Kaleidoscope MCP (SEC & financial data)
Server Details
Kaleidoscope by Memterra (not the Nathanael-R visual QA server): hosted SEC and financial data at mcp.kscope.io. Connects Claude, ChatGPT, Gemini, Cursor and any MCP client to cited, point-in-time data: ownership, activism, credit, private capital, filings, people and boards. Read-only tools, every answer cited to its source. Invitation-only: request a demo at mcp.kscope.io/request-demo, add the endpoint, then sign in with your allowlisted email.
- Status
- Healthy
- Uptime
- 100.0% over the last 24 hours
- Last Tested
- Transport
- Streamable HTTP · MCP 2025-11-25
- URL
- ChatGPT
- –
- Claude
- –
TDQS
Scored across 153 tools
Many tool families contain layered near-duplicates whose boundaries require memorization: thirteenf_holders vs thirteenf_holders_full vs thirteenf_who_holds vs thirteenf_smart_money_consensus; thirteenf_portfolio vs thirteenf_portfolio_full; company_deals vs deal_desk vs mars_mergers_for_company vs mars_mergers_screen; and several proxy-vote tools. Descriptions often include explicit 'use X for Y' guidance, but with 153 tools the agent must navigate substantial overlap and subtle scope differences, making misselection likely.
All names are snake_case and most are domain-prefixed, so they are readable and broadly organized by family. However, verb conventions are mixed: get_filing, search_activist_positions, screen_fundamentals, resolve_company_profile, discover_tools, and many bare noun phrases coexist, plus _full variants and superseded aliases.
At 153 tools, the surface is far beyond a well-scoped set and exceeds the rubric's extreme-mismatch threshold. While the server intentionally spans many SEC and financial subdomains, the count creates an overwhelming selection problem for an agent.
Coverage is exceptionally broad across filings, fundamentals, 13F/13D/G ownership, insiders, proxy voting, credit/BDC data, biotech, power, REITs, mining, Canada/SEDAR, macro, and graph relationships. Remaining gaps are mostly acknowledged caveats and foreign-issuer limitations, with workarounds available.
Available Tools
153 toolsactivism_campaignsActivism Campaigns (activist campaign timelines: 13D trail, demands, proxy fights, settlements)ARead-onlyIdempotentInspect
Activist campaign timelines — Console's Corporate Activism (ACTV) report. Two modes:
company (
ticker_or_cikorcompany_id): every activist campaign at the company, newest first.activist (
activistname, e.g. 'Elliott', 'Trian', 'Starboard'): every campaign that activist ran.
Each campaign: activist (name, mars investor_id when resolved), start / end, status (settled / withdrawn / election result / ongoing / no outcome reported) with the outcome text — set only by a withdrawal, settlement / cooperation agreement or contested election result; a proxy fight takes the company's election result even when it was reported under another activist on the same ballot (outcome_source says so), demands (board demands and nominations, plus 13D intent types), latest 13D stake %, and events in date order — 13D and 13D/A filings (stake %), demands and nominations, proxy solicitations, settlements / cooperation agreements, withdrawals, contested election results, and company responses (poison pill, anti-takeover provisions) — each with its SEC accession. Repeated proxy-solicitation filings are collapsed (n_filings). A campaign is one activist family at one company, split after a year of no activity.
Sources: mars activism events (contested proxy filings, vouched) + holdings 13D trail with intent. Takeover defenses beyond what filings disclose (SharkRepellent-style) are not available. For current 13D/13G positions use search_activist_positions.
Inputs: limit, since, activist, company_id, +3 more.
adv_custodiansADV Custodians (every custodian + prime broker an adviser or 13F family names)ARead-onlyIdempotentInspect
Every custodian and prime broker an investment adviser names on Form ADV, for one adviser (crd) or a 13F filer's whole adviser family (ticker or cik, up to 60 advisers). One row per firm (names matched without legal suffixes): roles (sma_custodian = Schedule D 5.K(3), firms holding 10%+ of separately managed account assets; custodian / prime_broker = named for a private fund in Schedule D 7.B(1)), place, sma_assets_usd, n_funds and fund_gav_usd (current private funds it serves), the advisers that name it (with CRD), and since (first ADV naming it; history from 2017-10). Firms no longer named follow with former=true and until. Also returns the advisers list. Each current firm row carries source_filing_ids (the Form ADV filings behind its current rows, newest first, max 10).
adv_screen_managers filters on an adviser's main custodian only; this lists them all.
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: cik, crd, ticker, include_former.
adv_manager_disciplinary_detailAdviser Disciplinary Detail (Form ADV regulatory DRPs — regulator/sanction/date/monetary_amount/allegations per action)ARead-onlyIdempotentInspect
Structured regulatory disciplinary history per adviser — the queryable detail behind adv_manager_profile's has_serious_disciplinary flag. Returns one row per regulatory DRP (Disclosure Reporting Page) currently on file at the adviser's latest ADV filing: regulator (e.g. SEC / FINRA / CFTC / state AG), principal_sanction (cease-and-desist / fine / suspension / bar / censure), date_initiated, monetary_amount, resolution + resolution_date, status (active / closed / pending), allegations (free-text narrative), and is_firm_level (firm vs affiliate DRP).
The lift over adv_manager_profile: profile tells you 'X has serious disciplinary' (boolean). This tool tells you 'X has a $200M CFTC cease-and-desist (2024-03), 3 FINRA actions (2019/2021/2023) totaling $4M, all closed' — auditable, with dates and amounts.
Coverage (2026-06-13): 125,498 regulatory DRP rows / 2,137 firms with at least one firm-level DRP. ~77% carry a date, ~76% the regulator, ~69% a monetary amount. Criminal + civil DRPs are NOT here — they're narrative-only on Form ADV (no structured fields exist to surface them). For firm-level disciplinary counts across the population, use adv_screen_managers(clean_only=false).
Pre-2025 caveat: only firms whose LATEST ADV filing is 2025-2026 are covered (the SEC's monthly-ZIP window the adv team ingests). Pre-2025 dormant/deregistered firms' regulatory history is phase-2 (historical backfill in flight). Live advisers are intact. Each DRP row carries source_filing_id (the Form ADV filing it is on) and reference_id (the DRP's reference id within that filing).
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: crd, limit, ticker_or_cik, firm_level_only.
adv_manager_feesAdviser Fee Schedule (full Form ADV Part-2A: mgmt/perf rates + components + raw_excerpt)ARead-onlyIdempotentInspect
Full Form ADV Part-2A brochure fee schedule for one adviser — header (firm-level rollup: headline mgmt + perf fees, hurdle / high-water-mark / negotiable / ceiling flags, defers_to_governing_docs, is_private_fund_adviser, source filing date) PLUS every component row (per-product / per-account-tier fee with its verbatim raw_excerpt and extractor confidence).
Coverage: 23,266 advisers with extracted fee records (one per adviser from their latest trailing-12-mo Part-2A); 2,236 are 13F-joinable via cik. 78% have a mgmt-fee; 13% have a perf-fee. 79% locator hit rate (21% misses are brochures not in S3 / Part-2B supplements / no locatable Item 5).
SEVEN LLM-facing gotchas — bake into any prompt that consumes this output:
Quote, don't paraphrase. Cite
raw_excerptverbatim. The structuredrate_pctis for screening, NOT for narrative quoting.rate_is_ceiling=truemeans 'up to X%', NOT 'charges X%' — say so explicitly.defers_to_governing_docs=truemeans real rate is in the LPA / PPM and the brochure won't disclose. Honest answer is 'not disclosed in the brochure' — do NOT report the structured rate as the actual rate in this case.tierspopulated means headline rate is the top-line; effective rate depends on account size. Reproduce tier breakpoints when account size matters.confidence+fee_source_filed_at— surface both. Fee schedules go stale; a 2022 brochure ≠ current rate.fee_data_available=false(no row in firm_fees for this CRD) is the explicit empty signal. NEVER treat absent fee data as 'no fees' or 'charges zero' — the brochure just hasn't been extracted yet.NULL ≠ no fee. Most important for perf fees: 91% of private-fund advisers with a fee record have
defers_to_governing_docs=truebecause carry lives in the LPA. Whenheadline_perf_fee_pct IS NULL, readdefers_to_governing_docsfirst. true → 'performance fee exists but the brochure defers the rate to fund documents.' Onlyfalse + NULLsupports 'no performance fee disclosed.' NEVER say 'charges no performance fee' from a NULL alone.
Honest tail caveat: ~4% of perf-fee firms read >30% (e.g. '50% of carried interest', '100% subject to clawback') — these are real brochure values with carry-split semantics that don't map cleanly to a 'performance fee %'. The headline is a max-rollup. Dollar-amount-as-percent extraction artifacts (e.g. $10,000 flat tier landing as 10000%) were fixed by adv on 2026-06-11 — rate_pct >100 now reroutes to rate_flat_usd per-tier (verified live: 0 firms with headline >100, 538 13F-joinable advisers match ≤1% mgmt). Still: always inspect raw_excerpt + confidence before quoting any rate as the actual fee. The fees block carries source_filing_id + source_brochure_id (the Form ADV filing and Part 2A brochure the header and components were extracted from; if_adv_brochure opens it).
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: crd, ticker_or_cik.
adv_manager_peopleAdviser People & Ownership (Form ADV Schedule A/B — executives, control persons + corporate ownership chain by firm CRD)ARead-onlyIdempotentInspect
Who runs / owns a management company — the people + ownership complement to the other adv_manager_* tools (which give RAUM / fees / disciplinary). From Form ADV Schedule A (direct owners + executive officers) and Schedule B (indirect ownership chain up to the ultimate parent). Given a firm CRD (the same key the other adv_manager_* tools take), returns:
• people — the individuals (executives / control persons): name, title, is_control_person, ownership_band, and owner_person_id when resolved (chain to donor_person_* / bio_person tools).
• ownership_chain — the corporate owners / parent entities, each with ownership_band and company_id when resolved.
Both are sorted control-persons-first. Verified: CRD 169355 → Clarke Futch (Chairman & CEO), Christopher White (President/CFO) + the full KKR ownership chain.
Name format: ADV stores owner names surname-first, comma-split ('FUTCH, CLARKE, BRANNEN' = Clarke Brannen Futch). Individuals' name is reordered for display; name_raw is the verbatim ADV string. schedule = 'A' (direct) or 'B' (indirect chain). Absence of rows means the firm hasn't filed Schedule A/B ownership (not covered), not 'no owners'. The response carries source_filing_id (the Form ADV filing whose Schedules A/B these rows come from).
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: crd (required), limit.
adv_manager_profileAdviser Profile (Form ADV: RAUM, disciplinary, custody flags — 13F manager quality)ARead-onlyIdempotentInspect
Form ADV manager identity card — for a 13F filer (by ticker, CIK, or CRD), return SEC-registered adviser metadata: legal firm_name, registration_type (SEC / ERA / STATE), current RAUM (regulatory assets under management) + as-of date + account count, private-fund footprint (n_private_funds, total gross asset value, top fund type), disciplinary history (has_serious_disciplinary + DRP counts), and custody red-flag indicators (self_custody, related_person_custody, auditor_pcaob_registered, n_redflags).
The 13F lift: 13F shows you positions; ADV shows you whether the manager behind them is a credible $20B clean shop or a $200M outfit with a disciplinary record. Use to filter '13F new position from manager X' for actual quality.
Key semantics (worth surfacing to the LLM):
The 13F-join key is
cik. 5,501 of 52,228 advisers have cik resolved — RIA-only advisers (don't manage qualifying institutional accounts) havecik=null; those don't file 13F anyway, so the null is consistent.Counts and flags are point-in-time off the adviser's LATEST ADV filing. RAUM history is in
adv_manager_raum_trend.has_serious_disciplinary(as of 2026-06-13) covers criminal + civil + registered-IA regulatory DRPs — firm-level serious doubled 291 → 1,261 when regulatory landed. Pre-2025 dormant/deregistered firms' regulatory is phase-2 (still backfilling), so this metric still undercounts on inactive advisers. For the queryable detail (regulator, sanction, date, monetary amount, allegations) calladv_manager_disciplinary_detail.n_redflags(range 0-N, max-1-or-2 in practice) is a rare weak-auditor signal — 387 firms have ≥1. Combine withauditor_pcaob_registered=falsefor the strongest red-flag screen.total_gav(gross asset value of private funds) is denominated in USD;current_raumis the regulatory metric. They can diverge — RAUM is the headline number 13F LPs care about.
Compact fee block (2026-06-11, when present): each adviser carries a fees block from their latest brochure. Live coverage 23,266 advisers / 2,236 13F-joinable. fees.fee_data_available=false is the explicit empty signal — never treat as 'no fees.' Critical NULL semantics: when headline_perf_fee_pct IS NULL, read defers_to_governing_docs FIRST. true → 'performance fee exists, rate is in the LPA/PPM' (91% of private-fund advisers do this). Only false+NULL supports 'no performance fee disclosed.' For component-level detail + raw_excerpt + confidence, call adv_manager_fees.
Each adviser row carries source_filing_id (the Form ADV filing the profile is read from) and its fees block carries source_filing_id + source_brochure_id (the brochure the fee data was extracted from; if_adv_brochure opens it).
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: crd, as_of, ticker_or_cik.
adv_manager_raum_trendAdviser RAUM Trend (annual time series — forced-seller detection)ARead-onlyIdempotentInspect
Per-adviser RAUM (regulatory assets under management) time series from Form ADV filings. The forced-seller signal: when a concentrated 13F holder's RAUM is in steep decline, its 13F positions are under redemption-driven selling pressure. Cross-reference the time-series with thirteenf_portfolio (or thirteenf_portfolio_full) for the names at risk.
Each row carries as_of (the PIT date the RAUM was reported), raum_total, raum_discretionary, raum_nondiscretionary, n_accounts. Returned newest-first. Each point also carries source_filing_id (the Form ADV filing that reported it).
Key semantics:
RAUM cadence is ANNUAL + amendments — NOT quarterly. So
limit=8is roughly 8 years of history, not 2 years. 19,707 advisers have ≥4 filings of history.Wind-downs appear as sharp drops (e.g. AJO went $19B → $0.6B over one filing cycle when the firm shut down).
Pass
ticker_or_cik(resolved via Atlas),cikdirectly, orcrd(the Form ADV primary key — use when the adviser has no SEC CIK).
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: crd, limit, ticker_or_cik.
adv_private_fundsAdviser Private Funds (Schedule D funds + service providers)ARead-onlyIdempotentInspect
Schedule D private funds reported by one adviser, with their service providers (auditor, prime broker, custodian, administrator, marketer). Covers both registered advisers and exempt reporting advisers (ERAs — most VC/PE funds).
This is a CURRENT-STATE register, not a per-filing time series: each fund row reflects its latest-known filing, globally, across every source that's ever reported it. A fund the adviser stopped reporting (closed/liquidated) still appears, frozen at its last reported state — there is no flag or parameter to show 'as of filing N' or to exclude closed funds; everything the adviser has ever had on file is always included.
role on each service provider is a TYPE (auditor/prime_broker/custodian/administrator/marketer), not an Item 7.B(1) sub-question number — don't present it as a literal filing-item citation.
Provider CRD/company resolution is partial (~2.4% of provider rows resolve to a company_id today) — a null provider_company_id means unresolved, not that the provider doesn't exist.
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: crd, ticker_or_cik.
adv_raum_declinersAdviser RAUM Decliners (forced-seller / Redemption Watchlist — discretionary-RAUM drops + basis-change guard)ARead-onlyIdempotentInspect
Forced-seller / Redemption Watchlist tool. Returns advisers whose discretionary RAUM fell between their two most recent ADV filings, ranked by percent decline. Pair with thirteenf_portfolio_full(cik=...) to surface specific equities under redemption-driven selling pressure ('which stocks face forced selling this quarter').
Backed by a pre-computed table (proc2.adv.firm_raum_decliners, 3,423 rows / 816 with cik). Daily blue/green refresh. Sub-2ms warm.
Two correctness-critical design choices baked in upstream by adv — surface these in any prompt that consumes the output:
Decline is computed on DISCRETIONARY RAUM, not total. The Sageview case is the canonical false-positive class: −84% on total RAUM ($238.6B → $37.5B) but +0.5% on discretionary — the entire $201B drop was a nondiscretionary reclassification, NOT a redemption. Sageview never enters the table because the filter is discretionary-only. Don't second-guess the table by re-computing off total.
basis_change_suspect=trueflags the residual cases where discretionary DID fall but the discretionary/total mix also shifted >20pts. These are likely reporting-basis reclassifications, NOT redemptions (K1 Investment Mgmt, GraniteShares ETF reclass are typical examples). The tool default excludes these (include_basis_changes=false). Setinclude_basis_changes=trueonly if you're auditing the data; never include them in 'who's facing forced selling' answers.
Killer acceptance case (per spec): Politan Capital Management (crd 316047, cik 1885245) → drop_pct=46.04, basis_change_suspect=false. The Masimo (MASI) campaign — clean redemption signal. Chain into thirteenf_portfolio_full(cik=1885245) for the at-risk equity book.
No as_of param — by design (adv team's position 2026-06-12, futures agreed): the table always compares each adviser's two MOST RECENT filings to detect discretionary-RAUM declines. Adding as_of would conflict with that purpose. For replay backtests at a past date, reconstruct equivalent decline signals from adv_manager_raum_trend(crd, limit=8) — the trend tool's rows carry their own as_of (Pattern B PIT, ~300K rows across all filings), so a client-side cutoff to as_of + take the two most recent ≤ as_of + compute (1 - latest_disc/prior_disc) reproduces the decliner detection PIT-correctly.
Filters (all index/scan-fast):
min_drop_pct(default 30) — minimum % discretionary declinemin_prior_raum_disc(default 1e9) — prior-filing discretionary RAUM in USD; use to anchor on size (1e9 = $1B+ advisers, which is the futures team's default for the Redemption Watchlist)min_latest_asof— only include advisers whose latest filing is on/after this date (YYYY-MM-DD); use to scope to a recent vintagewith_cik_only(default true) — restrict to advisers with a resolved SEC CIK (the 13F-joinable subset; the whole point is chaining intothirteenf_*). Coverage gap: only a minority of >$1B-prior-disc decliners resolve to a cik today — adv's CRD↔CIK bridge backfill is in flight; universe widens over time.include_basis_changes(default false) — set true to opt back in (audit only)sort∈ {drop_pct_desc(default),prior_raum_disc_desc,latest_asof_desc}limit(default 50, max 200)
Spec-validated counts (for sanity): no filters → 3,423; with_cik_only=true → 816; min_drop_pct=30 + min_prior_raum_disc=1e9 + with_cik_only=true + NOT basis_change_suspect (the worked default) → 60 advisers.
Each row carries prior_source_filing_id / latest_source_filing_id (the two Form ADV filings the prior / latest RAUM come from).
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: sort, limit, min_drop_pct, with_cik_only, +3 more.
adv_screen_managersAdviser Screener (cross-sectional Form ADV: size, type, clean-record, fees)ARead-onlyIdempotentInspect
Adviser Screener — rank and filter Form ADV advisers with AUM (one row per CRD, ~21.6K) by size, growth, client base, fees, custodian, location, regulatory record and adviser moves. Matches Console's Adviser Screener (/ledger/advisers).
Each row: aum (+ aum_as_of), g1 / g3 (1y / 3y AUM growth as a fraction, 0.25 = +25%; null when AUM at the start was under $50M), focus (wealth / institutional / private_funds / mixed — the client type holding 50%+ of RAUM), n_clients, n_employees, n_iar, mgmt_fee_pct / perf_fee_pct (headline brochure fees), n_private_funds, main_custodian (holds the most RAUM), city / state, events_5y (regulatory events, firm- and affiliate-level, in the last 5 years), has_serious_disciplinary, net_advisers_1y (advisers joined minus left; covers only ~5K firms — null means not covered, not zero).
Presets (preset; explicit filters override a preset's values): growing_wealth (wealth focus, $100M+, 1y +15%), large_wealth (wealth, $10B+), emerging ($100M–$1B with 3y +50%), private_funds, institutional, adding_advisers (net_advisers_1y ≥ +5), losing_advisers (≤ −5), shrinking (1y −15% or worse), regulatory (1+ event in 5 years, $100M+).
Filters: aum_min / aum_max (USD), g1_min / g1_max / g3_min (fractions), focus, state (2-letter), custodian (name substring), fee_max (max management fee %), pf_min (min private funds), net1_min / net1_max, events_min, name (substring) or crd. Older filters still work: min_raum / max_raum, clean_only, with_cik_only (13F-joinable — chain into thirteenf_* tools), registration_type, top_fund_type_in, firm_name_substring, max_mgmt_fee_pct, min_perf_fee_pct (NULL perf fee does not match; for private-fund advisers it usually means carry is in the LPA, so this undercounts), has_hurdle, fee_type_in, fee_data_only.
Sort: raum_desc (default), raum_asc, g1_desc, g1_asc, g3_desc, net1_desc, net1_asc, events_desc, fee_asc, n_private_funds_desc, n_accounts_desc (by clients). Growth / mover presets default to the matching sort. limit default 25, max 200. n_universe is the full match count.
as_of (YYYY-MM-DD) switches to the point-in-time Form ADV history screen (2017-10 → 2026-05, replay only); the new Console fields and presets are current-only and are rejected with as_of. The current screen covers SEC-registered advisers only, so registration_type='ERA'/'STATE' needs as_of.
Each row carries source_filing_id (the Form ADV filing behind aum / raum) and, on the current screen, source_brochure_id (the brochure the headline fees were extracted from; if_adv_brochure opens it).
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: crd, name, sort, as_of, +28 more.
bdc_summaryBDC Summary (every public BDC side by side: mark, below-90 share, non-accruals, NAV)ARead-onlyIdempotentInspect
Every covered public BDC side by side at its latest quarter (Console Credit Monitor BDC table): debt fair value, amortized cost, mark % (FV / cost), share of FV marked below 90% of cost, non-accrual % by FV and by cost, NAV per share, borrower count, as-of quarter. From each BDC's 10-Q Schedule of Investments (53 BDCs, data through 2026-09-30; e.g. ARCC, BCRED, OBDC, FSK, BXSL).
Source-gap flag: a book of 100+ borrowers reporting zero non-accruals is almost always an extraction miss, so non-accrual % comes back null with non_accrual_source_gap=true — never read it as 0%. BDCs report on staggered quarters; report_date is per BDC. For one BDC's positions, seniority mix and trend call credit_bdc_loan_book(ticker); for shared loans marked differently across BDCs, credit_cross_held_loans.
Params: tickers (subset), as_of (latest quarter on or before), sort (fv_desc default / mark_asc / below_90_desc / non_accrual_desc). Each row carries accession + filing_url (the BDC's 10-Q/10-K for nav_report_date, the same quarter's filing as the marks when nav_report_date = report_date; open with get_filing).
Inputs: sort, as_of, tickers.
bio_cash_runwayCash-runway trajectory (years + declining-streak) per biotech (PIT)ARead-onlyIdempotentInspect
Cash-runway trajectory in YEARS per biotech — cash ÷ annual operating burn, per quarter, with the QoQ change and the consecutive-declining-quarter streak. Answers 'how much runway, and is it deteriorating?'. 34,338 quarters / 1,423 companies; PIT.
Inputs: ticker_or_cik (required), as_of (YYYY-MM-DD — PIT gate on when the fundamentals became public; only quarters with pub_date < as_of), last_n (default 8).
Contracts / caveats (surface these): (1) runway_years is YEARS, not quarters (cash ÷ ANNUAL burn) — label it years. (2) Read the TRAJECTORY, not a single point. runway uses TTM operating cash flow, so commercial-stage names transitioning to profitability legitimately OSCILLATE (e.g. a name can swing 3 → 58 → 2.5yr as product revenue lands against R&D spend) — that's real, not a bug. (3) declining_streak ≥ 3 is the meaningful DISTRESS signal (a steady pre-revenue burner running down cash); declining_streak = 1 is usually just cash-flow lumpiness. Lean on streak≥3 / consistently-low runway for the 'deteriorating' read, especially for pre-revenue clinical-stage names (most of the universe — stable + meaningful). (4) PIT on pub_date. (5) Universe is survivorship-complete, so a company's 'latest' quarter may be its terminal filing before delisting — quarter_end tells you how current it is.
atm_usage (#2189): the at-the-market program — selling_now, last quarter's shares sold, net proceeds and remaining capacity (cash the company can still raise), sales agents, quote + accession, quarterly history; null when no ATM is on file. Each trajectory row carries accession + filing_url (the 10-Q/10-K that quarter's cash and burn come from).
Inputs: ticker_or_cik (required), as_of, last_n.
bio_catalyst_dossierNext FDA catalyst + how-to-think-about-it dossier (index with drill-down pointers)ARead-onlyIdempotentInspect
The next FDA catalyst for a biotech, plus how to think about it — as an INDEX with drill-down pointers. One call returns the nearest PDUFA/AdCom date AND a compact headline for each context pillar (prior CRLs + the base rate for that reason, comparable historical approval setups with realized returns, management track record, manufacturing/inspection risk, pipeline) — each with an exact drill tool call to go deeper. Read the index, then follow the drill for whichever pillar matters to your question; each drilled tool surfaces its own detail (and often its own pointers).
Real-time fan-out over already-cooked data (base rates, PIT snapshots, CRL reasons, the people graph are all pre-computed) — nothing baked ahead, nothing stale. Token-cheap by design: headline + pointer per pillar, NOT the full record of each.
Inputs: ticker_or_cik (required), as_of (YYYY-MM-DD — PIT gate for the CRL/pipeline/comparables reads; default today).
Each pillar carries its headline fields + drill (the ready next call) + available (false when that read errored or is empty — the drill still works). Pillars: next_catalyst · prior_crls · crl_base_rate (for the most-recent prior reason) · target_base_rate (the target's historical approval base rate — the key prior for an AdCom efficacy question) · comparable_approvals · management · manufacturing_risk · pipeline. next_catalyst carries source_accession + filing_url; prior_crls carries most_recent_accession + most_recent_filing_url when that CRL was disclosed in an SEC filing.
Caveat: the drill calls are the source of truth for detail — the dossier headlines are deliberately lossy summaries. Approval-anchored comparable returns are YES-arm only (see bio_similar_setups); CRL absence isn't proof of none (see bio_prior_crl_reason).
Inputs: ticker_or_cik (required), as_of.
bio_company_peopleOfficers/directors of a biotech, each with their track record elsewhereARead-onlyIdempotentInspect
The officers & directors at a biotech, each with their cross-company track record. For every exec/director who filed a Form-4 here, returns their role + how the OTHER companies they ran/sat on resolved (approval / acquired / failed / active). The money read: a new CEO who ran 3 biotechs that got approved. Complements insiders_of_company (SEC identity+count only) with the biotech outcome graph.
Inputs: ticker_or_cik (required), officers_only (default false — set true to drop directors and keep only officers, the operator-skill signal), limit.
Returns per person: insider_name, owner_cik, role, is_officer / is_director, first/last date here, and their elsewhere-record (n_companies, n_cos_approval / acquired / failed / active, top_role, is_entity). Ranked officers-first then by approval count. Pass an owner_cik to bio_person_track_record for that person's full company detail.
Contracts / caveats: (1) coverage = Form-4 filers 2003+ (pre-IPO founders absent); (2) the track record is company-level (credentialed-by-association for directors) — weight officers over directors for operator skill; (3) is_entity=true rows are fund/VC board seats (distinct smart-money signal, not individual skill).
Inputs: ticker_or_cik (required), limit, officers_only.
bio_crl_outcomesPost-CRL abnormal returns + approve-later rate, by deficiency reasonARead-onlyIdempotentInspect
Realized post-CRL outcomes by deficiency reason — the base-rate table over 419 FDA Complete Response Letters (2015+): given a CRL for reason_category X, the median/mean SPY-adjusted abnormal return at 21/63/126/252 trading days AND the eventual-approval rate. The load-bearing finding: reason predicts BOTH axes — efficacy is fatal (worst move ≈ −0.43, ~32% approved later); inspection_483 / cmc_manufacturing are fixable (mild move, high approve-later). Cross-sectional companion to bio_prior_crl_reason (one sponsor) and the NO-arm of bio_similar_setups.
Inputs: reason_category (filter to one reason — see enum), ticker_or_cik (a sponsor's own CRL events), limit. With neither filter you get the full by-reason base rate across the corpus.
Returns: by_reason (n, median/mean abn per horizon, %down at 252d, and approved-later rate) + events (the individual CRL rows).
Contracts / caveats (surface these): (1) abn base rate uses 126/252d — ~57% of rows are date_precision='month' (SEC month-only dates), so their 21/63d abn are unreliable and surfaced flagged; the 126/252d numbers + the aggregate are sound. (2) approve-later is 'of CRLs with known resolution' — eventual_approval is populated for openFDA rows but NULL for most sec_disclosure rows (SEC filings don't state eventual resolution), so the denominator (n_resolved) is < n. Never read the rate as over all CRLs. (3) Reason is classification-grade (Haiku at ingest), not a legal determination.
Each event carries accession + filing_url (the SEC filing that disclosed the CRL; open with get_filing) on sec_disclosure rows, and source_filing_id (the FDA letter file name) on openfda rows.
Inputs: limit, summary_only, ticker_or_cik, reason_category.
bio_entity_resolveBio Entity Resolve (gene/target alias → canonical HGNC)ARead-onlyIdempotentInspect
Superseded by resolve(identifier, kind='gene') (the shared resolver) — kept callable for one more release so existing prompts don't break, but new callers should use resolve instead. Resolve a user-facing biotech name (gene alias like 'PD-1' or 'HER2', modality, or indication) to the canonical identifier the rest of the bio tools expect. Returns ALL matching candidates — some queries are genuinely ambiguous (PD-1 → PDCD1 AND SNCA); the caller should pick. Resolves via an exact-normalize lookup against the curated HGNC gene gazetteer (sub-ms); on miss, falls back to semantic search over the targets collection so fuzzy queries like 'checkpoint inhibitor' still produce candidates. Use this first whenever a user passes a name to any other bio_* tool — the other tools want canonical symbols (PDCD1, ERBB2, ...).
Inputs: query (required), k_fallback.
bio_fda_approvalsFDA Approvals (NDA/BLA action dates by ticker/CIK/drug — authoritative post-event anchor)ARead-onlyIdempotentInspect
Historical FDA NDA/BLA approvals — the PIT-authoritative action dates (submission_status_date) per sponsor / brand / active_ingredient. Use to answer 'what has this company been approved for' / 'when did evolocumab clear' / 'how many BLAs in oncology since 2020.' Pairs with catalyst_calendar(pdufa) for the forward calendar and bio_similar_setups for the post-approval comparable cohort.
Coverage (substrate refreshed 2026-06-17): 1,388 NDA/BLA originals since 2015 — 1,363 CDER (openFDA drugsfda pull) + 25 CBER (curated cell/gene/CAR-T cohort). 645+ matched to cik/ticker (CBER match rate 72% — cell/gene sponsors are biotech-pure-play) via super-bio's brand_exact / brand_alias / ingredient_exact / brand_partial matchers with matched_confidence ∈ [0.5, 1.0]. Supplements (sBLA / sNDA) excluded by default — pass originals_only=false to include them.
Inputs (all optional, at least one of ticker_or_cik / drug / modality recommended):
ticker_or_cik— equity ticker or SEC CIK. Filter to one company.drug— ILIKE substring match againstbrand_nameORactive_ingredient(case-insensitive). Either brand (REPATHA) or INN (evolocumab) works.modality— drug-class filter: 'CAR-T' / 'gene_therapy' / 'cell_therapy' / 'mAb' / 'mAb_bispecific' / 'small_molecule' / 'siRNA' / 'antisense_oligo' / 'peptide' / 'protein'. Use this for 'every approved cell/gene therapy' or 'every CAR-T approval' queries — these are the CBER-class rows the modality column was added for. Filter is exact-match on the modality column.since— ISO date floor onsubmission_status_date(default2015-01-01).originals_only— bool, default true. False to include supplements (sNDA / sBLA).min_confidence— float in [0.0, 1.0], default 0.85. Higher = stricter ticker/cik match precision. Important caveat (super-bio note): whenmatched_confidence < 0.9, prefersponsor_namefor filer-correctness — the ticker/cik might be a licensee rather than the original developer. This is especially true for CBER (cell/gene) rows where the approval may have been transferred to a downstream developer.limit— max rows returned (default 50, max 500).
Foreign-private-issuer caveat (super-bio note): 20-F filers in the bio cohort (ARGENX, BNTX, LEGN) aren't in super-bio's universe yet (queued as FUTURE_DIRECTIONS §5 #13). For those, the ticker/cik match either lands on a US licensee (Vyvgart → ZLAB) or cik=NULL. When cik IS NULL but the approval is clearly the foreign filer's drug (recognize via sponsor_name), interpret accordingly.
Response shape: list of {application_number, submission_number, fda_action_date, sponsor_name, brand_name, active_ingredient, modality, source, cik, ticker, matched_method, matched_confidence} rows sorted by submission_status_date DESC. source ∈ {'CDER', 'CBER'} disambiguates row origin. Plus as_of_effective, n_returned, filters_applied.
Live verified spec cases: ticker_or_cik='TVTX' → 3 originals (THIOLA EC 2019-06-28, FILSPARI 2023-02-17, VOYXACT 2025-11-25); drug='evolocumab' → REPATHA AMGN 2015-08-27 (+ 2021 supplement when originals_only=False); ticker_or_cik='REGN', since='2020-01-01' → INMAZEB BLA761169 2020-10-14, EVKEEZA BLA761181 2021-02-11, +6 more; modality='CAR-T' → 7 rows (KYMRIAH 2017-08-30, YESCARTA 2017-10-18, TECARTUS 2020-07-24, BREYANZI 2021-02-05, ABECMA 2021-03-26, CARVYKTI 2022-02-28, AUCATZYL 2024-11-08); modality='gene_therapy' → 12 rows (LUXTURNA 2017-12-19 → BEQVEZ 2024-04-25); modality='cell_therapy' → 6 rows (STRATAGRAFT → RYONCIL — the CAPR-class reference set).
Inputs: drug, limit, since, modality, +4 more.
bio_inspectionsFDA Inspections (NAI/VAI/OAI classifications — the EARLY enforcement signal)ARead-onlyIdempotentInspect
FDA inspection classifications (NAI / VAI / OAI) by recipient — the EARLY enforcement signal, often months before a Warning Letter. Use to answer 'every OAI in oncology pharma since 2024' / 'has this CDMO had any OAIs' / 'what does this sponsor's inspection record look like.'
Classification codes (the load-bearing field):
OAI(Official Action Indicated) — regulatory/legal action warranted. The bad one. Often the upstream signal of a coming Warning Letter or import alert.VAI(Voluntary Action Indicated) — observations noted, voluntary correction.NAI(No Action Indicated) — clean inspection.
Use worst_only=true as a convenience for classification='OAI' — the high-signal subset for sponsor risk screening.
Inputs (all optional; pass at least one of ticker_or_cik / company / classification for a meaningful query):
ticker_or_cik— equity ticker or SEC CIK. Filters to inspections where super-bio's matcher resolved the establishment to a super_bio sponsor.company— ILIKE substring onlegal_name. Use for CDMOs (Catalent, Lonza, WuXi, Patheon, Thermo Fisher), foreign API makers, or BIMO clinical investigators that aren't in super_bio's sponsor universe.classification— exact match onclassification_code: 'OAI' / 'VAI' / 'NAI'. Mutually compatible withworst_only(worst_only wins if both set).worst_only— bool convenience forclassification='OAI'. Default false.project_area— substring match: 'Drug Quality Assurance' / 'Bioresearch Monitoring' (BIMO clinical trial sites) / 'Human Cellular, Tissue, and Gene Therapies' / etc.since— ISO date floor onclassification_date(default2015-01-01).min_confidence— float in [0.0, 1.0], default 0.85. Only applies when filtering by ticker_or_cik (forces tight matches). Set to 0.5 to also see name_overlap matches.limit— default 50, max 500.
Response shape: list of {inspection_id, fei_number, legal_name, city, state, country_name, fiscal_year, project_area, product_type, classification_code, classification, classification_date, posted_citations, matched_cik, matched_ticker, matched_method, matched_confidence, is_cdmo} rows sorted by classification_date DESC. fei_number is FDA's Establishment Identifier — useful for joining unmatched sites and for tracking site-level history across multiple owners.
Coverage (substrate live 2026-06-18): 26,133 classified inspections since 2015 across Drugs + Biologics (Biologics scope captures CBER cell/gene-therapy site inspections). Mix: OAI 453 / VAI 9,313 / NAI 16,367. ~50% matched to a super_bio sponsor; unmatched rows are mostly foreign API makers, CDMOs, and BIMO clinical investigators (surface via company ILIKE).
Pairing with bio_warning_letters: same recipient, two enforcement stages. Recommended consumer pattern: query bio_inspections(worst_only=true) for the EARLY signal, then bio_warning_letters for the escalation. An OAI with no subsequent Warning Letter is itself informative (issue may have been resolved at the 483 stage).
Verified spec cases (live 2026-06-18): company='Catalent', classification='OAI' → Catalent Indiana, LLC · 2025-11-24 · OAI · Drug Quality Assurance · ticker=CTLT — the canonical CAPR-class case, sitting 4 days BEFORE the CTLT Warning Letter 2025-11-20 (the inspection IS what the Warning Letter later escalated); worst_only=true, since='2024-01-01' → all OAIs since 2024 (highest-signal subset for sponsor risk screening); project_area='Human Cellular, Tissue, and Gene Therapies' → cell/gene therapy site inspections.
Inputs: limit, since, company, worst_only, +4 more.
bio_landscapeBio Landscape (every company with a target / indication, as-of date)ARead-onlyIdempotentInspect
Find every company with a drug asset targeting a gene (axis='gene') or treating an indication (axis='indication'), as of a date — with source filings cited. The cross-sectional companion to bio_pipeline (which is one-company-over-time).
Returns one row per (company, asset) so a single company with three assets against the target appears three times. Pre-resolve gene aliases (PD-1, HER2) with bio_entity_resolve first — landscape wants the canonical symbol (PDCD1, ERBB2). Indication matching (updated 2026-06-15): packed canonicals like 'immunoglobulin a nephropathy (igan), primary membranous nephropathy (pmn), focal segmental glomerulosclerosis (fsgs), …' are now exploded at build time in super-bio's pipeline_snapshots_by_indication matview, so an exact match on any single disease (e.g. 'immunoglobulin a nephropathy (igan)') finds every company whose canonical includes it. Original packed string is preserved as indication_packed_canonical for display. Tier-3 disease ontology (Cancer / Oncology / Solid tumors unification) is still future work on the super-bio side.
Inputs: axis (required), value (required), as_of_date.
bio_patent_cliffLoss-of-exclusivity / patent-cliff date per approved drug (Orange Book + biologic 12yr)ARead-onlyIdempotentInspect
Loss-of-exclusivity (patent-cliff) date per approved drug for a sponsor — the revenue-at-risk date when a marketed drug is exposed to generic / biosimilar entry. Small molecules carry REAL FDA Orange Book data (listed patent-expiry + regulatory exclusivity); biologics carry the computed BPCIA 12-year regulatory floor. 521 drugs.
Inputs: ticker_or_cik (required), before (YYYY-MM-DD — only drugs with loe_date ≤ this; use with a +3yr date for a 'rolling off patent soon' screen), drug (ILIKE substring on brand or ingredient).
Contracts / caveats (surface these): (1) source tells you what the date IS — orange_book = real listed patent + exclusivity data (the true cliff, loe_date = max(latest_patent, latest_exclusivity)); biologic_12yr_rule = regulatory FLOOR only (approval + 12yr), NOT the real patent cliff — actual biosimilar entry can differ via patent-dance settlements. Never present a biologic_12yr_rule date as a hard patent expiry. (2) has_substance_patent true = a composition-of-matter patent is listed (the strong one). (3) Absence ≠ on-patent-forever — the Orange Book delists expired patents, so a drug with no row is either already off-patent OR never listed patents; the two can't be distinguished. (4) ticker/cik is inherited from the approval match — mostly conf≥0.95 but a few sponsor-match artifacts exist upstream (e.g. IBRANCE tagged TRAW not PFE).
Inputs: ticker_or_cik (required), drug, limit, before.
bio_person_track_recordBiotech exec/director cross-company outcome record (Form-4 owner_cik identity)ARead-onlyIdempotentInspect
A biotech exec/director's cross-company outcome record — how many companies they've been an officer/director at, and how those companies resolved (drug approval / acquired / failed / active). The 'proven operator' read: a CEO who ran 3 biotechs that got approved, surfacing at a new one, is signal. Identity is authoritative (SEC Form-4 owner_cik — no fuzzy name matching). 16,943 people; 1,593 proven individual operators (2+ approval cos).
Inputs: name (ILIKE — returns all matching people) OR owner_cik (exact), one required; include_funds (default false — by default returns individuals only; set true to include funds/VCs like OrbiMed / Baker Bros whose board-seat record is a distinct smart-money signal).
Returns per person: insider_name, owner_cik, is_entity, top_role, n_companies, n_cos_approval / acquired / failed / active, and companies (the per-company detail: ticker, company, role, first/last year, outcome).
Contracts / caveats (surface these): (1) Coverage = Form-4 filers (officer/director at a universe biotech, 2003+) — pre-IPO founders who never filed a Form-4 are absent. (2) Outcome is company-level, not drug-attribution — a director at 12 approval-cos is credentialed-by-association; weight officers (CEO / CSO / CMO) over directors for operator skill vs board-network signal. (3) is_entity=true = a fund/VC, not an individual — their record is a distinct signal; keep it separate from individual operator skill.
Inputs: name, limit, owner_cik, include_funds.
bio_pipelineBio Pipeline (point-in-time reconstruction + full trajectory)ARead-onlyIdempotentInspect
Reconstruct a biotech company's drug pipeline at a point in time, or its full trajectory over time, with source filings cited on every fact. This is super-bio's killer capability — point-in-time pipeline reconstruction is something a frontier LLM provably cannot do without our extraction graph.
Two modes: • mode='at' (default): the company's pipeline AS OF a given date — every drug asset, its modality, current phase, targets (canonical genes), indications, and the SEC accession + verbatim snippet that established each fact. Returns the latest snapshot ≤ as_of_date. • mode='history': the full snapshot stream (per-quarter pipeline state) PLUS the resolved program_outcomes — for each drug program: first_snapshot / last_snapshot / max phase reached / final outcome (approved / advanced / acquired / discontinued).
Accepts a ticker (resolved via Atlas → CIK) or a raw CIK integer. Returns an empty assets list (NOT an error) when the company exists but has no extracted snapshot yet for the window. Resolve a target/gene name first with bio_entity_resolve if you want to cross-reference targets here against canonical genes.
Inputs: ticker_or_cik (required), mode, as_of_date.
bio_prior_crl_reasonFDA Prior CRLs with deficiency reason (PIT — the discriminator behind a 'saw prior CRL' gate)ARead-onlyIdempotentInspect
Prior FDA Complete Response Letters (CRLs) for a sponsor, PIT-strict, with the deficiency reason as the load-bearing signal. The 'why did this CRL happen' field that lets a caller distinguish fixable issues (CMC / manufacturing) from fatal ones (efficacy). Pairs with bio_similar_setups(event_type='approval') as the no-arm complement — that tool gives realized post-approval returns conditional on approval; this one surfaces the prior-CRL pattern that often preceded a no-arm outcome.
Philosophy: ingredients, not verdicts. Returns the reason category + the verbatim deficiency sentence from the CRL + a classification confidence. The caller LLM weighs them. The tool never says 'this means X.'
Coverage (super-bio initial 2026-06-27; expanded 2026-07-03 with Tier-1 coverage-breadth work): 934 letters 2002-2026 / 220 distinct companies. Two sources — surfaced per row as source:
source='openfda'(439 rows) — the original authoritative FDA CRL transparency-feed letters.application_number+approval_statuspopulated. Deficiency reason classified from the OCR'd letter text.source='sec_disclosure'(495 rows / ~132 net-new companies since 2026-06-27) — the company's OWN SEC-filing disclosure of a CRL. Filer CIK match,matched_confidence=1.0. Populatesdrug(surfaced per row).application_number+approval_statusare NULL by design (SEC disclosures don't reliably state the NDA/BLA number or resolution), so the caller-facing computed fields degrade honestly:same_application_as_current→null/false,eventual_resolution→'unknown'. That's correct behavior, not a bug — the FACT (company + drug + date) is still there. ~27% of SEC rows havereason_category='unknown'(reason not stated in the disclosure).
The corpus still skews to generics/specialty pharma (particularly on the openfda side); thin/unapproved biotechs are less absent than before but coverage remains incomplete. absence != "no prior CRL" — always surface the coverage caveat when an empty result is the answer.
PIT hygiene caveat on SEC-disclosure rows (flagged by futures 2026-07-06, substrate fix pending on super-bio side): letter_date on source='sec_disclosure' rows is the CRL EVENT date narrated in the filing ("In April 2026, we received a CRL..."). The disclosing SEC filing was made public LATER — often days, sometimes weeks. Until super-bio adds a disclosing_filing_date column to gate on, PIT replays with as_of set to a date BETWEEN the event date and the disclosing filing date may surface a CRL that wasn't yet publicly known. Buffer of ~60-90 days typically avoids the leak on 10-Q disclosures; strict PIT should treat SEC-disclosure rows within ~90 days of as_of with caution. openFDA rows are gated correctly.
Inputs:
ticker_or_cik— equity ticker (Atlas-resolves to CIK) or SEC CIK integer. Required.as_of— ISO date YYYY-MM-DD. The PIT anchor. Required — letters withletter_date < as_ofonly.eventual_resolution='approved_later'is also PIT-gated against this date (only asserted if the resolution itself predates as_of, else 'unknown').application(optional) — current event's NDA/BLA identifier (e.g.'BLA 761303'). When passed, each returned row carriessame_application_as_current(bool) so the caller can flag same-application repeats — the strongest signal short of identical molecule. Normalization strips/Original N//Resubmission Nsuffixes so submission cycles don't break the match.drug(optional) — current event's drug name (INN, brand, code-name). When passed, each returned row carriessame_molecule_diff_formulation(bool) — case-insensitive substring match either way. Catches the INCY lesson (ruxolitinib cream vs XR were the same unresolved PK issue across formulations).limit— max rows returned (default 50, max 200).
Returned reason categories (the reason_category field):
cmc_manufacturing / efficacy / safety / bioequivalence_pk / clinical_design / labeling / inspection_483 / other
Response shape: list of {crl_id, application_number, company_name, letter_date, approval_status, reason_category, key_deficiency_sentence, classify_confidence, file_name, same_application_as_current, same_molecule_diff_formulation, eventual_resolution} sorted by letter_date DESC. Plus as_of_effective, n_returned, lookup, and a coverage_caveat block whenever the sponsor returns zero rows (so the caller doesn't infer absence-as-evidence).
Verified spec case (super-bio): bio_prior_crl_reason(ticker_or_cik='REGN', as_of='2026-01-01') → 5 CRLs including BLA 761303 odronextamab same-molecule repeat pattern (2024-03-22 clinical_design Unapproved → 2025-07-30 inspection_483 Unapproved) — exactly the 'saw prior CRLs, same unresolved issue' discriminator.
Inputs: as_of (required), ticker_or_cik (required), drug, limit, +1 more.
bio_risk_scoreBio Risk Score (composite de-risking — magnitude/sizing, NOT direction)ARead-onlyIdempotentInspect
Composite de-risking score for a biotech company at a point in time — a single number summarizing pipeline-breadth × has-approved-drug × runway × partnership × target-crowdedness, validated against catalyst-day |returns|.
⚠️ MAGNITUDE / SIZING signal — NOT a direction predictor. Higher score = more de-risked = smaller expected |moves|. Does NOT say which direction the stock will go. Use for sizing / vol bands / options input. The methodology caveat is in the response — carry it into any brief that quotes the number.
Returns the latest score ≤ as_of_date with components (n_assets, runway_quarters, has_approved, log_market_cap, max_target_crowdedness, partnership_present) and source_refs (which inputs the score was joined from).
Inputs: ticker_or_cik (required), as_of_date.
bio_target_base_rateBio Target Base Rate (historical phase-progression rate, survivorship-complete)ARead-onlyIdempotentInspect
Historical phase-advancement rate for a target gene — fraction of resolved programs that reached approval or were acquired-while-active, out of all resolved (positive + discontinued). Discontinuations ARE in the denominator — programs quietly dropped, delisted, shelved, OR that recorded a documented Phase-3 endpoint FAILURE (v2, from SEC trial-readout extraction) show up as negatives. Substrate: 21 years of 10-Ks across 1,676 CIKs, keeping programs from delisted / bankrupt / spun-out issuers.
⚠️ CRITICAL — NOT a trial-success rate, and the residual bias is CENSORING (conservative), not inflation. POSITIVE = approved or acquired-while-active. NEGATIVE = discontinued (company died, program dropped, or a documented Phase-3 miss). A program at a still-alive company still ADVANCING with no failure signal is CENSORED — excluded from the denominator entirely, NOT scored positive. So the metric never counts advancement as success; its residual bias is that it still censors un-resolved and earlier-phase / undisclosed failures, so it leans conservative-optimistic on targets with lots of live pipeline. v2 (methodology phase_progression_v2) closes the Phase-3-disclosed slice by reclassifying documented Phase-3 misses to discontinued (approved programs untouched). n_advanced in the response is always 0 (advanced = censored, never scored) — there is no non-zero case to discount. Read the returned methodology_note field for the canonical caveat text.
Pass exclude_cik to get the base rate after excluding the company being evaluated (e.g. 'how often do TTR programs advance for OTHER companies?'). Returns verdict: insufficient_priors when fewer than 3 priors. Resolve gene aliases with bio_entity_resolve first — wants the canonical HGNC symbol.
Inputs: gene (required), as_of_date, exclude_cik.
bio_target_trajectoryBio Target Trajectory (quarterly crowdedness + max-phase time series)ARead-onlyIdempotentInspect
Quarterly trajectory for a target gene: how many companies and how many assets have been pursuing it each quarter, the maximum clinical phase reached, and the top-ranked companies at each point. Use to answer 'is this target getting more crowded?' or 'when did Phase-3 work on this target peak?' Pass a canonical gene symbol; resolve aliases first with bio_entity_resolve.
Inputs: gene (required), compact, end_date, start_date, +1 more.
bio_trial_readoutsTrial endpoint outcomes (met/missed, p-value) from SEC filings (PIT)ARead-onlyIdempotentInspect
Clinical-trial endpoint outcomes for a sponsor — primary endpoint MET / MISSED / MIXED, the p-value (as stated), and the endpoint name — extracted from that filer's SEC disclosures. The actual trial RESULT per program (a tier deeper than 'assets by phase' or 'approvals'): numbers a model reading a 10-K rarely surfaces. 2,442 readouts (Ph3 1,240 / Ph2 1,126 / Ph1 65; MET 1,224 / MISSED 1,182; p-value stated on ~1/3, endpoint named on ~99%).
Inputs: ticker_or_cik (required), phase ('1'|'2'|'3'), outcome ('MET'|'MISSED'|'MIXED'), as_of (PIT — only readouts disclosed before this date), limit.
Contracts / caveats (surface these): (1) Extraction, not ground truth — outcome is classification-grade. The drug↔outcome pairing can be wrong when one disclosure window mentions multiple trials. pvalue_outcome_conflict=true marks exactly these rows — the stated p-value contradicts the outcome (MISSED with a significant p, or MET with non-significant), so the (outcome, pvalue) pairing is a likely mispair (e.g. Keytruda MISSED p<0.001). Discount the outcome↔pvalue pairing on flagged rows (~12% of rows with a p-value); the drug / phase / endpoint fields are still fine. The sentence field is the announcement text — use it to sanity-check. (2) PIT anchor is filing_date (the disclosure date), same discipline as the CRL tool — as_of gates filing_date < as_of, leak-safe. (3) Tickered universe only (argos), so delisted / private sponsors are underrepresented. (4) Filer-anchored (is_own_trial) — competitor trials narrated in a filing are excluded.
Pairs with bio_pipeline (asset → its readouts) and bio_target_base_rate (v2 — the Phase-3 misses here are what reclassify its censored programs to negatives).
Inputs: ticker_or_cik (required), as_of, limit, phase, +1 more.
bio_warning_lettersFDA Warning Letters (late-stage enforcement by recipient — CDMO + sponsor)ARead-onlyIdempotentInspect
FDA Warning Letters by recipient — late-stage public enforcement actions. Use to answer 'has this company received a Warning Letter' / 'every CGMP warning letter to a CDMO since 2024' / 'OMQ enforcement against this sponsor.' Returns letter date + issuing office (CDER/CBER/OMQ/etc.) + subject (typically CGMP / Finished Pharmaceuticals / Adulterated) + response / closeout flags.
Inputs (all optional, but pass at least one of ticker_or_cik / company / issuing_office for a meaningful query):
ticker_or_cik— equity ticker or SEC CIK. Filters to letters where super-bio's matcher resolved the recipient to a super_bio sponsor.company— ILIKE substring match on company_name. Use this for CDMOs / contract manufacturers (Catalent, Lonza, Patheon, Thermo Fisher, etc.) that aren't in super_bio's sponsor universe but still receive Warning Letters.issuing_office— substring match: 'Office of Manufacturing Quality' / 'OMQ' / 'CDER' / 'CBER' / etc. The OMQ subset is the high-signal cohort for supply-chain trades (manufacturing-quality warnings frequently precede import alerts or product shortages).since— ISO date floor onletter_issue_date(default2015-01-01).min_confidence— float in [0.0, 1.0], default 0.85. Only applies when filtering by ticker_or_cik (forces tight matches). Set to 0.5 to also see name_overlap matches.limit— default 50, max 500.
Response shape: list of {company_name, letter_issue_date, posted_date, issuing_office, subject, has_response, has_closeout, matched_cik, matched_ticker, matched_method, matched_confidence} rows sorted by letter_issue_date DESC.
Critical caveat — interpret a NULL result correctly. Warning Letters are the LATE-stage enforcement signal. Most pharma FDA enforcement appears in inspection classifications (NAI / VAI / OAI) which are NOT in this table (super-bio §5 #25, pending FDA Data Dashboard API key). An empty result for a sponsor means 'no Warning Letter (yet)' — NOT 'no FDA issues.' For inspection classifications you'd need to wait for the OAI ingestion to land.
Verified spec cases (live 2026-06-18): company='Catalent' → 1 row (Catalent Indiana, LLC · 2025-11-20 · CGMP/Finished Pharmaceuticals/Adulterated · Office of Manufacturing Quality — the CAPR-class verify case); issuing_office='Office of Manufacturing Quality', since='2024-01-01' → all manufacturing-quality letters since 2024 (the supply-chain catalyst cohort).
Inputs: limit, since, company, ticker_or_cik, +2 more.
canada_feedCanada Feed (SEDAR+ material changes + Canadian issuers' US 424s)ARead-onlyIdempotentInspect
Kaleidoscope Console's Canada Feed: one timeline of Canadian issuers' SEDAR+ material change reports (every vouched report, typed and summarised by MARS) and their US 424 prospectuses (EDGAR; bank structured notes left out). Newest first. Each row: date, kind (mc material change / p424), event type and label, company (name, ticker, CIK, market cap), MARS id, SEDAR+ profile, province, sector, one-paragraph summary, accession (material changes: pass to sedar_document).
Filters: window days 7 / 30 / 90 / 180 / 365 (default 30); type: financing_offering Financing; acquisition_divestiture Acquisition / divestiture; merger_takeover Merger / takeover; management_change Management change; board_change Board change; operational_project Operations / project; legal_regulatory Legal / regulatory; earnings_guidance Earnings / guidance; dividend_distribution Dividend; other Other; p424 US prospectus (424); province (e.g. 'British Columbia', 'Canada, province not known', 'Outside Canada', 'Not known'); sector (GICS-like: 'Materials', 'Energy', 'Financials'... or 'Unclassified'); cap '100m' / '1b' / '10b' (minimum market cap; issuers with no cap drop out); q (issuer name substring, exact ticker or CIK); notes (show 424B2 note supplements — RY / TD file several a day); tickers / ciks (only these issuers; replaces Console's watchlist toggle).
Also returns Console's KPIs (events, issuers, financings, M&A, management & board, US 424s, equity raised in 424s) and facets (type / province / sector counts, each computed with the other filters on). Console shows the newest 1,500 rows; limit (default 25) / offset page through them.
Province: atlas EDGAR address when it says Canada, else MARS's record folded (names, postal codes, EDGAR codes), else GLEIF jurisdiction. Sector: MARS sector / industries by keyword, else SIC. Market cap: US listing's cap, else TSX / TSXV cap. Big issuers rarely file material change reports (TD 10 ever): for them the feed is mostly 424s.
Inputs: q, cap, ciks, type, +8 more.
capital_structureCapital Structure (market cap to EV, leverage, coverage, 8 quarters)ARead-onlyIdempotentInspect
A company's capital structure: share price, cover-page shares outstanding (with the filing they come from), market cap, short- and long-term debt, cash, operating leases and enterprise value; trailing-twelve-month revenue, EBIT, EBITDA, interest expense, net income and free cash flow; leverage (debt / EBITDA, net debt / EBITDA, debt / equity) and interest coverage (EBIT / interest, EBITDA / interest); EV / revenue, EV / EBIT, EV / EBITDA. Plus history — the same figures at each of the last 8 quarter ends — and recent_debt_issuance (notes and loans announced in 8-K Item 2.03, with rate and maturity; each row carries accession + filing_url of its 8-K). Same data as Kaleidoscope Console's Capital Structure report.
Banks, brokers and insurers (SIC 6000-6499) get book value, price / book, ROE and ROA instead of EV and leverage (is_financial).
Rules: TTM sums the last 4 reported quarters; year-to-date or full-year figures filed under a quarter label are converted to the quarter (listed in adjusted_fields). EBITDA = EBIT + D&A and is null when D&A is not tagged. EV = market cap + total debt - cash; preferred stock and minority interest are shown but not added (as in Console); ev_incl_leases adds operating lease liabilities. An ADR's market cap uses the ordinaries-per-ADS ratio. For filers reporting in another currency, EV and the EV multiples are null (market cap is USD, the balance sheet is not); ratios that do not mix currencies stay. checks flags a market cap that moved much more than the share price between quarters — a spin-off, reverse split the price data does not adjust for, or a share-count error; read the filing before relying on that quarter. as_of gives the picture on a past date using only filings and prices known then.
atm_usage (#2189): at-the-market program use from the 10-Q / 10-K — selling_now (sold in one of the last two reported quarters), the last quarter's shares sold, net proceeds, remaining capacity, sales agents, quote + accession as evidence, and a quarterly history; null when the company has no ATM on file.
Inputs: as_of, company_id, ticker_or_cik, history_quarters.
catalyst_calendarCatalyst Calendar (PDUFA / IPO lockup / debt maturity — forward + trailing dated catalysts with provenance)ARead-onlyIdempotentInspect
Forward (or trailing) calendar of dated NON-EARNINGS catalysts — PDUFA goal dates / AdCom dates, IPO lock-up expiries, and corporate-bond maturity walls. The 'what's coming' companion to upcoming_earnings, merged across event types into one uniform record so an analyst-LLM can reason across catalyst types with one shape.
Eleven event types (event_type filter):
pdufa — FDA decision dates. 8,177 events 2015→ from SEC filings + 2,654 from press releases, LLM-extracted; outcomes auto-resolved against Drugs@FDA. Status: pending / extended / approved / crl / withdrawn / projected (v1.5 calibrated PDUFA windows, is_confirmed=false) / unresolved (still pending or extended more than 180 days past the goal date with no approval, CRL or withdrawal found: outcome unknown, no longer a forward catalyst; since 2026-11-09). Confidence is certainty of the DATE — stated PDUFA ~0.9+, vague window ~0.4-0.6. is_confirmed=true means an explicitly stated date. Calibration vs Drugs@FDA: conf≥0.9 bucket = medAE 5d / 23% exact-on-date / median −1d (FDA acts on or before deadline). Big-pharma names (BMY/PFE/MRK/AZN/REGN) come from the news pass.
adcom — FDA Advisory Committee meeting dates (added 2026-06-18). Sibling of pdufa; status='adcom_scheduled' for forward or window-only dates, 'adcom_held' once an exact-date meeting is before as_of or its vote is public by as_of (since 2026-11-09). Filter by event_type='adcom' if you only want AdCom dates (previously these were nested under pdufa's detail.catalyst). Sample evidence: 'FDA's Pulmonary-Allergy Drugs Advisory Committee (PADAC) met on...' — narrative text from the SEC filing or PR.
ipo_lockup — IPO lock-up expiry dates. 2,572 genuine IPOs since 2020 (SPAC-excluded by default). Dates are deterministic (anchor date + lockup days). Terms enriched from IPO filings for 362/480 operating 2024-2026 IPOs; pre-2024 carry the 180-day default (confidence 0.5, is_confirmed=false). Status: upcoming / early_release_eligible / passed.
debt_maturity — corporate bond maturities (the 'debt wall' question). Wraps the tranche N-PORT-P panel — par values are fund-held sample only (~$3T), NOT total debt outstanding. Dates are contractual (exact). Confidence 1.0. Use min_par to floor on fund-held par (default $50M).
clinical_readout — forward Phase 2/3 trial data-readout dates (topline / interim / proof_of_concept) extracted from biotech press releases. Mostly vague windows — date_range is the signal, not event_date alone. Confidence reflects certainty of the READOUT DATE (exact stated date ~0.75; firm 'on track for Q4 2026' ~0.6; vague 'data in 2027' ~0.4). is_confirmed=true only on exact stated dates (rare). detail carries drug, trial_name, indication, phase, readout_type, is_preclinical (true = lower-value animal/NHP readout — filter via min_confidence if you want to exclude).
device_panel — FDA device advisory-committee meetings (added 2026-06-28). Device sibling of adcom. In both snapshot and history (past as_of queries route to history just like adcom).
device_pma — device PMA approvals (status='approved') + forward decision windows (status='projected'). Device sibling of pdufa. In both snapshot and history.
cms_coverage — Medicare coverage events. Status: granted / denied / restricted / delayed / pending. Snapshot-only (no history table for CMS) — past-as_of queries fall back to snapshot for this type, same as debt_maturity and clinical_readout.
cms_coverage and cms_star_ratings are the two CMS catalyst types. cms_star_ratings (added 2026-06-28, second batch) is the annual CMS Medicare Advantage Star Ratings release (forward ~October), per public insurer (UNH/HUM/CVS/ELV/CNC/MOH/ALHC/CLOV). detail carries overall stars + YoY + 4.0-cliff proximity. Snapshot-only (no history table).
foreign_permit (added 2026-07-03) — foreign government-approval calendar. MVP scope: CNSC nuclear licensing hearings (Canada), covering uranium/nuclear developers with a US ticker (e.g. NXE Rook I, DNN Wheeler River). detail carries {regulator, sub_type, us_ticker, is_developer, applicant, matter, decision_class, estimated_decision, outcome}. Snapshot-only — past-as_of falls back to snapshot, same as cms_coverage / cms_star_ratings. Broader jurisdictions may follow.
cms_rate_notice (added 2026-07-10) — CMS Medicare Advantage Rate Notice, per public insurer (HUM/UNH/CVS/CNC/ELV/...). Advance + Final milestones; the Final Rate Announcement lands ~statutory first Monday of April. Snapshot-only (no history table) — past-as_of falls back to snapshot.
tariff (added 2026-07-13) — Section 232 / 301 tariff actions across 12 commodities (copper, steel, aluminum, semiconductors, critical minerals, pharma, timber, autos, aircraft, trucks, polysilicon, drones). Federal-Register-sourced (proclamations / BIS / USTR notices — every row carries its FR document number + URL), deterministic, no LLM. History-eligible — the core query is a past-as_of read: "what was copper's 232 status on 2025-09-01?" → in_effect (50% since 2025-08-05). Milestones: investigation_opened → proclamation → (rescinded); status ∈ under_investigation / in_effect / etc. detail carries statute / commodity / milestone / status / effective_date / proclamation_ref / federal_register_url + the live current_status / current_rate_pct / current_since so one call answers "status on date X" without interval math. ⚠ rate_pct is deliberately NULL on multi-rate documents — a 232 proclamation cites market-share statistics, recommendations, and multiple enacted rates in one legal instrument; a wrong tariff rate on an auto-consumed feed is unrecoverable, so when a document enacts >1 rate we hand over rates_enacted[] + the FR link instead of guessing (copper's enacted 50% IS populated). Per-HTS rates are a v2 ask (needs HTS tables, not proclamation prose).
Inputs (all optional):
tickers — list of tickers; omit for cross-name screen. NB: ~30% of debt_maturity rows have ticker=null (debt-only financing subs without an equity ticker — Ford Motor Credit, Toyota Motor Credit, etc.); use the cik field on the returned row to identify those.
event_types — list of {pdufa, adcom, ipo_lockup, debt_maturity, clinical_readout}; omit for all.
window_days — forward (or trailing when include_passed=true) window in days from as_of. Default 90. Ignored when window_start/window_end set.
window_start / window_end — explicit YYYY-MM-DD bounds on event_date.
as_of — point-in-time date (YYYY-MM-DD). Default today. PIT replay is now honest (2026-06-15 + widened 2026-06-17): whenever the query REACHES INTO THE PAST — as_of < today OR include_passed=true OR window_start before today — pdufa + ipo_lockup queries route to catalyst_calendar_history (~5.4k rows, validity ranges; full resolved-outcome record approved/crl/ withdrawn back to 2015). Snapshot carries resolved outcomes only within its ±731d bake window, so deep include_passed base-rate sweeps MUST hit history to get the full picture. as_of today/future + forward-only window, or any debt_maturity / clinical_readout query, hits catalyst_calendar_snapshot. source_table per row + sources_used at the top reflect which path produced each row. debt_maturity + clinical_readout are NOT in history (maturity dates are static/public from issuance; clinical readouts are too new and disclosure-sourced — past-as_of of either falls back to snapshot).
data_health (on every response): per-event-type feed freshness + audit status so an auto-consuming system can detect a STALE or BLIND feed and refuse to trade on it. status ∈ {ok, stale, critical, unknown}; per_type reports only the event_types returned (freshness_class + is_stale + hours_since + row_count); audit_passing says whether what we serve matches reality (Federal Register / Drugs@FDA recall). When degraded, a top-level health_alert ⚠ leads the response. row_count=0 for a type ⇒ critical (zero rows is indistinguishable from 'nothing scheduled' — that's exactly how the adcom bug hid for weeks).
min_confidence — 0.0-1.0 floor (default 0.3). Drops vague-window rows for the conservative path; pass 0.0 to see everything.
include_passed — when true, include events whose event_date < as_of. Default false (forward-only). For pdufa+ipo_lockup, also routes the query to catalyst_calendar_history for full resolved-outcome coverage back to 2015.
min_par — for debt_maturity rows, minimum fund-held par in USD (default $50M = 50_000_000). Applies ONLY to debt rows; pdufa/lockup rows pass through.
sort_by — date (default, ASC) or confidence (DESC).
limit — max rows returned (default 50, max 500).
Response shape — one row per catalyst, common record: {cik, ticker, company_name, event_type, event_date, date_range, days_until, confidence, is_confirmed, status, headline, detail (jsonb per-type — see below), evidence, source_accession, source_type, source_date}.
detail blob per type:
pdufa: {catalyst: pdufa|adcom, drug, indication, review_type}
adcom: also vote (list of {question, yes, no, abstain, favorable} per voting question; favorable = the majority outcome favors approval, polarity handled), vote_outcome (favorable / unfavorable / mixed), vote_source (news:, the sponsor press release) and vote_source_date. All null when no tally is known; a vote shows only when vote_source_date <= as_of. Coverage: AdComs with ticker-tagged news, so some older or no-vote meetings have no tally.
ipo_lockup: {ipo_date, lockup_days, tranche, ipo_type, shares_locked, locked_pct_of_total, early_release_trigger, holder_types}
debt_maturity: {cusip, coupon, fund_held_par, n_funds, note} (note flags the fund-held N-PORT sample caveat)
Provenance: every row carries source_accession, source_type (filing/news/fda/tranche), source_date (PIT key — what made this knowable).
Console Catalyst Calendar parity (2026-10-12) — added inputs window (7 / 30 / 90 / 180, alias of window_days), types (alias of event_types), sector (SIC description substring or 2-digit SIC), min_market_cap (USD, approximate cap), q (company name / ticker substring). Two more event types: listing_deadline — open exchange-compliance deadlines (a notice is open until its deadline passes, the company regains compliance, or for a late-filing notice a later 10-K / 10-Q is filed); earnings — PROJECTED earnings dates (estimates, $10B+ market cap by default, earnings_min_market_cap to change; any cap when tickers is given). Every row now carries group (Healthcare / Capital markets / Policy / Credit / Earnings), market_cap_approx (earnings panel, not Console's guarded cap) and sector_sic; the response has by_group counts. Console defaults: confidence ≥ 0.4 and debt maturities OFF unless types / event_types names debt_maturity or include_debt=true.
Inputs: q, as_of, limit, types, +14 more.
chart_eventsChart Events (filing-native markers: earnings/dividend/split/insider/stake/campaign/formd/debt/contract/auditor)ARead-onlyIdempotentInspect
Filing-native event markers for one company, meant to be plotted on a price chart: earnings (beat/miss), dividends, splits, insider transactions (open-market buy/sell, 10b5-1 plan — grouped one marker per day+kind), 13D/13G ownership stakes, activist-campaign events, new Form D private offerings, debt obligations, material contracts (commercial/M&A/capital-markets/real-estate/credit only — employment and equity-award contracts are excluded as chart noise), and auditor changes.
Every marker carries accession (not a SEC.gov URL) so a caller opens the actual filing, plus filing_url; insider markers (one per day+kind) carry accessions (the Form 4s grouped into the marker, max 10). tone is +1 (beat/buy) or -1 (miss/sell) where applicable, null otherwise. Contracts are capped at 150 by dollar value and insider markers at 150 keeping every buy first — both per Console's own noise-control rules, not an arbitrary truncation.
Each lane degrades independently: if one source is unavailable or empty the rest of the response is unaffected — check warnings for any lane that failed, don't assume a missing lane means 'no events.'
Inputs: ticker_or_cik (required), end, start, types.
company_address_historyCompany Address History (relocation timeline from filing headers, 1994-2026)ARead-onlyIdempotentInspect
Business-address relocation timeline for one company, derived from SEC filing headers (1994-2026, ~27M filings). Each distinct address gets a first_seen / last_seen date and a count of filings.
Real moves are easy to spot: Apple's Cupertino → One Apple Park (~2017), Tesla's Palo Alto → Austin (2021), AOL's Vienna VA → Dulles VA, etc. For companies that never moved (Berkshire's Kiewit Plaza, JPM's 270 Park), you'll see a single address spanning decades.
Accepts ticker or CIK. Result is sorted chronologically by first_seen. Address normalization is whatever EDGAR shipped — ALL CAPS / inconsistent state codes / typos are common; identical addresses may appear as separate rows if a comma or unit-number style changed mid-filing-history. Each address carries accession + filing_url (the latest filing at that address, last_seen; open with get_filing) and first_accession (the first, first_seen).
Inputs: ticker_or_cik (required), limit.
company_dealsM&A History (one row per deal, SEC trail first, news second)ARead-onlyIdempotentInspect
A company's M&A history — one row per real deal, SEC first, news second (same rules as Kaleidoscope Console's M&A History). Each deal: the company's role (acquirer / target / seller), counterparty (company_id, cik, ticker, name, aka), kind (Merger / Stock purchase / Asset purchase / Acquisition), status (Closed / Pending / Terminated / Announced / Letter of intent / Talks / Rumor / Reported), announced and closed dates, value + value_source, whether a definitive agreement is on file, confirmed, communications count (425 / DEFA14A), advisers (Financial / Legal, which side) and an event trail (agreement, SEC filings, 8-K milestones, news) with accessions.
Rules: SEC merger / stock-purchase / asset-purchase agreements come first; merger vehicles are never shown as counterparties. Agreement amounts are not deal values (they are often termination fees) — values come from linked news coverage. News deals are deduplicated; investments reported as acquisitions are hidden (investments_hidden). Status comes from the filing trail when there is one (delisting / 8-K 2.01 = Closed, 8-K 1.02 = Terminated, proxy / tender / S-4 = Pending). A Closed deal always carries closed and closed_source (the filing label, or 'news' with a reported close date); news saying closed with no date shows Announced / Reported plus a status_note. 8-K milestones are the 8-Ks linked to that deal. When the company sold a business, role is seller and business_sold names it. A news-only deal still open after 2+ years with no filing is flagged stale. A news-only acquisition of a US public company with no SEC filing from the target is unconfirmed and hidden unless include_unconfirmed=true. Filings are returned by accession; open one with get_filing. For market-wide deal screens use mars_mergers_screen.
Inputs: limit, roles, status, company_id, +2 more.
company_derivative_exposureDerivatives on a Stock (funds' N-PORT options, swaps, CDS on one company: long / short)ARead-onlyIdempotentInspect
Derivatives that mutual funds and ETFs hold on one company's stock or credit, from their N-PORT filings: options, total-return and equity swaps, credit default swaps, futures and forwards that reference the company (or its parent). Index and sovereign references are left out.
Uses each fund's latest report in the last 270 days. Returns a summary (funds, families, contracts, long / short / unstated gross value, USD notional, how many funds lean long vs short), a rollup by kind, fund families, funds (adviser, family, gross / long / short / net value, USD notional, share of net assets, lean, accession and filing_url), the largest contracts (fund, counterparty and LEI, notional and currency, value, weight as a fraction of net assets, expiration, direction) and a quarterly trend (each fund's last report in each calendar quarter, history from 2019).
Direction: the filed payoff profile when there is one; options: a bought call or written put is long, a written call or bought put is short; swaps: long when the fund receives the reference leg (for a CDS, paying the other leg = protection sold = long credit); otherwise unstated. Gross = sum of absolute values; notional is summed only where filed in USD. For one fund's whole derivatives book use fund_derivatives.
Inputs: cik, limit, ticker, funds_limit, +1 more.
company_dilutionDilution & Capital Markets (offerings, converts, reverse splits, IPO)ARead-onlyIdempotentInspect
How a company raises money and what it does to the share count, over a lookback (default 5 years). Returns registered equity offerings (date, type, security, amount, shares, price, secondary / ATM / warrants flags, underwriters, accession; amended filings folded into the original), debt offerings kept apart, private placements and converts (instrument, amount, conversion price, discount, variable / reset flag, investors, maturity), charter changes (reverse-split ratio, authorized shares before / after, effective date, reason), Form D private raises (PIPEs), the IPO (price, proceeds, underwriters, lock-up tiers), cards and coverage.
Rules (Kaleidoscope Console): a conversion price under $0.10 that is also under 2% of the stock's last price is a conversion rate or par value read as a price — it is dropped and its variable flag ignored (conversion_price_misread=true), with or without a stated discount. On one filing and instrument, a row whose conversion price is under 2% of its twin's price (or was blanked as a misread) is dropped — the twin keeps the real price. Implausibly large amounts are dropped (amount_dropped=true); generic underwriter names ('Sales Agent', 'Placement Agent') are removed. Registered offerings cover roughly 900 filers a year, mostly small caps — 'none on record' does not mean 'none filed'. Filings are returned by accession; open one with get_filing.
atm_usage (#2189): how much of the ATM program was actually sold, per reported quarter (10-Q / 10-K) — selling_now, last quarter's shares sold, net proceeds, remaining capacity, sales agents, quote + accession, history; null when no ATM is on file.
Inputs: ticker_or_cik (required), years.
company_filingsCompany Filings (classifier tags + evidence sentence per filing)ARead-onlyIdempotentInspect
A company's SEC filings over a lookback window, each tagged with the classifiers that fired on it (score >= 0.8) and an evidence sentence per tag. Covers both the filer role (the company's own filings) and the subject_company role (13D/G filings naming it as the target).
Groups by classifier (NOT a 'theme' — Console's curated theme/direction subset lives in Console's own codebase, not here; each tag carries the raw classifier name + direction from classifier_names so a caller can map to themes on their own side).
424B / FWP (structured-note free-writing prospectuses) are capped at 100 — some issuers (notably big banks) file thousands of these per year and they are rarely what 'the company's filings' means to a human reader; they're still counted and included, just bounded.
Inputs: ticker_or_cik (required), forms, limit, years, +2 more.
company_fund_votesCompany Fund Votes (how funds voted at a company's meeting, N-PX)ARead-onlyIdempotentInspect
How mutual funds and ETFs voted at one company's shareholder meeting (SEC Form N-PX; Console's Fund Votes). Input a ticker_or_cik and optionally a meeting_date (default: the latest past meeting).
Predecessor CIKs (a reorg's old holding company) are included, so pre-reorg meetings show up.
Returns the meeting (date, ballot items, fund votes cast, managers voting, the lowest-support proposal), then one row per proposal (small wording variants are flagged minor and listed last): side (management or shareholder), description, fund votes, support by shares (FOR shares / (FOR + AGAINST/WITHHOLD shares)), support by fund votes, and how many managers voted For / Against / Abstain / Split. Pass proposal (a row's number) to list each manager's vote on it, with shares and fund votes. Also lists the company's other recent meetings.
Funds word the same proposal differently, so N-PX items MARS linked to the company's 8-K voting-results proposal are grouped as that one proposal; unlinked items group by side + wording. A linked row takes its side from the 8-K proposal type (funds sometimes mislabel management items as shareholder ones) and carries meeting_result: the meeting's own support (for / (for + against + abstain)), vote counts, the outcome as reported (null when not yet recorded, never inferred) and the 8-K accession. Unlinked rows have meeting_result: null.
Managers are fund families (an adviser's family name, or the fund's own name when it has no adviser on record). For unlinked rows, side is the N-PX vote source as filed; a few funds leave it blank, and those votes form their own 'unspecified' row. A manager is Split when its funds voted differently. N-PX support is among reporting registered funds only; compare it with meeting_result.meeting_support_pct.
layout='matrix' returns the vote matrix instead: one row per voting entity (the fund adviser, or the voter when it has none; rollup='family' rolls them up to fund family) and one column per ballot item, with director slates split into one column per nominee (nominees are matched on surname across filers' wordings). Each cell is the entity's vote across its funds (For / Against / Abstain / Split / No vote) with shares voted and the side most shares took (majority, majority_pct); a Split cell also gives the shares on each side. Small wording variants are kept as columns flagged minor and listed last. Every N-PX vote row counts, so an adviser's own report adds to its funds'. Investment managers' own N-PX reports cover say-on-pay only. Rows are sorted by shares voted; page with offset / next_offset.
Inputs: ticker_or_cik (required), limit, layout, offset, +3 more.
company_private_capitalPrivate Capital (Form D offerings, deduped funding rounds, Reg CF, placement agents)ARead-onlyIdempotentInspect
One company's private financing — Console's Private Capital (PVT) report. Returns:
form_d_offerings: SEC Form D Reg D offerings, amendments grouped into one offering by SEC file number (first sale, first/last filed, n_amendments, latest offering/sold amounts, investors, exemptions, related persons, placement agents with CRD). An offering a news round matched carriesmatched_roundsandform_d_days_before_news(positive = the Form D was on EDGAR before the round was reported).funding_rounds: news-derived rounds (mars, verified/vouched only), duplicates merged (reported_n= how many raw rows/reports folded in), each in onebucket— equity / debt / secondary / grant / fund_formation / other. Equity and debt are never blended. Aspirational or outlier valuations are hidden (valuation_hidden).reg_cf: Reg CF (Form C) crowdfunding offerings with amendments, progress updates, withdrawals, portal (intermediary + CRD), target and max amounts.related_persons,placement_agents: rolled up across every Form D.totals: equity vs debt sums from rounds, Form D amount sold.
Input: company (name / domain / UUID / ticker / CIK), or ticker_or_cik, or company_id. Works for private companies with no ticker. For the raw round list with investors use mars_funding_history; for an investor's deals use mars_investor_portfolio.
Inputs: limit, company, company_id, include_other, +1 more.
company_proxy_votesProxy Votes (meeting results, say-on-pay, next ballot)ARead-onlyIdempotentInspect
Shareholder-meeting vote results for one company (8-K Item 5.07) plus the items on the next ballot (DEF 14A filed after the latest result). Per meeting: each proposal with for/against/withheld/abstain/broker, support %, outcome, and whether it went against management; director elections are one row per nominee. Cards: last meeting, say-on-pay support and change vs prior, director support (per nominee: support %, withheld, slate average, lowest nominee, count under 80%), auditor ratification, shareholder proposals (5 years).
Rules (same as Kaleidoscope Console): support = for ÷ (for + against + withheld + abstain), broker non-votes excluded (ISS / Glass Lewis convention); withheld counts against and is also shown as its own number. Slate average = unweighted mean of the nominees' support %. Say-on-pay < 70% = warn (ISS responsiveness line); any nominee < 80% = warn. When a meeting's director table could not be split by nominee, director_support.granularity = 'slate' (one combined row). Directors are never marked failed by the vote count — they are plurality-elected and can stand below 50%. For say-on-pay and auditor votes the count overrides a contradictory extracted outcome (see outcome_note). A passing shareholder proposal or a failing management proposal is 'against management'. When the meeting date is missing or impossible the 8-K filing date is used (meeting_date_estimated=true).
Filings are returned by accession; open one with get_filing.
Inputs: years, company_id, ticker_or_cik.
company_red_flagsRed Flags (going concern, weaknesses, listing, risk drift, related party)ARead-onlyIdempotentInspect
Accounting, listing, risk and related-party trouble for one company in one call. Returns cards (going concern, material weakness, restatement, impairment, listing status, new risk factors, related party, bankruptcy 8-K — each with state clear / info / warn / hot, latest date, accession, count); financial events; exchange-listing events with the currently open notice; the latest 10-K/10-Q risk-factor comparison (new paragraphs ordered high → medium → low materiality) with history; related-party transactions (3 years); and the newest date per source.
Rules (Kaleidoscope Console): rows flagged low-confidence ('warn') are dropped. Going concern counts only when the text says going concern / substantial doubt, isn't about a subsidiary, segment or country, and market cap is under $2B (cap here is an approximation from the earnings panel). A listing notice is open until its deadline passes, the company regains compliance, or — for a late-filing notice — a later 10-K / 10-Q is filed. Event dates outside the filing's plausible window are dropped (date_dropped=true). No rows means no extracted event, not proof of a clean record. Filings are returned by accession; open one with get_filing.
Inputs: ticker_or_cik (required).
compsComps (trailing valuation and operating metrics vs peers, peer medians)ARead-onlyIdempotentInspect
Comparable-company analysis, as Kaleidoscope Console's Comps report shows it: trailing valuation and operating metrics for a company and its peers, with peer medians (subject excluded). Trailing only, no estimates.
Per company: market cap, EV, P/E, EV/Sales, EV/EBIT, EV/EBITDA, gross / operating / EBITDA / net / FCF margins, revenue and net-income growth (TTM vs the TTM a year earlier), ROE, net debt / EBITDA, dividend yield, 1-year total return (split- and dividend-adjusted close), plus the quarter used and its accession. Figures are computed exactly as capital_structure does (YTD quarters de-cumulated, cover-page shares, ADR caps via the ADS ratio). EV and EV multiples are null for non-USD reporters and when no debt is tagged; banks and insurers (SIC 6000-6499) get P/B instead of EV multiples.
Peers: pass peers (tickers or CIKs); by default the board's own peer group from the latest proxy (compensation peer group first, then the TSR / performance-graph set), else companies with the same 4-digit SIC (widening to 3 then 2 digits) closest in revenue. peer_selection says which, with the proxy accession. named_as_peer_by lists companies whose proxies name this one as a peer. as_of rebuilds the table on a past date. For one company's full capital structure and 8-quarter history use capital_structure.
Inputs: as_of, peers, n_peers, company_id, +2 more.
counterparty_exposureCounterparty Exposure (one LEI's funds, contracts, references + quarterly trend)ARead-onlyIdempotentInspect
One derivatives counterparty (a bank or dealer, by LEI) across funds' N-PORT filings. Uses each fund's latest report in the last 270 days. Returns the filed name and other spellings, a summary (funds, families, contracts, gross and net value, USD notional), a rollup by kind (FX forward, future, swap, CDS, call, put, swaption...), what the contracts reference (company, index, sovereign, fund), fund families, funds (adviser, family, contracts, gross / net value, USD notional, share of net assets, accession and filing_url), the largest contracts (fund, notional and currency, reference, value, weight as a fraction of net assets, expiration) and a quarterly trend (history from 2019).
Find an LEI with derivative_counterparties.
Inputs: lei (required), limit, funds_limit, trend_quarters.
credit_bdc_loan_bookBDC Loan Book (10-Q Schedule of Investments — mark / non-accrual / seniority / top positions)ARead-onlyIdempotentInspect
A public BDC's 10-Q Schedule of Investments, deterministically — portfolio mark (fair_value/cost, the headline credit-quality number), non-accrual % by value, seniority mix, top positions with terms (seniority / spread / maturity / mark / non-accrual), and a QoQ trend of the headline metrics. Primary-source SEC 10-Q data for the covered BDCs (53 BDCs, data through 2026-09-30; e.g. ARCC, BCRED, OBDC). Reading a Schedule of Investments by hand is miserable — this section is a genuine differentiator.
Input: ticker (required, a covered BDC), as_of (quarter cutoff, default latest), top_n (default 15), include_trend/include_sector_mix (default true), include_provenance (SEC accession per position, default false). The summary also carries non-accrual % by cost, share of FV marked below 90% of cost and NAV per share. The summary, each top position and each trend quarter carry accession + filing_url (the BDC's 10-Q/10-K for that quarter). A 100+ borrower book with zero non-accruals is flagged non_accrual_source_gap=true and its non-accrual figures are null — an extraction miss, never 0%. For every BDC side by side call bdc_summary. Unknown ticker → clean error: ticker not covered + the supported set. n_quarters as low as 2-5 is legitimate for newer BDCs (surface it, not a data gap). sector_mix is best-effort (private borrowers sparsely tagged — carries coverage_pct). Echoes as_of_effective + methodology_version.
Inputs: ticker (required), as_of, top_n, include_trend, +2 more.
credit_borrower_financing_historyBorrower Financing History (private middle-market — who lent, terms, mark trend)ARead-onlyIdempotentInspect
Who financed a private middle-market borrower, on what terms, and how the marks trend — for the ~6,300 private companies in BDC loan books. Which BDCs lent (facilities: seniority / principal / spread / PIK / maturity / origination / non-accrual (TERMS ONLY — marks live in mark_history + worst_current_mark)), a per-lender QoQ mark_history, and a summary with worst_current_mark (min mark across live positions — the single most useful credit signal). Breadth the news pipeline can't reach (~95% of these borrowers have no news footprint).
Resolve by ONE of company_id (strongest), name (trigram — CONSERVATIVE: different real companies share names, e.g. a $1M 'Eli Lilly & Co' middle-market loan ≠ the pharma giant; prefer a candidate that actually has a BDC position, and the response echoes match + sim so a report can hedge — do NOT assume across a fuzzy match), or domain. originated_date is NULL on ~66% of facilities (not disclosed — sorted last, not invented; the mark_history first quarter is the reliable lower bound on 'when'). people = the latest-Form-D-filing officer roster (person_id → persons_v2) when the borrower matches a Form-D issuer (~12% coverage; coverage:'none' otherwise — never faked). Echoes as_of_effective. Each facility and mark_history row carries accession + filing_url (the lender BDC's 10-Q/10-K).
Inputs: name, as_of, domain, company_id, +2 more.
credit_cross_held_loansCross-Held Loans (BDC NAV front-run screen — stale holders on a marked-down shared loan)ARead-onlyIdempotentInspect
BDC NAV front-run screen — shared private loans where an early-reporting BDC just marked the credit down / to non-accrual, but late-reporting BDC holders (still on an older 10-Q quarter) haven't caught it yet. Those stale holders are the short/put candidates over the [t_early, t_late] window. The same loan sits in multiple BDCs' books at STAGGERED report dates — the gap between them is the tradeable window.
Per shared loan returns each holder's latest mark (bdc_ticker, bdc_cik, report_date, filing_date, mark_ratio, mark_qoq_delta, nav_per_share, is_non_accrual, is_stale, tradeable, accession) + a signal block: mark_spread (max-min mark across holders; marks above 130% of cost are left out), fresh_markdown (biggest cut by a holder at the freshest quarter), stale_public_holders (holders on an older quarter who haven't reflected it — the front-run targets).
All from public SEC 10-Q Schedules of Investments — the signal is public-filing cross-referencing, NOT MNPI. Coverage: 53 BDCs, 4,009 loans held by 2 or more, data through 2026-09-30. tradeable=false on non-traded BDCs (e.g. BCRED / ASIF / MSDL — no liquid equity). Staggered dates are the point — holders are NOT aligned to a common quarter.
Params: signal ('all' default / 'markdown_divergence' / 'non_accrual_divergence'), holder_ticker (only loans this BDC is exposed to — the short-candidate screen), company_id (single-loan drill-down), min_holders (default 2), as_of (PIT cutoff), include_series (full per-holder mark history), limit.
Inputs: as_of, limit, signal, company_id, +3 more.
deal_deskDeal Desk (every new M&A deal with a listed side: terms, transfer agents, advisers, solicitors)ARead-onlyIdempotentInspect
Every new M&A deal (merger, stock or asset purchase) with a listed company on either side, acquirer and target side by side, the way a transfer agent or proxy solicitor reads a deal. Per deal: deal_id, kind, status (signed, proxy_filed, definitive_proxy, tender_open, voted, terminated, closed), value, sector (SIC-based, target's else acquirer's), sec_confirmed (false = news-only, $100M+), the agreement and the latest deal filing (accession, filing_url), terms (consideration, cash price per share, exchange ratio, which side's holders vote, premium as a fraction, HSR, expected close, outside date, break fees). Per side: transfer agent (and the filing it comes from), law firms (office, attorneys), bankers (fairness opinion), agents (exchange / paying / information / depositary), the merger meeting, and the solicitor: solicitor_merger_proxy once that side filed its merger proxy, else solicitor_last_annual_meeting (firm and fee; none_named when the proxy names none). Also counts (with transfer agent, holder vote at a meeting, acquirer issues shares, news-only), deals per day, sectors and transfer agents in the window.
Deals stay on the desk 90 days after signing; the desk is rebuilt every 30 minutes. Use deal_work(deal_id) for one deal's trail, vote result, contacts and holder base, and merger_votes_calendar for upcoming merger meetings.
Inputs: q, limit, offset, sector, +6 more.
deal_workWork This Deal (one desk deal: trail, vote, meetings, contacts, holder base)ARead-onlyIdempotentInspect
Work one Deal Desk deal (deal_id from deal_desk): the deal as the desk lists it, its stage (signed, proxy_filed (proxy or S-4), definitive_proxy, tender_open, voted, terminated, closed), the public trail per side (agreement, merger proxies, S-4 / F-4, tenders, delisting, 8-K items 1.01 / 1.02 / 2.01 / 5.01 / 5.07, each with accession and filing_url), holder vote results (for / against / abstain, for as a fraction of shares outstanding and of votes cast, turnout). Per listed side: whether its holders vote, its merger meeting from its own merger proxy (date, time, virtual, record date, vote required, majority of the minority, adjournment), solicitor and fee, transfer agent, law firms, bankers, agents, who to call (general counsel, corporate secretary, CEO, CFO), IR and media contacts from its press releases (name, title, IR firm, email, phone, source 8-K), and the solicitation sized up (shares outstanding, what a majority takes, institutions / insiders / everyone else as fractions, top 12 13F holders).
Deals leave the desk 90 days after signing.
Inputs: deal_id (required).
derivative_counterpartiesDerivative Counterparties (banks / dealers on funds' N-PORT derivatives, by LEI)ARead-onlyIdempotentInspect
Counterparties (banks and dealers) on funds' derivatives from N-PORT filings, one row per LEI: the most-filed name and other spellings, how many funds use it, contracts, gross value and the latest report date. Most funds first. Search with q (name words or an LEI; every word must match the main name or another filed spelling; main-name matches come first).
contracts and gross value add up every report in the last 270 days, so they run higher than counterparty_exposure, which uses each fund's latest report. Use the lei with counterparty_exposure for one counterparty's funds, contracts and trend.
Inputs: q, limit, offset.
discover_toolsFind the right toolARead-onlyIdempotentInspect
Call this FIRST when you're not sure which tool fits a question. Returns the top tools matching your query (semantic search over tool descriptions), each with its full input schema — so you can call the right tool in one round trip without a follow-up list_tools. The catalog has many tools spanning SEC-filing signals, fundamentals, 13F institutional ownership, 13D/13G activist positions, news/press-release search, and (incoming) biotech pipeline data. Use this any time the question is fuzzy or could plausibly match multiple tools.
Example queries:
'who are the largest institutional holders of a ticker'
'find news about FDA approvals in 2026'
'how is smart money positioned in semiconductors'
'manager track record vs the S&P'
'pipeline of a biotech as of a past quarter'
Returns top-k matches sorted by similarity, each with name, title, description, and inputSchema. Pick the best match and call it directly.
Inputs: query (required), k.
earnings_surprise_historyEarnings Surprise History (Foster SUE + post-earnings drift base-rate per ticker)ARead-onlyIdempotentInspect
For one ticker: its earnings-surprise track record (Foster SUE per quarter) and its post-earnings-drift base-rate — does it tend to beat (vs its OWN history), how big is the latest surprise, how does the stock behave in the weeks after it reports. Pairs with upcoming_earnings ('who reports → what to expect').
CRITICAL labels (worth reinforcing in any prompt):
Surprise is OWN-history SUE, NOT vs-Street consensus. EPS minus same-fiscal-period-prior-year EPS, scaled by trailing 8q stdev. Standard academic PEAD measure when no analyst estimates are available. Never say 'beat consensus' — say 'beat its own seasonal trend' or 'positive SUE.'
Both abnormal AND raw drift returned, labelled. abn_drift_41_bps is the alpha read (vs IWM beta); raw_drift_41_bps is what a long-only earns (beta-dominated). Don't conflate.
is_restated carried per row. PIT-validated drift signal SURVIVES dropping every restated surprise-quarter (+88bps full-universe, monotonic; survivors-only was +105bps — delisted names show no drift). Callercan exclude/flag if needed.
report_date and drift fields are DECOUPLED. report_date is set whenever the earnings 2.02 is matched (~79% of cik-quarters carry it). Drift fields (announce_return_bps, abn_drift_*, raw_drift_*) are populated only AFTER the +41-trading-day window elapses. So a JUST-REPORTED quarter reads report_date SET + drift NULL = 'reported, drift in progress' — NOT 'no earnings event.' Distinguish from non-filer (both NULL = no 2.02). True drift misses on settled quarters are ~1%; the rest of the 21% null are genuine non-filers reporting via 10-Q only.
cadence per row ('quarterly' or 'annual', new 2026-06-14). Foreign private issuers (20-F filers — ICLR, ASML, BABA, …) carry annual SUE (fiscal_period='FY'); domestic 10-K/10-Q filers carry quarterly. Never compare an annual SUE to a quarterly SUE as if same-cadence — present the summary's cadence_mix block and frame accordingly. SUE itself is currency-INVARIANT (own-history Δ/σ; the unit cancels) so SUE is comparable across currencies, but eps/revenue are in reporting_currency (ICLR USD, ASML EUR, BABA CNY).
Annual-cadence rows have report_date + ALL drift fields NULL by design. 20-F filers file no 8-K Item 2.02 so there's no announcement to anchor drift on. 'Annual SUE, no drift' is the correct, expected state for foreign filers — NOT a data gap. Coverage: us-gaap 20-F live; still empty: ifrs-full-only names (HAFN, BN — pending xbrl's IFRS extractor) and 6-K quarterly interims (futures v2 funnel).
Per-quarter drift is NULL'd for fiscal/calendar-collision names (since 2026-06-12; previously the adjacent quarter's value). Drift events are keyed by calendar quarter; when two fiscal reports collide in one calendar quarter (e.g. KMX Feb-FYE — fiscal-Q4 in April + fiscal-Q1 in June, both calendar-Q2), the orphaned fiscal quarter's drift fields read NULL. Treat NULL as 'drift not available for this quarter,' NOT zero. ~2.6% of quarters affected. The aggregate SUE→drift signal (+88bps full-universe) and sue/revenue_sue/beat/report_date are all correct — re-anchoring on the 2.02-per-fiscal-quarter was tested and rejected (it would have degraded the aggregate meaningfully; the calendar anchor sometimes catches the guidance pre-announcement, which is closer to where drift actually begins).
Rows are found by the CURRENT ticker (META returns every quarter, including the FB years); each row's era_ticker is the symbol traded at that quarter's end_date, and former_tickers lists the earlier symbols.
Returns N most recent quarters for the ticker (default 12, max 80) + a summary block: beat_rate, drift_after_pos_sue, drift_after_neg_sue, latest_sue, latest_eps, sue_trend (slope of recent SUEs), data_quality.confidence, cadence_mix (count of quarterly vs annual rows returned), and reporting_currency (the row's native currency for eps/revenue).
Verified on CROX: beat_rate 0.63, drift after +SUE +249bps vs after −SUE +67bps, n=57 — the name-level PEAD pattern consistent with the cross-section.
Inputs: cik, as_of, ticker, n_quarters, +1 more.
ecm_monitorECM Monitor (market-wide offerings, splits, converts, lock-ups, league table)ARead-onlyIdempotentInspect
Market-wide equity capital markets and dilution for listed US companies over a window (30 / 90 / 180 / 365 days). KPIs: primary offerings, $ raised ex-ATM, ATM programmes, resale registrations, reverse splits, authorized-share increases, variable-price converts, IPO lock-ups ending in the next 120 days. Lists for each, an underwriter league table (deals, deals as lead, $ credited) and a by-sector rollup. Filters: sector (SIC description or 2-digit SIC), under_2b (approximate market cap under $2B), q (company name or ticker).
Rules (Kaleidoscope Console): conversion prices under $0.10 that are also under 2% of the last price are misread conversion rates and are not counted as variable-price; implausibly large amounts are dropped; generic underwriter names are removed; $10 blank-check IPOs and warrant / unit lines are left out of lock-ups. Registered offerings cover roughly 900 filers a year, mostly small caps — 'none on record' does not mean 'none filed'. Sectors are SIC-based; market cap is approximate. For one company use company_dilution.
Inputs: q, limit, sector, window, +1 more.
fund_derivativesFund Derivatives (a fund's N-PORT forwards, futures, swaps, options + counterparties)ARead-onlyIdempotentInspect
Derivatives book for one fund (by SEC series_id, S000…) from its N-PORT filing: forwards (FX legs), futures, swaps (fixed rate / floating reference), options / swaptions / warrants and other derivatives. Returns net assets, the derivative value and its share of net assets, a by_type rollup (count, value, unrealized appreciation), the top 10 counterparties by gross value, and positions sorted by absolute value: counterparty, notional and currency, reference instrument, expiration, value_usd, unrealized appreciation and weight (a fraction of net assets).
Uses the fund's latest report on or before as_of; amendments (NPORT-P/A) are already applied. A fund with no derivatives returns empty lists and a no_derivatives warning. For the fund's bonds use tranche_fund_portfolio (include_noncusip=true adds loans, repos, cash and other positions without a CUSIP, and every call returns the asset mix).
Inputs: series_id (required), type, as_of, limit.
fund_equity_bookFund Equity Book (an ETF or fund equity positions, N-PORT)ARead-onlyIdempotentInspect
A fund's / ETF's equity holdings — given a fund SERIES id (S000…), return its equity positions (ticker, issuer, shares, value_usd), ranked by position size, amendment-resolved to the fund's latest N-PORT report. The mirror of fund_equity_holders ('who holds X' ↔ 'what does fund Y hold').
Source: SEC N-PORT-P fund-holdings panel (equity slice), fund-series grain. series_id is the FUND, not the fund family — one registrant (e.g. iShares Trust) files a separate N-PORT per series (iShares Core S&P 500 ETF = one series_id). Find a series_id from fund_equity_holders (each holder row carries it) or a fund's N-PORT.
Caveats: equity slice only (this is the fund's EQUITY book — its bond holdings are in the tranche corp-debt panel, its full book incl. derivatives/other isn't here); as_of is the fund's latest report_date; value_usd is the marked position. Coverage: latest-quarter + ~3.5y history (2022-11 → current); full 6.5y backfill in progress.
The header carries accession + filing_url (the N-PORT for as_of; open with get_filing) and each position its own accession + filing_url (the N-PORT it was read from).
Each fund row carries fund_type from the fund's latest SEC N-CEN: etf, mutual_fund, closed_end, money_market, interval, both (one series with ETF and mutual-fund share classes, e.g. Vanguard Total Stock Market = VTI + VTSAX; share_classes then lists each class with ticker and is_etf_class), other (separate account, UIT), or null when N-CEN doesn't cover the series (e.g. liquidated before filing one).
Inputs: series_id (required), limit, include_preferred.
fund_equity_holdersFund Equity Holders (which ETFs / fund series hold a stock, N-PORT fund-series grain)ARead-onlyIdempotentInspect
Which fund series / ETFs hold an equity ticker — from the SEC N-PORT-P fund-holdings panel, at FUND-SERIES grain (per-ETF, per-mutual-fund), amendment-resolved to each fund's latest report. Returns the ranked holder list (series_name, series_id, registrant_cik, shares, value_usd, as-of date) + breadth stats (n_funds, total shares, total fund-held value, top-5 concentration). Each holder row also carries its adviser and sub_advisers (name, CRD, CIK, role) from N-CEN; use funds_managed for every fund an adviser runs. cik beside each crd (#2287) is the adviser's SEC CIK: the CIK of the 13F filer that reports under this CRD (mars thirteenf_filer_crd_family, own_adv_crd / ncen_own_name; can be the parent's filer, e.g. Vanguard Capital Management -> The Vanguard Group 102909), else the CIK on its own Form ADV (investors_v2.adv_cik); null when neither exists (never inferred). When those two differ, ciks lists both, 13F filer first.
This is the granularity 13F cannot give. thirteenf_who_holds returns institutional MANAGERS (BlackRock, aggregated across all its funds); this returns the actual FUND SERIES (iShares Core S&P 500 ETF, Fidelity 500 Index Fund, …) with per-fund shares. Use this for 'which ETFs own X'; use 13F for 'which managers own X'.
Scope caveats (surface these): (1) SEC-registered funds only (mutual funds + ETFs) — insurers / pensions / direct holders / foreign funds are NOT in N-PORT, so this is the fund slice of ownership, not the full holder base. (2) Funds file N-PORT quarterly on staggered fiscal calendars, so 'as_of' varies per fund (each row is that fund's latest report) — it's a rolling snapshot, not a single date. (3) value_usd is the fund's marked position. Coverage: latest-quarter universe + ~3.5y history (2022-11 → current); full 6.5y backfill in progress.
Input: ticker (resolved to CUSIP internally) OR cusip (037833100-style, faster). include_preferred=false by default (common stock only; true adds preferred 'EP').
Each fund row carries fund_type from the fund's latest SEC N-CEN: etf, mutual_fund, closed_end, money_market, interval, both (one series with ETF and mutual-fund share classes, e.g. Vanguard Total Stock Market = VTI + VTSAX; share_classes then lists each class with ticker and is_etf_class), other (separate account, UIT), or null when N-CEN doesn't cover the series (e.g. liquidated before filing one). fund_type filters holders and the breadth stats — for 'which ETFs hold X' pass fund_type=['etf','both']. by_fund_type gives n_funds, value and shares per type across all holders. Each holder row carries accession + filing_url (the fund's N-PORT behind the row; open with get_filing).
Inputs: cusip, limit, ticker, fund_type, +1 more.
funds_managedFunds Managed (every registered fund series an adviser family runs, N-CEN + N-PORT)ARead-onlyIdempotentInspect
Every registered fund series (mutual funds, ETFs, closed-end funds) an adviser family runs, as adviser or sub-adviser — from SEC N-CEN adviser links joined to the N-PORT fund registry. Input a 13F manager cik (expanded to its adviser family of CRDs, e.g. State Street → SSGA Funds Management + State Street Global Advisors Ltd + …) or a single adviser crd.
Returns, per fund series: series_id, fund name, registrant (the filing trust — NOT the adviser), advisers and sub-advisers (name, CRD, CIK, role), net assets from the latest N-PORT and its date, and the N-CEN as-of date. Also per-adviser rollups (n_funds, net assets, roles) and family totals.
A CIK missing from the adviser-family table falls back to the CRD whose Form ADV carries that CIK (family_via: family_table / adv_cik_link). If no CRD is linked at all, n_funds is null (not 0) and candidate_crds lists ADV firms matching the filer name.
Notes: advisers are grouped by CRD under one canonical name (N-CEN spells some advisers several ways). Terminated adviser links are excluded. A series with no N-PORT registry row is still counted but has null net assets (n_funds_without_net_assets). Net assets are each fund's own latest N-PORT figure, so dates vary by fund. Pair with fund_equity_book(series_id) for a fund's holdings.
Each fund row carries fund_type from the fund's latest SEC N-CEN: etf, mutual_fund, closed_end, money_market, interval, both (one series with ETF and mutual-fund share classes, e.g. Vanguard Total Stock Market = VTI + VTSAX; share_classes then lists each class with ticker and is_etf_class), other (separate account, UIT), or null when N-CEN doesn't cover the series (e.g. liquidated before filing one). fund_type filters the funds list and totals; by_fund_type (family-wide and per adviser) gives n_funds and net assets per type. Each fund row carries accession + filing_url (the N-CEN filing that links the family to the fund; open with get_filing) and nport_accession + nport_filing_url (the N-PORT whose net assets the row shows; null when that filing is not on file).
cik beside each crd (#2287) is the adviser's SEC CIK: the CIK of the 13F filer that reports under this CRD (mars thirteenf_filer_crd_family, own_adv_crd / ncen_own_name; can be the parent's filer, e.g. Vanguard Capital Management -> The Vanguard Group 102909), else the CIK on its own Form ADV (investors_v2.adv_cik); null when neither exists (never inferred). When those two differ, ciks lists both, 13F filer first. The same cik is on the family and advisers rollups.
Inputs: cik, crd, role, limit, +1 more.
fund_vote_matrixFund Vote Matrix (every fund adviser's N-PX vote per ballot item at one meeting)ARead-onlyIdempotentInspect
How every mutual fund adviser voted at one company's shareholder meeting, from SEC Form N-PX, as a matrix: one column per ballot item and one row per voting entity.
Columns: director elections split one column per nominee (matched on surname, so 'Ron Sugar' = 'Ronald D. Sugar'), with block votes in 'All nominees'; then the other proposals. Each column has the nominee or description, side (management / shareholder), category, fund votes, for_pct_by_shares (a fraction) and the 8-K result when linked. Rows: the fund adviser (by CRD), else the filer: name, crd, kind, family / family_cik (the 13F filer it reports under, by brand name, else MARS's adviser family), shares, n_against_mgmt, and per column For / Against / Abstain / Split / No vote with shares and fund counts (a split shows shares per side and the majority side).
Minor items (few votes) are hidden unless include_minor=true. For proposal-by-proposal tallies or a family rollup use company_fund_votes. Each row carries accessions (the entity's N-PX filings behind its votes, max 10; open with get_filing).
Inputs: cik, limit, offset, ticker, +2 more.
geographic_revenueGeographic Revenue (by country and region as filed, de-overlapped)ARead-onlyIdempotentInspect
Revenue by country and region exactly as the company filed it in XBRL (10-K / 10-Q / 20-F), per fiscal year or quarter — no estimates, no gap-filling. Same data as Kaleidoscope Console's Geographic Revenue report.
Each period lists the slices as filed (country, region, or residual such as 'Rest of world'), each with value, % of total revenue, super-region and developed/emerging. Overlaps are removed: a region that contains listed countries counts only its remainder (contribution; e.g. Americas minus the United States), a 'Non-US' total that equals the listed foreign slices counts as their subtotal, and when a filer tagged two breakdowns the one with the most country detail is used. Then exposure sums the non-overlapping contributions by super-region (north_america, latin_america, europe, middle_east, africa, asia, oceania, multi_region, unallocated), by developed / emerging / mixed, and domestic vs foreign.
coverage_pct = geographic sum / total revenue. Many filers disclose only part of revenue geographically, so under 100% is normal and the gap is NOT estimated; over 120% is flagged is_anomalous (an overlap the reader could not untangle). Values are in the filer's reporting currency (unit), not converted. Negative slices are usually intersegment eliminations. Each period cites the accession and revenue tag used.
Inputs: as_of, years, period, company_id, +1 more.
get_company_profileCompany ProfileARead-onlyIdempotentInspect
One-company card: who the company is (CIK, name, ticker, SIC, exchange, entity kind such as operating company / fund / SPAC, sector, location, website, employees, short description) plus its latest reported quarter of fundamentals (revenue, margins, net income, EPS, cash flow, debt, YoY changes, with the source 10-Q/10-K). Ends with next: the exact calls for filings, holders, activist stakes, proxy votes and full fundamentals history. Start here for a single-company question.
Fundamentals are in the company's reporting currency (reporting_currency) and standard (accounting_standard, us-gaap or ifrs-full); don't compare margins across standards as if they were the same metric. With as_of, only a quarter first reported on or before that date is returned (point-in-time). Identity is current, not as of. No signal scores, prices or credit spreads (that data was retired in 2026-11).
Inputs: ticker (required), as_of.
get_filingGet Filing (open any filing by accession — text, exhibit, section, passages)ARead-onlyIdempotentInspect
Open one SEC filing by accession number and read its text from Kaleidoscope's own filing store (never a sec.gov link). Use it whenever another tool hands you an accession (company_deals, company_red_flags, company_dilution, insider_liquidity, company_proxy_votes, ...).
With just accession: filing metadata (company, CIK, form, filing date), a table of contents — exhibits (type, title, size) and sections (10-K / 10-Q / 20-F / S-1 items) — and the first page of the main document's text. Then:
exhibit ('EX-10.1', '10.1', or the exhibit_seq) reads that exhibit — merger agreements, credit agreements, press releases (EX-99.1) usually live here;
section (item code '1A' / '7' / '2.02', or a name like 'Risk Factors' / 'MD&A') reads one item;
query returns up to 5 passages where a phrase occurs instead of a page, each with the offset to page from;
offset + max_chars (default 20,000, max 100,000) page through long text; follow next_offset until it is null.
Coverage: main documents 1994 onward, exhibits 1994 onward, sections for the forms sections has itemised (10-K / 10-Q / 20-F / S-1 family, 8-K items like '1.01' / '2.02', SEC comment letters). A filing missing here is a coverage gap, not proof it doesn't exist.
Inputs: accession (required), as_of, query, offset, +3 more.
get_fundamentalsQuarterly FinancialsARead-onlyIdempotentInspect
Get financial fundamentals for a company from SEC XBRL data — quarterly values with year-over-year growth rates, margins, margin deltas, and PIT-integrity metadata. Includes COGS, inventory, capex, operating lease assets/liabilities, interest expense, dividends paid, research_development, total_current_assets and total_current_liabilities (for working capital / current ratio), total_assets and total_liabilities alongside the original revenue/margin/EPS/cash-flow set. period='annual' returns fiscal-year rows (10-K/20-F/40-F) instead of quarters; annual rows carry the values and PIT metadata but no YoY, margin or flag fields. Use this when asked about actual financial performance, profitability, balance sheet, cash flow, capex intensity, lease obligations, or interest burden.
ROIC + beta (2026-11-09): every row carries roic = operating_income x (1 - tax rate) / (total_debt + stockholders_equity), a ratio (0.25 = 25%). On quarterly rows it is ONE quarter, not annualized (AAPL ~16%); for 'what is the ROIC' use period='annual' (AAPL FY2026 69%). Null for filers that don't tag operating income (JNJ, CVX, XOM, most banks and insurers); runs high for captive-finance filers whose finance-arm debt isn't in total_debt (Ford). The top-level beta block is the stock's beta vs SPY over 2 years of daily total returns (with r_squared, n_obs, as_of); null under 250 observations; low r_squared is real, not an error.
Point-in-time integrity (2026-06-10): each row carries is_restated (bool — True iff any of revenue/net_income/operating_income/eps/operating_cash_flow for this period was later filed with a >2% difference) and first_reported_date (the date the value first became knowable, = filed_at of accession_number). Fundamentals already store as-first-reported by default (XBRL's extractor picks the earliest-filed accession), so for backtests that want the strictly-PIT subset, filter is_restated=False. IBM 2019-Q4 is the canonical restated case (re-presented in the post-Kyndryl-spinoff 10-K comparatives).
YoY + margin deltas (2026-06-10): rows now carry operating_income_yoy, gross_profit_yoy, fcf_yoy, capex_yoy, shares_outstanding_yoy (the numeric buyback/dilution rate that complements the existing diluting boolean), and gross_margin_delta / operating_margin_delta / net_margin_delta (margin-space deltas, NOT percent-change of margin — so a 0.0125 means the margin expanded by 1.25 percentage points, not 1.25%).
Sector-specific columns surface only when non-null (so a tech filing won't carry pension fields, an oil filing will carry exploration_costs): deferred_revenue / capitalized_contract_cost / contract_cost_amortization (SaaS-style), exploration_costs / mineral_properties (oil/mining), government_assistance (semis / subsidized industries), and pension fields: pension_liability (quarterly — net unfunded amount), pension_plan_assets and pension_obligation (annual / 10-K only — gross plan asset FV and gross PBO/ABO). Coverage of pension fields is ~5-7% of filers (only companies with defined-benefit plans).
Foreign-filer 20-F + IFRS coverage (2026-06-13 → 2026-06-17): the XBRL alias covers 20-F (foreign private issuers — ICLR / ASML / BABA / HAFN / BN / BNTX / TSM / Nubank / argenx / etc.) and 40-F (Canadian MJDS — BCE / Rogers / Teck / Kinross / Pan American Silver / Fairfax / Methanex / Eldorado / etc.). Each row carries TWO honest-labeling keys: reporting_currency (3-letter ISO — USD for domestic filers, CNY/EUR/HKD/JPY/CAD/etc. for foreign filers) and accounting_standard (us-gaap or ifrs-full).
Two read-it-first rules for foreign filers:
(1) Monetary values are in NATIVE currency, NOT USD-converted — BABA revenue comes back in CNY, ASML in EUR, BCE in CAD. Always read reporting_currency before quoting a dollar figure (no risk of misreading €32B as $32B). Caller does any FX conversion.
(2) NEVER compare margins or growth rates across accounting_standard — IFRS and us-GAAP differ on lease classification, asset revaluation, financial-instruments measurement, business combinations, etc. A 32% gross margin under IFRS is NOT comparable to 32% under us-GAAP without analyst-side normalization.
IFRS bank coverage uses revenue_bank_computed fallback (net interest + fee income — HSBC / BBVA / Lloyds / Barclays / ING / Santander / Westpac / Sumitomo / etc.). Bank revenue ~75-100% of true depending on filer — trading + insurance income missing. HSBC has the biggest gap (~25% under-stated). Banks that tag Revenue directly (WBC, NWG, Lloyds) match published numbers exactly. IFRS REITs use RentalIncomeFromInvestmentProperty fallback (Gazit-Globe etc.). Filers still missing: mixed-currency tag filings (AZN, NVO — file NetIncome in USD but Revenues in GBP/DKK) and 6-K interim results (futures v2 funnel).
Coverage: ~4,400 public companies + foreign-filer 20-F additions, quarterly data back to ~2009. Class-share aliases (GOOGL/GOOG, BRK-A/BRK-B, etc.), preferreds, and other ticker variants are auto-resolved via Atlas before the XBRL lookup — when aliasing is in play, the response includes a resolved_via_atlas block showing the queried ticker, the XBRL canonical ticker, and the CIK.
Inputs: limit, period, ticker, company, +2 more.
get_sentencesFiling Sentences by ThemeARead-onlyIdempotentInspect
Retrieve the actual sentence text from SEC filings that triggered a classifier. Use this after screening to see exactly what language caused a signal to fire. Requires a classifier name plus either a ticker or accession number. Returns the matching sentences ranked by classifier score (highest first). Use search_classifiers first to find valid classifier names.
Inputs: classifier (required), limit, ticker, filed_before, +1 more.
graph_aggregateGraph Aggregate (rank entities by an aggregate over their edges — the screen primitive) [beta]ARead-onlyIdempotentInspect
The graph SCREEN primitive — rank entities BY an aggregate over their edges (the cross-node complement to graph_neighbors' single-node walk). 'Which companies have the most 13D filers', 'most widely-held stocks', 'busiest placement agents', 'most-interlocked directors'.
Pick an edge_type (discover via graph_schema), a group_side ('src' or 'dst' — which END of the edge to rank), and a metric:
• n_edges — count of edges (e.g. # 13D filings on a company)
• n_counterparties — count of DISTINCT other-side entities (e.g. # distinct 13D filers — usually what you want for 'most-targeted')
• sum:<signal> — sum of a numeric edge signal (e.g. sum:value_usd on holds/dst = biggest aggregate institutional $ position; signals per edge come from graph_schema).
Examples: holds / dst / n_counterparties → most widely-held stocks; holds / dst / sum:value_usd → biggest $ held; filed_13d_on / dst / n_counterparties → most-targeted companies; placed_offering_for / src / n_edges → busiest placement agents; board_member_of / src / n_edges → most-interlocked directors.
Each row: resolved name + node_id + n_edges + n_counterparties + signal_sums + as_of_max (freshness of the underlying edges). Data is a nightly rollup (refreshed_at echoed) — current to ~last night, which is right for these rankings. Catalog-driven: every graph edge is screenable, new edges auto-join. min_value floors the metric; node_type filters the grouped node type.
Inputs: edge_type (required), group_side (required), top_n, metric, +2 more.
graph_entityGraph Entity (node + edge-type histogram) [beta]ARead-onlyIdempotentInspect
Node detail + what it's connected to — 'look before you leap'. Returns {node_id, node_type, name, edge_summary:[{edge_type, direction, target_type, count}]} (e.g. MSFT → insider_of:9235, invested_in:4, board_member_of:…). Edge-count histogram over every edge in mars.graph_edge_catalog this node touches — new catalog edges appear automatically. Pass a graph_resolve node_id (+ optional node_type to skip the probe).
Inputs: node_id (required), node_type.
graph_ledgerGraph Ledger (company/person overview — pre-joined investors, employment, peers, liquidity, health signals in one call) [beta]ARead-onlyIdempotentInspect
Tell me about this company / person / investor — call this tool for a full company overview or background profile, by name, ticker, or domain. Accepts a bare company name/ticker/domain directly via identifier (resolves internally, no separate resolve call needed) and returns a complete overview: profile, investors (with portfolio context), employment, peers, liquidity windows, and health signals, pre-joined in one call. Replaces mars_company_resolve + mars_funding_history + mars_search_articles — those three separate calls collapse into this one.
Each neighbor section is fused with the multi-table facts that make it interesting (a board member arrives with their OTHER board seats and whether those companies went bankrupt/delisted/were acquired), plus next_hops — the exact node_ids to dig deeper with graph_neighbors.
Pilot scope: serves a precomputed set of 50 companies/people/investors (mars.entity_ledger_v2) — growing, not yet full graph coverage. Every other company or person returns an honest found: false with a pointer to graph_neighbors for a live walk instead, never an error.
Each neighbor section carries a dig pointer (a ready graph_neighbors(...) call) for the rest of that edge type beyond what's itemized here — pointers are checked against this server's own live catalog before being returned, so a dig string you get back is always actually callable.
Inputs: node_id, node_type, identifier.
graph_neighborsGraph Neighbors (typed weighted-signal edges — the traversal primitive) [beta]ARead-onlyIdempotentInspect
The traversal primitive — a node's linked entities via typed edges, each with a signals dict carrying that edge's NATURAL weight (no unified edge weight; you pick). Returns edges:[{edge_type, direction, target_node_id, target_type, target_name, signals, as_of, source}]. Second-hop = call again on a returned target_node_id (an insider's OTHER board seats, a fund's co-investors, an institution's alumni). Filters: edge_types (discover valid ones + their signals via graph_schema), order_by a signal DESC, since+as_of (date RANGE — windowed/PIT walks), min_signal+signal_key floor, include_rumored (default false → acquired is definitive-only), limit. Edge set is catalog-driven (mars.graph_edge_catalog) — new edges are usable with no code change; consumer-contract filters (integrity, is_definitive) are baked in per the catalog.
Inputs: node_id (required), as_of, limit, since, +7 more.
graph_pathGraph Path (bounded relationship-path finder between two people/companies — warm-intro / degrees-of-separation) [beta]ARead-onlyIdempotentInspect
Find the relationship path(s) between two people (or a person and a company) — the 'Paths In' / Find-Path / warm-intro-meter primitive. Bounded BFS over the person↔person connected_to graph (professional/board/deal co-occurrence ties), depth ≤ max_hops (default 3). Returns the shortest paths as node sequences with the per-hop tie strength and an overall score (strength_product = product of ties, favors short strong paths; strength_min = weakest link).
from_id + to_id are graph_resolve node UUIDs. If to_id is a company, the final hop is a leadership edge (board member / officer / employee) into it, so a path reads person →connected_to→ person →board_member→ company. Top-level connected + min_hops are the warm-path meter ('reachable via N hops'). Bidirectional beam search — bounded, returns strong short paths (not every possible path). to_type optional (inferred).
Inputs: to_id (required), from_id (required), to_type, max_hops, +1 more.
graph_resolveGraph Resolve (any id → canonical entity node) [beta]ARead-onlyIdempotentInspect
Mirrored by the cross-family shared resolve tool (same underlying logic, callable regardless of which family you're granted) — kept under this name too since graph-family callers already use it. Resolve any identifier to a canonical graph node. ticker / cik / domain / crd / linkedin_url / UUID → {node_id, node_type, name} over the mars entity DAG — the entry point for traversal (feed node_id to graph_entity / graph_neighbors). Strong-ID first; a bare NAME matches company then institution (resolve investors by crd). When >1 canonical matches, all are in candidates. node_type ∈ company / person / investor / institution ('Duke University' → institution node for the alumni walk). Prototype.
Inputs: identifier (required).
graph_schemaGraph Schema (discover traversable edge types + their correct-use metadata) [beta]ARead-onlyIdempotentInspect
Discover what's traversable in the entity graph — the searchable catalog of edge types (from mars.graph_edge_catalog), so you know which edges exist AND how to use them SAFELY before walking. Pass a natural-language query ('edges from a person to a company', 'board seats', 'M&A') → the relevant edges ranked by meaning (E5 semantic search — searchable, not a dump: at hundreds of edges you can't read the whole schema). Each result carries the correct-use metadata inline: signals (rank axes), integrity_filter + edge_filter (the mandatory consumer gates, already applied by graph_neighbors), pit_column, src/dst_node_type, coverage_rows (density), freshness (live vs nightly-materialized-view — how current a walk of this edge is). Omit query to list all. Feed an edge_type to graph_neighbors.
Inputs: limit, query, node_type.
holder_profileHolder Profile (one 13F manager: style, portfolio, contacts, campaigns, position in a company)ARead-onlyIdempotentInspect
One 13F manager's profile (cik; a re-registered sub-CIK reads as its canonical parent), optionally against one company. header (brand and filed name, address, phone, website, investor type, adviser CRDs); details (style, turnover band and annual turnover, market-cap focus, family equity AUM, 13F AUM, positions, activism band with 5-year campaign / fight / settlement counts, top-10 concentration, return_4q and 4Q / 2Y excess return vs the S&P 500, all as fractions); overview_text; characteristics (weighted P/E, P/B, dividend yield, average market cap); sectors (weights and change vs the prior quarter); market-cap distribution (Mega >= $200B, Large $10-200B, Mid $2-10B, Small $300M-2B, Micro); industries; top holdings, buys and sells; contacts (compliance officer, regulatory contact, 13F signer, adviser executives); people (bios); campaigns (the family's activist campaigns). With company: sector_holdings / sector_buys, position (shares, estimated average cost, unrealized as a fraction, fraction of shares outstanding, first seen, last 12 quarters split-restated) and last_vote (its funds' N-PX votes at the company's latest meeting). 13F rows (holdings, buys, sells, sector_holdings, sector_buys, position and its quarters) carry accession + filing_url (newest 13F behind the row) and accessions (every family 13F behind it). Coverage artifacts (#2168, 2025Q4 on): when a security joins the SEC 13(f) list (or its old CUSIP leaves as the new one joins), a holder that already owned it starts or stops reporting it; that reads reporting_started / reporting_stopped (cusip_changed when the old CUSIP left and the new one joined) with no flow, not new / exit.
Inputs: cik (required), company.
holders_combinedHolders — Combined (13F + 13D/G + Form 4 insiders, one merged register)ARead-onlyIdempotentInspect
One register of who holds a stock: 13F institutional holders + 13D/G beneficial-ownership stakes + Form 4 insider positions, merged into a single list sorted by holder significance (13F by value, then 13D/G, then insiders). Each row carries source so you know which filing regime it came from, and its OWN as_of — 13D/G rows are as-of the FILING date, NOT a 13F quarter-end; do not treat the three sources as the same point in time. Composes thirteenf_holders_full + search_activist_positions + insider_holdings under the hood (reuses their existing canonical-CIK rollup / dedup logic rather than re-deriving it) — call those directly if you only need one slice with its full field set (13F flow signals, 13D/G campaign/intent fields, Form 4 derivative detail all get dropped down to the common fields here).
pct_os and value_usd are null where the underlying source doesn't carry them (13D/G doesn't report a value; 13F doesn't report %-of-shares-out in this tool) — null means not reported by that source, not zero.
Form4 rows are one line per insider per vehicle: ownership is 'direct' or 'indirect: ' (e.g. 'indirect: By Limited Liability Company 1'), and value_usd is shares x the last close (value_basis says which). Emptied vehicles (0 shares) and relabeled vehicles are dropped, and so are former insiders (no Form 4 at the issuer in 548 days, insider's owner_is_current) unless include_former_insiders=true. Each drop is counted in warnings. An indirect line the owner's later filings stopped reporting is KEPT with line_status = 'last reported , absent from N later filings' and missed_later_filings = N — weigh it as possibly stale. Every row carries accession + filing_url (its source filing: the holder's 13F for the quarter, the latest 13D/G, or the latest Form 4 behind the line; open with get_filing); 13F rows also carry accessions (all 13Fs for that quarter).
Inputs: ticker (required), limit, report_date, holder_types, +1 more.
insider_cluster_screenARead-onlyIdempotentInspect
Cross-issuer discovery of insider CLUSTER BUYING — issuers where min_insiders+ distinct insiders made DISCRETIONARY open-market buys in the window. The "multiple insiders are buying their own stock" signal (the highest-conviction insider pattern). Open-market buys ONLY — option exercises, awards, and planned sales are excluded as noise. Backed by 11.7M Form 4/5 transaction lines (incl. derivative) from ~4.6M filings, ~10-min fresh.
Returns issuers ranked by distinct-buyer count then total buy value: issuer_cik, ticker, sector, n_distinct_buyers, n_csuite_buyers, open_market_buy_usd, first_buy / last_buy. PIT-honest on filing_date; as_of defaults to today. Aggregates the price-sanitized value_usd.
Each cluster carries buyers: the insiders in it (largest first, up to buyers_per_cluster) with insider_cik, name, title, value_usd and up to 5 recent accessions (SEC accession numbers), so the cluster can be traced to filings.
Inputs: as_of, limit, window_days, min_insiders, +2 more.
insider_flowARead-onlyIdempotentInspect
Per-issuer Form-4 insider-transaction rollup — the Flow signal. "Is insider activity at company X bullish or bearish over the last N days?" Separates discretionary open-market buys/sells (the signal) from planned 10b5-1 sales, option exercises, tax-withholding, and awards (noise). Backed by 11.7M Form 4/5 transaction lines (incl. derivative) from ~4.6M filings, mid-2003 to present, ~10-min fresh.
Returns: n_buyers / n_sellers / n_csuite_buyers / n_csuite_sellers, open_market_buy_usd / open_market_sell_usd / planned_sale_usd, net_discretionary_usd (open-market buys − sells; the one number the Flow formula consumes — planned sales are EXCLUDED because a pre-scheduled sale says nothing about what the insider thinks today), n_transactions, plan_flag_coverage, as_of_effective.
PIT-honest: keyed on filing_date (Form 4s land up to 2 business days after the trade, so trade-date keying is lookahead). as_of defaults to today. plan_flag_coverage=false means the window reaches before 2023-04 (when the SEC 10b5-1 checkbox became mandatory) — planned-vs-discretionary is less reliable there. Dollar figures aggregate the derived, price-sanitized value_usd (never raw filer totals, which reach $300T from price-column typos).
include_participants=true adds participants: the top insiders behind the numbers (by dollar value, up to participants_limit), each with insider_cik, name, title, side (buy / sell / planned_sale), value_usd, n_transactions, last filing date and up to 5 most recent accessions (SEC accession numbers) so every figure can be traced to a filing. For every row, call names_transactions(issuer_cik=...).
Inputs: as_of, ticker, issuer_cik, window_days, +2 more.
insider_historyARead-onlyIdempotentInspect
One insider's Form-4 transaction trail + their OWN-history baseline — the audit surface behind an 'unusual buy'. Returns recent transactions (class, shares, value_usd, filing_date, is_planned) plus baseline stats computed from the insider's history up to as_of: n_open_market_buys, n_open_market_buys_5y, median_buy_usd, last_buy_date, is_first_buy_in_5y, is_outsized_vs_own_median. A CEO who buys every quarter is not a signal; a first buy in five years is — this lets you see which.
PIT-honest: both the trail and the baseline are computed from filing_date <= as_of, so a historical query never sees future trades (the baseline is NOT read from the as-of-today baselines table). Aggregates price-sanitized value_usd.
Code G rows are class gift with gift_kind = gift | intra_owner_transfer; their value_usd is ESTIMATED (value_estimated=true, market close x shares) and labeled in value_note. price_market_ratio far from 1 means the price was filed in local currency (e.g. BRL); value_usd is already converted. Each transaction carries accession + filing_url (its Form 4; open with get_filing).
Inputs: insider_cik (required), as_of, limit.
insider_holdingsARead-onlyIdempotentInspect
Each insider's current standing stake, from their initial Form 3 and every later Form 4 and 5, not a sum of trades (a sum understates founders who owned stock before becoming insiders). Ask by person (owner_cik, person_id or name) for their positions across companies, or by company (ticker, issuer_cik or company_id) for that company's insiders. Natural persons only by default (real_people_only).
A derivative position's shares are the underlying share count; never add them to common shares (both carry is_derivative). shares_bounded=false means an implausible filed share count was withheld. A missing position is not a zero holding (filings before 2003 are not covered). Holdings are as of today; a past as_of returns unsupported_as_of.
Each row is one owner and holding vehicle (nature_of_ownership). line_status flags a doubtful line: 'continued as ' (the vehicle was relabeled; only shown with include_inactive=true, and not to be added to the line it continued as) or 'last reported , absent from N later filings' (still counted, possibly stale). owner_is_current means the owner filed at this company within 548 days. Each row carries accession + filing_url (the latest Form 3/4/5 behind the line, latest_filing_date; open with get_filing).
Inputs: name, as_of, limit, ticker, +7 more.
insider_liquidityInsider Liquidity (Form 144, vesting/option calendar, 10b5-1 plans)ARead-onlyIdempotentInspect
When a company's insiders can sell, and have said they will — one call per company. Returns: Form 144 notices of proposed sale (last 12 months: seller, relationship, shares, stated value, value at last close, earliest sale date, broker, accession); the vesting / option-expiry calendar for the next 365 days with a monthly rollup; Rule 10b5-1 trading-plan adoptions and terminations since 2023 with which plans are still active; and cards (Form 144 $ last 90 days, vesting next 90 days, options expiring in 12 months, active plans).
Values (same rules as Kaleidoscope Console): when MARS flags a Form 144 value implausible, the stated value is withheld and shares × last close is shown with estimated=true. When a stated value is more than 3× off shares × last close it is kept but flagged mismatch, and totals use the smaller. Calendar values are shares × last close — an estimate at today's price, not the value at the event. A person has at most one active 10b5-1 plan: a newer adoption, a termination or a cancellation closes the earlier one; a plan with no stated end is treated as expiring 15 months after adoption.
For one insider's liquidity across companies, or a market-wide scan of who becomes liquid soon, use insider_liquidity_calendar. Filings are returned by accession; open one with get_filing.
Inputs: ticker_or_cik (required).
insider_liquidity_calendarARead-onlyIdempotentInspect
When does an insider become liquid — option expiries, vesting dates, and active 10b5-1 plan sales. A trading desk reads these as scheduled supply; prospect research (the donor family) reads them as a capacity+timing signal ("who comes into money this quarter"). Thin reader over a daily-materialized, real-people-only, forward-looking calendar (91K events, ~2ms).
Three shapes: pass a PERSON (owner_cik / person_id / name) for their upcoming liquidity; an ISSUER (ticker / issuer_cik / company_id) for that company's insiders' events; or NO identifier for a forward date-window SCAN of everyone becoming liquid in the next days_ahead (optionally csuite_only=true) — the prospect-research entry point.
Contracts: shares is PROVENANCE, not sanitized (OTC penny-stocks file billions) — describe an event with it, never rank people by it. plan_sale event_date is a PAST anchor (currently-selling status, trailing 180d), NOT a future date; option_expiry / vesting are future-only. integrity_tier = verified (structured/rule) | vouched (qwen from a footnote). Historical as_of returns unsupported_as_of (a calendar is as-of-today).
Inputs: name, sort, as_of, limit, +8 more.
insider_proven_buysProven Insider Buys (recent buys by insiders with good records)ARead-onlyIdempotentInspect
Market-wide: recent open-market buys by insiders whose PAST buys beat the S&P 500. Returns rows (insider × company: first/last buy in the window, days bought, value, role, C-suite, company with approximate market cap and SIC sector, the insider's record summary), companies with 2+ such buyers, KPIs (buys, buyers, graded, proven, $), and leaders — the best records among this window's buyers (5+ scored buys, bought in the last 2 years).
grade: 'proven' (strong + good, default), 'strong', 'weak', or 'all'. Plan-buying days (employee purchase programmes) are excluded from the buy list as well as from the scoring. Market cap is approximate (earnings panel), sector is SIC-based.
Console's scoring of open-market buys vs the S&P 500 (SPY), price-only on unadjusted closes, calendar windows of 30/90/180 days. Buys by one insider in one stock within 30 days of the first count once, priced at the first buy. Excess = stock return − SPY return over the same window. Left out: returns above +200/300/400% (reused-ticker artifacts) and plan-buying days (8+ insiders buying the same stock on one day at an issuer with 3+ such days — employee purchase programmes). Grade (needs 3 scored buys at 90d): strong = 4+, ≥65% beat SPY, median excess ≥ +5 pts; good = ≥55%, median > 0; weak = median ≤ −5 pts and <45%; otherwise mixed. Past returns, not a prediction.
For one insider's full history or everyone who bought one stock, use insider_track_record.
Inputs: role, grade, limit, sector, +3 more.
insider_sector_pulseARead-onlyIdempotentInspect
Sector-level insider buy/sell flow — which SIC sectors are seeing net insider buying vs selling over a window. Per sector: n_issuers_bought / n_issuers_sold / n_issuers_active, n_buyers / n_sellers, buy_usd / sell_usd, net_discretionary_usd, pct_issuers_with_buying. Backed by 11.7M Form 4/5 transaction lines (incl. derivative) from ~4.6M filings. Pre-computed for the standard windows (30/90/180/365 days) at 4-digit SIC; custom windows, historical as_of and coarser rollups run a live PIT query (~0.1-0.2s).
rollup: sic (4-digit industry, default), major_group (2-digit SIC, e.g. 28 Chemicals incl. Pharma, 73 Business Services incl. Software, 60 Banks) or division (letter A-J: Mining, Manufacturing, Finance, Services...). sort by net_discretionary (default), buy_usd, sell_usd or pct_buying; order desc (default) or asc — asc on net_discretionary puts the heaviest NET SELLING sectors first. Paginate with offset/limit; total_sectors is the full count and totals sums buy/sell/net USD and issuer counts across ALL sectors (not just the page). n_buyers/n_sellers are per-sector distinct insiders and are not summed into totals (one insider can sit at issuers in several sectors).
Grouped on the SIC code with canonical sector labels; aggregates the price-sanitized value_usd; PIT-honest on filing_date.
Inputs: sort, as_of, limit, order, +3 more.
insider_signalARead-onlyIdempotentInspect
The scored insider-activity packet for one issuer — a single correct-by-construction call that collapses a laborious, trap-laden Form-4 pipeline into a packet a model can immediately reason over. All the deterministic work (own-history 5y baselines, cluster detection, per-insider 'unusual vs their own history' scoring, discretionary-vs-planned separation) is precomputed daily by the insider team; this reads one materialized row.
Returns a signal_class (cluster_buy / csuite_buy / net_buying / neutral / net_selling — a deterministic classification, NOT a buy/sell recommendation), the discretionary flow (net_discretionary_usd = open-market buys − sells, planned 10b5-1 EXCLUDED), cluster + C-suite-conviction + unusual-buyer flags, and a notable[] array of the top insiders with per-person context (first-buy-in-5y, x-vs-own-median, C-suite, class). The model supplies the world knowledge and narrative; the tool supplies deterministic facts.
Contracts: dollars are the price-sanitized value_usd (never raw filer totals); planned_sale_usd is surfaced but never netted into net_discretionary; plan_flag_coverage=false means the window reaches pre-2023-04 so the 10b5-1 split is partial (don't read net_selling as all-discretionary there). PIT: a historical as_of is NOT served from the daily table — v1 returns unsupported_as_of and points you at the transaction-level tools for backtests. Each notable[] entry carries accessions (that insider's open-market / 10b5-1 Form 4s at this issuer in the window, newest first, max 10; open with get_filing).
Inputs: as_of, identifier, issuer_cik, window_days, +1 more.
insiders_of_companyInsiders of Company (Form 4 reporting_owners with last filing + count)ARead-onlyIdempotentInspect
List every Form 4 reporting_owner (insider) for one issuer, with last filing date + count of Form 4s filed. Derived from SEC filing headers (1994-2026, ~27M filings).
Answers questions like 'who's been filing Form 4s at Apple', 'how long has Cook been an insider', 'are there any new directors filing this year'. Sorted by last_filing DESC so currently-active insiders are at the top.
Accepts ticker or CIK. Returns insider_cik + insider_name + form4_count + last_filing for each distinct reporting_owner. Note: a single human insider can have multiple CIKs over their career (one per company they reported at) or sometimes a single CIK they reuse — depends on EDGAR's quirks.
This is filing-header-only data — does NOT include the actual Form 4 transaction details (shares acquired/disposed, prices, ownership type). For trade-level analysis you'd need the Form 4 body parse, which lives elsewhere. Each insider carries accession + filing_url (their latest filing at this issuer, last_filing; open with get_filing).
Inputs: ticker_or_cik (required), as_of, limit.
insider_track_recordInsider Buy Track Record (graded vs S&P 500)ARead-onlyIdempotentInspect
How an insider's open-market BUYS performed afterwards against the S&P 500 (SPY), graded strong / good / mixed / weak / few. Two ways in: • a person (person_id or owner_cik) → their record plus every buy episode (date, company, role, value, price, raw and excess return at 30/90/180 days, implausible flag); • a company (ticker_or_cik) → the record of everyone who bought there, each scored on ALL their buys (any company), best records first. Per insider: grade, scored buys at 90d, hit rate (beat SPY) and median excess return at 30/90/180d, median raw 90d return, total bought, number of stocks, first/last buy, number left out.
Console's scoring of open-market buys vs the S&P 500 (SPY), price-only on unadjusted closes, calendar windows of 30/90/180 days. Buys by one insider in one stock within 30 days of the first count once, priced at the first buy. Excess = stock return − SPY return over the same window. Left out: returns above +200/300/400% (reused-ticker artifacts) and plan-buying days (8+ insiders buying the same stock on one day at an issuer with 3+ such days — employee purchase programmes). Grade (needs 3 scored buys at 90d): strong = 4+, ≥65% beat SPY, median excess ≥ +5 pts; good = ≥55%, median > 0; weak = median ≤ −5 pts and <45%; otherwise mixed. Past returns, not a prediction.
For recent buys by insiders with good records across the market, use insider_proven_buys.
Inputs: limit, owner_cik, person_id, ticker_or_cik.
macro_calendarMacro Calendar (CPI / PCE / EMPLOYMENT / GDP / FOMC — forward macro release dates, PIT-by-construction)ARead-onlyIdempotentInspect
Forward (or trailing) calendar of major US macro releases — CPI, PCE, EMPLOYMENT (payrolls + unemployment), GDP, FOMC meetings, and EIA weekly energy inventories (WPSR crude Wed + natural-gas storage Thu). The macro counterpart to upcoming_earnings and catalyst_calendar. Use to answer 'what macro prints are coming in the next N days' / 'when was the last FOMC' / 'is CPI in this window.'
PIT semantics: PIT-by-construction. Major-release schedules are published ~1 year in advance, so filtering release_date > as_of IS 'as known then.' Response echoes as_of_effective = as_of (no resolution needed). Reference implementation parity with sec_power_supply_curve and the rest of the as_of-honored catalog.
Inputs (all optional):
window_days— forward window in days fromas_of(default 30, max 365).as_of— YYYY-MM-DD reference date (default today).categories— restrict to one or more of {CPI, PCE, EMPLOYMENT, GDP, FOMC, EIA_PETROLEUM, EIA_NATGAS}. Omit for all.include_passed— when true, returns a TRAILINGwindow_dayswindow (release_date in [as_of - window_days, as_of]). Default false (forward-only).limit— max rows returned (default 50, max 500).
Coverage: 5,383 release dates, 1947 → 2028. FOMC schedule is static (synthesized from FOMC's published meeting calendar; release_id = -1 on those rows); CPI/PCE/EMPLOYMENT/GDP come from FRED's release calendar. EIA weekly inventories now covered (added 2026-07-11): EIA_PETROLEUM (WPSR, Wed 10:30 ET) + EIA_NATGAS (gas storage, Thu 10:30 ET), 2010 → 2028, generated from EIA's pre-announced weekly schedule with federal-holiday shifts (PIT-by-construction, source='eia_static'). Release DATES only — inventory values live with the stock-data / prices feeds. Rare stacked-holiday weeks are a documented ±1d approximation.
Response shape: list of {category, release_name, release_date, source, release_id} rows sorted by release_date. Plus as_of_effective (= as_of), window (resolved bounds), n_returned.
Verified live (FRED team smoke goldens 2026-06-14): 30d after 2025-01-01 returns EMPLOYMENT 2025-01-10 + CPI 2025-01-15 + FOMC 2025-01-29 + GDP 2025-01-30 + PCE 2025-01-31. CPI as known on 2025-01-01, next 3: 2025-01-15 / 2025-02-12 / 2025-03-12.
Inputs: as_of, limit, categories, window_days, +1 more.
macro_seriesMacro Series (FRED data — ALFRED vintages, honest first-print vs revised values for backtests)ARead-onlyIdempotentInspect
Point-in-time historical values for a FRED macro series — backed by ALFRED vintages so a query at as_of=2020-05-15 for UNRATE returns the first-print value (14.7) not the later revision (14.8). The honest PIT macro reader.
Why this is the right shape: every other macro source returns 'current best value' which silently puts revised data into a backtest of a strategy that ran on first-print data — lookahead bias. ALFRED vintages let you ask 'what did everyone see on day X' and get the truth.
Inputs:
series_id(required) — FRED id, e.g.UNRATE,CPILFESL,DGS10,VIXCLS,NFCI. See covered list below.start/end— YYYY-MM-DD bounds onobs_date(observation period).startdefault isas_of - 1 year;enddefaults toas_of.as_of— YYYY-MM-DD vintage cutoff (default today). Returns latest vintage ≤ this date perobs_date. 0 rows is a valid answer — means the print hadn't been released byas_of(e.g. UNRATE for 2020-04-01 obs at as_of='2020-05-01' = empty; the April report was released 2020-05-08).limit— max rows returned (default 500, max 5000 — full daily series can be large).
as_of_effective = MAX(realtime_start) among the returned rows — the true currency of the data at as_of. If no rows: as_of_effective echoes as_of with a no_data_before_as_of: true flag (the 'print not yet released' signal).
Coverage (44 series total, daily refresh):
24 ALFRED-true vintages (revision-prone, reconstruct first-print vs revised exactly):
CPILFESL,PAYEMS,UNRATE,INDPRO,CFNAIMA3,UMCSENT,ICSA,GACDFSA066MSFRBPHI(Philly Fed mfg),GACDISA066MSFRBNY(Empire Manufacturing), BAML credit indices (BAMLC0A0CM,BAMLC0A1CAAA,BAMLC0A3CA,BAMLC0A4CBBB,BAMLH0A0HYM2,BAMLH0A1HYBB,BAMLH0A3HYC),DCOILWTICO(WTI),DTWEXBGS(USD broad index),ECBDFR,THREEFYTP10(Adrian-Crump term premium), some intl rates (IRSTCB01JPM156N,IRSTCI01AUM156N,IRSTCI01GBM156N),TEDRATE.20
final_synthseries (daily/weekly market series — non-revised so synthesized first-knowable date as obs_date + 1 business day): Treasury yields (DGS1MO,DGS3MO,DGS6MO,DGS1-DGS30),DFF(Fed Funds),DFII10(10Y TIPS),T10Y2Y/T10Y3M/T10YIE(curve + breakeven),VIXCLS,NFCI(Chicago Fed NFCI),STLFSI4(St. Louis Fed Financial Stress),USEPUINDXD(EPU index).
Important caveat on final_synth: for the 2 recomputed weekly indices (NFCI, STLFSI4), final_synth means intermediate revisions are NOT represented — final values only. For market-priced series (yields, VIX, FX, oil), the synthesized date is correct because those don't get revised.
Smoke goldens (FRED team 2026-06-14, verified live): UNRATE 2020-04-01 first-print 14.7 at as_of='2020-05-15' (realtime_start=2020-05-08); UNRATE 2020-04-01 at as_of='2020-05-01' returns 0 rows (April report released 2020-05-08, so not yet knowable). Pair with macro_calendar for 'when does the next print drop.'
Inputs: series_id (required), end, as_of, limit, +1 more.
manager_proxy_votingManager Proxy Voting (how a manager's funds voted, N-PX)ARead-onlyIdempotentInspect
How a manager's funds voted their proxies in a proxy year (SEC Form N-PX; Console's Proxy Voting holder report). Input a 13F manager cik (expanded to its adviser family), an adviser crd, or an investor name, plus proxy_year (2024-2026; a proxy year runs July 1 - June 30, so 2026 = Jul 2025 - Jun 2026).
Returns: fund votes cast and the share cast against management (of votes with a known direction); a breakdown by category (shareholder proposals counted separately) with proposals opposed per category; proposals where most of the family's votes opposed management; say-on-pay votes most opposed; shareholder proposals most backed (voted FOR), out of all shareholder proposals voted. A proposal is one issuer + meeting date + side + description, however many funds voted it; counts are fund vote rows.
Coverage: N-PX notice reports (funds that held nothing to vote) are left out and listed as 'notice report: no votes'; a voting report with 0 votes is a coverage gap (coverage_gaps). Funds are linked to advisers through N-PX voter records, so a fund sub-advised by the family counts toward it. Source filings: accessions (top level: the voting N-PX filings counted, newest first, max 10) and on each listed proposal (the N-PX filings whose votes it counts).
Every adviser crd in the output has the adviser's SEC cik beside it (#2287; x_crd -> x_cik, a CRD list -> ..._crd_ciks): the CIK of the 13F filer that reports under that CRD, else the CIK on its own Form ADV; ciks lists both when they differ; null when neither exists. A row that already had its own cik keeps it.
Inputs: cik, crd, top_n, investor, +1 more.
mars_form_d_flowForm D Flow (SEC private-offering pulse by month, PIT-clean, w/ breakdown) [beta]ARead-onlyIdempotentInspect
Aggregate SEC Form D private-offering flow as a time series — "how much private capital was raised via Reg D exemptions by month?" — with industry / jurisdiction / entity-type breakdowns. The cleanest PIT-anchored private-capital pulse mars ships: every row is a real SEC filing, so filing_date is a genuine point-in-time date (the day it hit EDGAR), with no extraction ambiguity, no dedup, and no coverage cliff. 765K filings (410K of them pooled-fund), 2006 → today.
Required: from_date + to_date (max 20-year window) + bucket (week|month|quarter|year). Optional: industry (industry_group_type exact — the live values are descriptive title-case strings, e.g. 'Other Technology', 'Commercial', 'Biotechnology', 'Oil and Gas', 'REITS and Finance', 'Residential', 'Other Health Care', 'Investing', 'Manufacturing', 'Other' — NOT uppercase codes), jurisdiction_of_inc (FULL UPPERCASE name, e.g. 'DELAWARE' / 'TEXAS' / 'CALIFORNIA' / 'NEVADA' / 'BRITISH COLUMBIA, CANADA' — NOT a 2-letter code), entity_type ('Corporation' / 'Limited Liability Company' / 'Limited Partnership' / 'Business Trust' — LP is a decent fund-vehicle proxy), breakdown_by (industry|jurisdiction|entity_type|fund_type), as_of (PIT gate on filing_date).
Contracts: (1) NO integrity_tier filter — Form D rows are SEC-clean by construction. (2) exclude_amendments defaults true — amendments are retroactive updates that would double-count. (3) total_offering_amount = the offering REGISTERED (a ceiling, not necessarily raised); total_amount_sold = raised as of the filing date (can grow via later amendments). Report both; don't call the offering amount 'raised.' (4) as_of gates filing_date <= as_of — a past as_of reconstructs the pulse as it was knowable then, leak-free.
Fund flows (2026-07-18): fund_type ('Hedge Fund' / 'Private Equity Fund' / 'Venture Capital Fund' / 'Other Investment Fund' — exact title-case) + is_pooled_fund (bool) are now populated (410K pooled filings, ~54% of corpus), so PE/VC/Hedge private-fund flow is one query — e.g. fund_type='Venture Capital Fund' for the VC-fund-formation pulse; each bucket also carries n_pooled_funds. Complements mars_sector_funding_flow (news-extracted operating raises); the two are NOT summable.
Inputs: bucket (required), to_date (required), from_date (required), as_of, +7 more.
mars_funding_historyPrivate-Co Funding History (per company timeline) [beta]ARead-onlyIdempotentInspect
Full funding-round timeline for one private (or recently-IPO'd) company. Returns every disclosed round with round type, funding_subcategory (early_stage / growth_stage / public_equity / other), round_date, amount_usd, currency, integrity_tier ∈ {verified, vouched, reported, warn}. Also article count, and canonical investors (joined through deal_investors_v2 → investors_v2 with post-2026-07-06 family cleanup applied).
Input: company accepts any identifier — UUID / domain / CIK / ticker / name — resolved internally via mars_company_resolve.
PIT: as_of filters rounds by round_date <= as_of. Rounds with NULL round_date are excluded from the PIT gate (return them but note in warnings).
Integrity gate: default min_integrity_tier='vouched' filters to {verified, vouched} — reported rows (the deal rests on one secondary news report (no filing, wire release, major outlet or second outlet) and returns to vouched automatically when corroborated) and warn rows (failed extract_guard / range_sanity) are excluded. min_integrity_tier='verified' (30,670 rows, ~40% of vouched population as of 2026-07-08) means SEC Form D amount match OR ≥2 independent article extractions corroborating the same event. Use it when you need the strongest corroboration signal — e.g. a hedge-fund analyst wanting confirmed round amounts rather than single-article extracts.
Amount fields: amount_usd_converted is the FX-normalized USD value (via exchangerate.host at round_date; >95% coverage). Use this for cross-currency comparability and hedge-fund analyst queries. amount_usd is Grok's raw extraction — for non-USD deals may be the stated amount in the original currency. amount_usd_source is 'grok_usd' (USD fast-path), 'fx_converted_<CURRENCY>', or NULL (rare edge-case, ~460 rows). currency is the original stated currency (unchanged, for provenance).
Investors: each round returns canonical_investors[] with investor_id / name / type / crd / position ('lead' when is_lead=true, else 'participant') / role_canonical (the investor's TYPE in this deal — mars 16-bucket taxonomy: venture_capital / corporate_vc / angel / private_equity / hedge_fund / bank / underwriter / institutional / … ; null ~10% when the source string was a position not a type). Post-2026-07-06 family cleanup means this is materially trustworthy. An empty canonical_investors[] is GROUND TRUTH, not a coverage gap — it means the round's source filing/article didn't itemize backers, which is STRUCTURAL for IPO / debt / Reg D / private-placement rounds (Reg D 100% unnamed, IPO 29%, Debt 34%) vs Series A/B/C/Seed (~95% named). Each row carries has_named_backers (bool) + an investors_note on empties so you can tell 'no backers named in source' from a gap. So a name with only IPO/debt rounds (e.g. SpaceX's recent IPO/Debt/PP rows) legitimately shows the funding TRAJECTORY without namable backers.
Composability: chain to mars_investor_portfolio(investor.investor_id) to drill into an investor's other deals; chain to SEC-side get_company_profile(cik) for public tickers.
Inputs: company (required), as_of, limit, offset, +3 more.
mars_funding_searchPrivate-Co Funding Search (v2 — filter by sector / stage / date / amount / investor) [beta]ARead-onlyIdempotentInspect
Cross-company funding round search. Filter deals by sector / industries[] / round_types[] / funding_subcategory / amount range / date range / investor / country. Returns paginated rows enriched with company identity.
At least one filter required — unbounded scans return invalid_input.
Sector vs industries: sector filters on companies_v2.sector (broad, e.g. 'Technology'). industries filters on companies_v2.industries[] (specific, e.g. ['Artificial Intelligence']). Use industries for tighter thematic slices.
Investor filter: fuzzy-matches on investors_v2.primary_name and name_variants[]. Ambiguous name (>1 canonical investor) returns rows for ALL matching investors + warning telling the caller the query fanned out.
Integrity gate: only verified / vouched deals. reported (the deal rests on one secondary news report — no filing, wire release, major outlet or second outlet) and warn (extractor sanity failures) are excluded; there is no opt-in flag.
PIT: as_of gates round_date <= as_of. NULL round_date rows always included (extractor couldn't parse — spec design).
Sort: round_date_desc (default), round_date_asc, amount_desc, amount_asc.
Country coverage: ~86.5% of companies have a resolved country (mars backfill completed 2026-07-11, up from ~73%). NULL-country rows (~13.5% — mostly domain-less / genuinely ambiguous private entities) are still excluded when country is set, but results are now a reasonable representation of the substrate, not a heavy lower bound.
Amount fields: amount_usd_converted is the FX-normalized USD value (via exchangerate.host at round_date; >95% coverage). Use this for cross-currency comparability and hedge-fund analyst queries. amount_usd is Grok's raw extraction — for non-USD deals may be the stated amount in the original currency. amount_usd_source is 'grok_usd' (USD fast-path), 'fx_converted_<CURRENCY>', or NULL (rare edge-case, ~460 rows). currency is the original stated currency (unchanged, for provenance).
Composability: chain each result's company_id to mars_funding_history for full trajectory.
Event-mix — use the deal_kind filter (2026-07-18): funding_deals_v2 now carries a 9-value deal_kind (100% populated), so pass deal_kind='operating_raise' for clean company/VC raises. Without it, results mix the four event classes below:
The funding_deals_v2 panel currently captures four semantically distinct event classes under one 'funding round' umbrella: (A) operating-company primary raise — the semantic 'funding round'; (B) fund formation — asset manager raised LP capital for a vehicle (e.g. Apollo Hybrid Value Fund, Antares senior loan fund); (C) treasury issuance — bank/insurance issued debt to the credit market (e.g. UBS $20B Debt, Citi $17.5B Debt); (D) structured product — securitization or one-off instrument. Consumers asking about 'companies that raised money' want (A) only, but the current panel returns all four. When results are dominated by financial-services incumbents / PE firms at high dollar tiers, warn the user that Classes B and C are polluting the result set — this is honest reporting, not a bug in your query.
Inputs: as_of, limit, offset, sector, +12 more.
mars_investor_portfolioInvestor Portfolio (all deals for one investor, family-rollup default) [beta]ARead-onlyIdempotentInspect
All deals an investor participated in, joined with company identity. Powered by mars.deal_investors_v2 (post-2026-07-06 family cleanup — 617 memberships in 39 contaminated families were unset, so include_family=true is now materially trustworthy).
Investor resolver order: UUID → CRD (numeric) → adv_cik (numeric) → name-fuzzy.
Family rollup: include_family=true (default) rolls up parent + subsidiaries via parent_investor_id and family_canonical_name. Response echoes family_members_included so caller knows how many vehicles contributed.
Position filter: 'lead' returns rows where deal_investors_v2.is_lead=true; 'participant' returns is_lead=false. null returns both.
PIT: as_of gates deal round_date <= as_of.
Response envelope: top-level investor object carries family_canonical_name + AUM band + CRD so downstream analysis has context without a second call.
Inputs: investor (required), as_of, limit, offset, +9 more.
mars_mergers_for_companyM&A For Company (deals as acquirer / target / either) [beta]ARead-onlyIdempotentInspect
Every M&A deal touching a company (as acquirer, target, or either). Returns role per row + counterparty (name + company_id) + announcement_date + deal_value_usd + status + source URL + snippet. Source: Mars-resolved mergers.
Useful for 'what has Microsoft acquired in the last 5 years' or 'has Twitter been a target.' role='either' UNIONs both halves; 'acquirer' / 'target' constrain. Mars covers both private and public-company M&A; deal_value_usd NULL when undisclosed (~40% of smaller deals).
v2 (mars-v2 methodology, 2026-06-17): substrate swapped to UUID-keyed company_id so multi-domain entities resolve cleanly. company accepts ANY identifier — a company_id UUID (the chain-in identifier every other mars_* tool emits), a domain, a CIK, a ticker, or a free-text name (incl. bare/short names missing the 'Inc.'/'Corp.' suffix — resolved via the shared mars_company_resolve, 2026-09-22) — the same universal resolver every other mars_* tool uses. Response carries resolved_via ∈ {uuid, domain, cik, ticker, name} so the caller knows confidence. If the name is genuinely ambiguous, returns error='ambiguous' with candidates[] instead of guessing. Counterparty UUID is included on each row when known (chain-able into another mars_funding_history(company_id=...) call). dedupe=true collapses near-duplicate rows from multi-source wire coverage by canonical_key (default off — matches v1). Each deal row carries accession + filing_url (the merger agreement's filing, when one is linked; open with get_filing) and accessions (every SEC filing linked to the deal, max 10); both are empty for news-only deals.
Inputs: company (required), role, limit, dedupe, +1 more.
mars_mergers_screenM&A Screen (by industry / deal value / status) [beta]ARead-onlyIdempotentInspect
Cross-company M&A screener with filters: date range, industry, minimum deal value, status ('announced'/'completed'/'terminated'/'regulatory_review'). Returns acquirer + target (name, company_id, ticker), announcement_date, deal_value_usd, status, industry, merger_type, strategic_goals, source article URL + snippet, published_at. Source: Mars-resolved mergers from wire services.
v2 (mars-v2 methodology, 2026-06-17): substrate swapped to UUID-keyed company_id. acquirer/target UUIDs surface when known (chain-able). Multi-wire duplicate rows collapse on demand via dedupe=true (keyed on (acquirer_company_id, target_company_id, announcement_date) — UUID-based so trivial name variants collapse correctly).
IMPORTANT: industry is grok3-extracted free-text and drift-prone — for thematic slices use mars_search_articles. deal_value_usd is NULL for undisclosed deals (~40% of small M&A). Status values per v2: 'announced' / 'completed' / 'terminated' / 'regulatory_review' (lowercase). Mars covers both private and public-company M&A. Each deal row carries accession + filing_url (the merger agreement's filing, when one is linked; open with get_filing) and accessions (every SEC filing linked to the deal, max 10); both are empty for news-only deals.
Inputs: limit, dedupe, status, date_to, +6 more.
mars_private_credit_flowPrivate Credit Flow (seniority / spread / deal-type pulse by month, w/ breakdown) [beta]ARead-onlyIdempotentInspect
Aggregate private-credit deal flow as a time series — "what's the private-credit pulse this quarter, and how are spreads moving on senior-secured vs mezzanine?" A window into a market public HY spreads can't see (94K deals from BDC schedules + direct originations + fund raises).
Required: from_date + to_date (max 20-year window) + bucket (week|month|quarter|year). Optional: seniority ('first_lien_senior_secured' / 'first_lien_unitranche' / 'second_lien' / 'subordinated' / 'mezzanine'), deal_type ('direct_loan_origination' / 'portfolio_disclosure_bdc' / 'fund_raise'), lender_investor_id (UUID), breakdown_by (seniority|deal_type|lender|benchmark_rate), include_spreads (default true — adds median cash + PIK coupon bps per bucket), include_bdc_holdings (default false), pit_gate_on + as_of.
Contracts: (1) include_bdc_holdings defaults false — BDC quarterly disclosures (87.5% of rows) re-report the SAME held loan every quarter and would massively inflate flow totals; the default filters them to leave direct originations + fund raises. Set true only for a stock (held-book) view, never for flow. (2) principal_usd is native USD (BDC + SEC forms file in USD — no FX conversion). (3) integrity_tier verified/vouched only. (4) coupon coverage is partial (cash ~79%, PIK ~12%) — medians are over populated rows only.
PIT: pit_gate_on = 'announcement_date' (default — the deal's real timeline, best for descriptive series) OR 'first_seen_at' (leak-free — when the deal became knowable in the corpus; use this for honest backtests, but note it has a coverage cliff at the ingest date). as_of gates the chosen column <= as_of.
Note: industry cut is NOT available — industry_group_type is ~0.4% populated on this table (flagged to mars). Not summable with mars_sector_funding_flow / mars_form_d_flow.
Inputs: bucket (required), to_date (required), from_date (required), as_of, +7 more.
mars_reit_deal_flowREIT Deal Flow (per-company RE capital activity — net buyer/seller series) [beta]ARead-onlyIdempotentInspect
Per-company real-estate CAPITAL-ACTIVITY time series — is a public REIT/operator a NET BUYER or SELLER this period, deploying vs raising capital, at what activity level, as a year-over-year (or quarterly) series. reit's capital-EVENTS surface (acquisitions / dispositions / financings / capital raises / JVs / M&A per bucket + net_deployed_usd = acquisitions − dispositions). NOT operating fundamentals (FFO/AFFO/occupancy/NOI — separate build).
Read the amounts honestly: the trustworthy signal is COUNTS + NET DIRECTION (the SIGN of net_deployed_usd) — buyer vs seller, activity cadence. The $ magnitudes are trade-press-sourced (vouched), noisy, ~1.5% implausible outliers that pass range-sanity. Lead a section with net direction + activity level; treat $ as approximate. Each bucket returns median_deal_usd + max_deal_usd — when max ≫ median the sum is outlier-driven, so narrate the median or the count. Currency: trade-press (GDELT) amounts were never FX-converted (a PHP 77B deal was stored as $77B). reit flags those non-US / unknown-country deals as unverified, so they are left out entirely (not counted, not summed); counts for non-US issuers therefore cover Form D, news-path and any US-company deals only. Form D + news-path amounts and US-company GDELT amounts are summed. For non-US issuers treat counts and $ as a lower bound. No counterparties in v1 (real_estate_deals_v2 stores one party). Direction comes from deal_type, not party role.
PIT: as_of gates on first_seen_at (knowability, NOT announcement date); buckets by when the deal happened (announcement/close). Echoes as_of_effective. Input: ticker (or company_id), from_year, to_year, bucket (year|quarter).
Inputs: to_year (required), from_year (required), as_of, bucket, +2 more.
mars_reit_fundamentalsREIT Fundamentals (FFO / AFFO / NOI / occupancy / WALT per fiscal period) [beta]ARead-onlyIdempotentInspect
REIT operating fundamentals per fiscal period: FFO (+ per share), AFFO, NOI, same-store NOI, occupancy, same-store occupancy, debt maturing in 12 months, WALT (weighted average lease term) and cash / GAAP re-leasing spreads (mostly office, retail and industrial REITs), as an ascending series for one REIT. From earnings 8-K supplementals, ~224 of reit's 307-REIT universe, quarterly 2020 onward. The operating companion to mars_reit_deal_flow (capital events: acquisitions / dispositions / raises).
FFO is not cross-REIT comparable by default: each row's ffo_variant says which figure the REIT headlines (core / nareit / normalized / affo / null). Compare FFO across REITs only when both are 'nareit'; within a REIT, compare like variants (many REITs switch variants mid-series — ffo_variants_in_series lists them and ffo_warning fires when they mix). NOI scope also differs by REIT.
Missing is not zero: a null metric means the REIT didn't report it that period. The coverage block says, per metric, whether this REIT reports it at all (reported / not_reported / gap_needs_review) and in how many of its filings. Fill rates across all REITs: FFO ~66%, occupancy ~56%, AFFO ~46%, NOI ~43%, WALT ~22%.
Default returns verified and vouched rows; include_warn=true adds warn rows with their failed_checks. fiscal_period follows the REIT's own fiscal year; rows are filtered and ordered by period_end. PIT: as_of gates on the 8-K filing date (filed_at); echoes as_of_effective. A REIT outside coverage returns its coverage_note (e.g. non-traded, no earnings 8-K pattern). Input: ticker (or company_id / cik).
Inputs: cik, as_of, ticker, to_year, +4 more.
mars_sector_funding_flowSector Funding Flow (aggregate $ by sector by month, w/ breakdown) [beta]ARead-onlyIdempotentInspect
Aggregate funding-flow time series by sector — the macro/thematic tool. "How much money flowed into AI infrastructure by month over the last 12 months?"
Required bounds: period_start + period_end + aggregation ('week'|'month'|'quarter'). Max period 5 years — larger errors invalid_input.
Optional breakdown: breakdown_by='stage' splits each bucket by funding_subcategory; breakdown_by='country' splits by companies_v2.country; breakdown_by='top_investor' adds a top-10-investors-by-total-deployed table alongside the base series.
PIT: as_of gates round_date <= as_of (applied AFTER period_end via LEAST).
Sparse buckets returned as zero-count entries so a time-plot renders monotone.
Currency: total_usd aggregates amount_usd_converted (FX-normalized via exchangerate.host at round_date; >95% coverage). A row without a converted value falls back to amount_usd ONLY when its own currency is USD or NULL (2026-10-10 fix — a non-USD row with no converted value is excluded from the aggregate rather than counted at its raw foreign-currency face value, which was silently inflating/deflating sums by up to ~1,457x on the ~0.6% of rows this hit per names/mars's #825). Long-window comparisons (>3y) may drift with historical FX rates.
deal_kind (2026-07-18 — the event-mix fix): funding_deals_v2 now carries a 9-value deal_kind classification (100% populated), so the old event-mix pollution is fixable at the query. Pass deal_kind='operating_raise' for a clean VC / company-primary series (strips bank_debt / fund_formation / pipe / grant — e.g. May-2026's $1.03T decomposes to 35% operating_raise / 29% bank_debt), or breakdown_by='deal_kind' to see the full mix per bucket. WITHOUT a deal_kind filter, broad-sector totals still mix all event types (bank debt + PE fund formations can dominate Financial Services) — so always set deal_kind for a company-funding series.
Inputs: period_end (required), aggregation (required), period_start (required), as_of, +8 more.
mars_trending_acquirersTrending Acquirers (most-active buyers in last N days) [beta]ARead-onlyIdempotentInspect
Most-active acquirers in a trailing window (default 90 days), ranked by deal count then total disclosed value. Returns acquirer_company_id + name + deal_count + sum of disclosed deal values + an embedded recent_targets[] list (target name / company_id / date / value + source URL + snippet). The 'who's been on a buying spree' question. Source: Mars-resolved mergers.
v2 (mars-v2 methodology, 2026-06-17): substrate swapped to UUID-keyed company_id, so multi-domain entities collapse to a single acquirer record. Joins to companies_v2 to bring in domain list for the acquirer; targets remain at the per-row name (target_company_id surfaced when resolved).
Ranking is deal_count-first because ~40% of small-deal values are NULL (undisclosed) — count is the more honest 'trending' signal. total_value_usd sums DISCLOSED values only and is NULL (not 0) when none of an acquirer's deals disclosed terms — n_disclosed tells you how many of the deal_count deals carried a value. Undisclosed ≠ $0. For value-first ranking, ask and we'll add a sort_by param.
Dedup is mandatory and on by default (mars team 2026-06-17): the underlying mergers_resolved_v2 carries one row per news article (multi-wire coverage of the same deal lands as N rows). We DISTINCT ON (acquirer_company_id, target_company_id, announcement_date) BEFORE counting so 'Merck 6 deals' doesn't actually mean the same Terns acquisition counted 6 times. When two wires disclose different deal values, we keep the row that has the value (NULLs last). Each recent_targets entry carries accession + filing_url (the merger agreement's filing, when one is linked; open with get_filing) and accessions (every SEC filing linked to the deal, max 10); empty for news-only deals.
Inputs: limit, window_days, include_warns.
material_contractsMaterial Contracts (filed agreements — parties, terms, clause types)ARead-onlyIdempotentInspect
A company's filed agreements — every EX-10 material contract and EX-2.1 M&A agreement it filed or signed as a party — with counterparties (companies and people), dates, term, amount and its basis, governing law, and which clause types each agreement contains (change of control, non-competition, termination, indemnification, ...). Same data as Kaleidoscope Console's Material Contracts report.
Filters: since / as_of (filing date), agreement_types (e.g. 'credit_agreement', 'employment_agreement', 'license_agreement', 'merger_agreement', 'supply_agreement', 'lease_agreement'), families ('credit', 'employment', 'commercial', 'ma', 'equity', 'capital_markets', 'real_estate', 'governance', ...), has_clause (clause_type keys from clause_facet, e.g. 'change_of_control_definition', 'non_compete', 'non_solicit_employee'; clause_match any|all) and counterparty (name substring).
Each agreement lists clauses [{type, label, group, n, has_passages}], grouped substantive-first. Only merger agreements carry clause passages today; other types name the clause type only. Each row's source is SEC or SEDAR (Canadian SEDAR+ filings, 2016 on, for companies MARS links to the filer); open_with names the tool that opens the text (get_filing for SEC, sedar_document for SEDAR; no pointer when the SEDAR+ document is not stored). Before 2016 only agreements with clause headings have clause types. clause_facet counts agreements per clause type across everything matching the other filters (use its keys in has_clause; the most common clause_facet_limit types, default 60 — clause_types_total says how many exist); type_facet does the same per agreement type.
Amounts are as written in the agreement with their basis (total / annual / per-unit, ...) and direction; 'undisclosed' / 'redacted' when withheld. For an M&A agreement the amount is often a termination fee, not the deal value — use company_deals for deals.
Inputs: as_of, limit, since, offset, +10 more.
merger_votes_calendarMerger Votes Calendar (desk votes, merger meetings ahead, recent merger votes)ARead-onlyIdempotentInspect
Merger votes in three lists. desk_votes: every Deal Desk deal whose holders vote or consent (stage, latest merger proxy, each voting side's meeting with solicitor and fee, the result once held); votes ahead first by date, then voted / closed / terminated. meetings_ahead: every company's merger meeting still ahead, market-wide, from merger proxies (date, time, virtual, record date, vote required, majority of the minority, solicitor and fee, proxy accession and filing_url), joint-proxy duplicates dropped. recent_votes: merger approvals voted in the last 30 days (outcome, votes for / against / abstain, for as a fraction of shares outstanding and of votes cast, turnout).
Use deal_work(deal_id) for one deal's full picture.
Inputs: limit.
mining_company_reservesARead-onlyIdempotentInspect
For NI 43-101 projects (any commodity, ~1,500 projects, stage / NPV / IRR / P&P / M&I) use mining_projects — this tool is kept for one release (#2487) for the 7 gold majors' company totals below.
Company-level gold reserves (proven+probable Moz) + annual production + AISC ($/oz) for the major gold producers — structured, reconciled, and provenance-backed. Replaces scraping a 10-K's reserve tables (which mangles them — the '271 Moz' / '46.5 Moz production' misreads). Every figure is reconciled (contained ≈ tonnes × grade) and carries its SEC accession + source doc; implausible values are WITHHELD (NULL), never guessed.
Coverage (7 gold majors, FY2025): reserves for NEM (134.4 Moz) / GOLD-Barrick (146.4) / AEM (55.4) / GFI (54.9) / KGC (19.7); production + AISC for all 7 (incl. AU / HMY / AEM). Source: SEC SK-1300 Technical Report Summaries + Canadian AIFs (SK-1300 is modeled on NI 43-101, so the same extraction applies).
Inputs: ticker (one of NEM / GOLD / GFI / KGC / AU / HMY / AEM — omit for all), commodity (default 'gold'), include_deposits (default false — add the per-deposit reconciled detail rows with grade/tonnage/contained + citation).
Contracts / caveats (surface these): (1) Withheld headline — when reserves_completeness != 'ok', reserves_pp_moz is NULL by design (value withheld, not the row): AU/HMY report Mt-Moz where the schema expects kt-koz (a uniform ~1000× scale error the ratio reconciliation can't catch → gated out; HMY's South-African reef units don't reconcile in the schema either); AEM now serves via a stated_total tier. Their production + AISC ARE trustworthy and still serve. reserves_completeness ∈ {ok, stated_total, incomplete, implausible} tells you WHY a reserve is NULL vs merely absent. (2) Basis not normalized — reserves_pp_moz is as-reported; some filers state 100%-project basis, others attributable (Barrick 146 is 100%-basis; attributable ≈ 89). A basis flag is a sedar v2 item. (3) Do NOT sum the deposit detail — it includes the filing's own subtotal rows (e.g. 'Total Peñasquito'); use the summary reserves_pp_moz for company totals. (4) effective_date is sparse (NEM/GOLD have it). Refreshes on annual filing cadence.
Inputs: limit, ticker, commodity, include_deposits.
mining_projectsMining Projects (NI 43-101: stage, NPV / IRR, P&P / M&I, screen or company)ARead-onlyIdempotentInspect
NI 43-101 mining projects from SEDAR+ technical reports — Kaleidoscope Console's 'Mining Projects' (app and company report). One project per company at its latest technical report; earlier reports listed under it. Each project: name, company (name, ticker, CIK, MARS id), stage (production / feasibility / PFS / PEA / exploration / unknown), commodities, country / region, study economics (NPV US$M with its basis, IRR %, capex US$M, mine life, discount rate, tax basis, price deck), headline reserves (P&P), resources (M&I) and inferred lines (contained metal + tonnes), report date and accession (sedar_document reads it).
With an issuer (issuer / cik / profile_number / mars_company_id): the company report — every project with the latest report's full headline reserve / resource table (category, zone, cutoff, tonnes Mt, grades, contained, effective date), production first then by NPV. Works for issuers with no CIK.
Without an issuer: the screen. Filters commodity (a name: 'Gold', 'Copper', 'Lithium', 'Uranium', 'Rare earths'...), stage, country ('Canada', 'United States', 'Not stated'...), econ (NPV stated), npv_min (US$M), irr_min (%), q (project / company substring, exact ticker or CIK). KPIs: projects, companies, with economics, total NPV, median IRR, in production; facets for commodity / stage / country. Rows newest report first (Console's order) or sort by npv / irr; Console shows 1,500; limit (default 25) / offset.
Rules: reserves = P&P (else probable / proven), resources = M&I (else indicated / measured), the combined line first then most tonnes, zero-tonnage lines dropped; project name variants folded ('Almas' = 'Almas Project'); commodity symbols mapped (sb -> Antimony, U3O8 -> Uranium), grade-only elements and metal equivalents dropped. Economics on ~1 project in 5; effective dates sparse. For the 7 US-listed gold majors' company-level reserves / production / AISC (SK-1300) see mining_company_reserves.
Inputs: q, cik, econ, sort, +12 more.
operating_metricsOperating Metrics (per-company KPIs in the industry-standard set — cruise; SIC-routed)ARead-onlyIdempotentInspect
A company's operating-KPI time series in the metric set standard for its industry — the structured operating depth a model can't reliably pull from a 10-K itself. You pass a company (ticker/CIK) and get THAT company's values (per-company, not industry aggregates), from MARS's KPI panel extracted from 10-Q / 10-K text.
Covered industries: cruise lines (occupancy, passenger cruise days, APCDs, fuel cost per ton, net cruise costs), P&C insurers (combined / loss / expense ratios, premiums written and earned), homebuilders (net orders, closings, backlog, average selling price, cancellation rate, active communities) and airlines (ASMs, RPMs, load factor, TRASM / PRASM / CASM, yield). The response's industry field says which; each period lists its KPI values and derived_kpis (Q4 values MARS computed as FY minus Q1-Q3). Latest 20 reporting periods by default, newest first. A company with no KPI rows returns status='not_covered'; an unresolvable input returns status='unresolved'.
Input: ticker_or_cik. Heavy verticals (biotech, mining) have their own dedicated tools.
Inputs: ticker_or_cik (required), max_periods.
price_historyPrice History (daily/weekly OHLCV, split-adjusted or total-return)ARead-onlyIdempotentInspect
Daily or weekly price (OHLC) and volume history for a US listing. Source: atlas market.prices_v.
adjust controls which close series you get: split (default — split-adjusted, NOT dividend-adjusted; comparable price-action across a split like NVDA's 2024 10-for-1), total_return (atlas's adj_close — splits AND dividends reinvested), or raw (as-traded, what the ticker tape actually showed that day — a pre-split price will look huge vs today).
splits in the response lists every split event found in the window (date + coefficient) so a caller can annotate a chart without re-deriving them from the coefficient trail.
Ticker variants (BRK.B / BRK-B) are tried automatically. Pass a CIK and it resolves to the current ticker first — note a CIK only maps to ONE ticker at a time, so a renamed/relisted company's older price history under a prior ticker won't be found via CIK lookup alone; pass the old ticker directly if you need pre-rename history.
Inputs: ticker_or_cik (required), end, range, start, +2 more.
resolveLook up a company or personARead-onlyIdempotentInspect
Look up a company, person, investor or institution and get its canonical id: {node_id, node_type, name} (every match in candidates when ambiguous; a person name returns up to 5 namesakes, each with cik, current_company and current_title to tell them apart). For a LIST of people, use resolve_people (one call, matched against each company's roster). Best for tickers, CIKs, CRDs, web domains, LinkedIn URLs and company names; person names work too ('Colette Kress', 'Kress, Colette', 'KRESS COLETTE M'). Most tools take a ticker or CIK directly, so you don't need this first. To find people by name with filters (company, title, school), use a people-search tool instead. For a biotech gene/target/indication name (e.g. 'PD-1', 'HER2') pass kind='gene' — a different identifier space (HGNC symbols), returns {canonical, hgnc_id, aliases, confidence, kind} candidates.
Inputs: identifier (required), kind.
resolve_company_profileCompany Identity Card (addresses + EIN + LEI + former names + filer status)ARead-onlyIdempotentInspect
Rich identity card for one company, sourced from SEC filing headers. Returns: legal name, ticker, entity_type, EIN, LEI, state_of_inc, fiscal_year_end, business address, mailing address, phone, website, investor_website, former names (with from/to dates — e.g. Facebook → Meta), current filer status (Large/Accelerated/Non-Accelerated + SRC/EGC/WKSI overlays).
Accepts either ticker_or_cik (ticker 'AAPL' or CIK integer). For typos or private companies / non-tickered filers, call resolve_company_name first to get a CIK.
Distinct from get_company_profile, which is the short card (identity + latest fundamentals quarter + next-call pointers) — this one is the full SEC-header identity. Use this when the question is 'where are they HQ'd / what was their former name / filer status'.
Coverage: 8,287 CIKs have a cik_profile row (mostly operating companies with EDGAR-submissions enrichment); 6,673 have former_names history. For address history over time (relocations), use company_address_history. For non-cik_profile filers (small / private / inactive), the response still returns ticker + filer_status + former_names if available.
Inputs: ticker_or_cik (required).
screen_fundamentalsFundamentals Screen (cross-sectional XBRL screen — revenue/margin/growth/cash-flow filters)ARead-onlyIdempotentInspect
Cross-sectional screen over XBRL fundamentals: 'which companies have X financial profile', not 'show me company X's history' (that's get_fundamentals). Defaults to each company's LATEST quarter within the given scope, then applies every filter to that one quarter — a company whose latest quarter doesn't qualify is excluded even if an earlier quarter would have.
Numeric filters take _min/_max suffixes (e.g. revenue_min=1e9, operating_margin_min=0.15). Boolean filters (revenue_growing, margin_expanding, earnings_positive, cash_generation_positive, diluting, deleveraging) restrict to true when set.
us_only=true excludes Foreign Private Issuers (20-F) and Canadian MJDS filers (40-F) via form_type — no state-of-incorporation lookup needed, form_type is the SEC's own structural distinction.
Revenue filtering/sorting always applies a non-configurable revenue < $1T sanity guard — some non-USD filers (e.g. Telkom Indonesia, Ecopetrol) have nominal nine-figure-trillion nonsense in USD terms purely from unconverted foreign-currency reporting; reporting_currency is returned per row so you can see which currency a number is actually in.
view='full' returns every fundamentals_cooked column; default 'compact' returns identity + the standard margin/growth/flag set. fields adds specific extra columns (ignored if unknown). Every row carries accession_number (the 10-Q/10-K the quarter comes from) and filing_url.
Inputs: sic, ciks, view, limit, +30 more.
search_activist_positionsActivist & 13D/13G PositionsARead-onlyIdempotentInspect
Search 13D/13G beneficial-ownership stakes (>5% positions) and activist campaigns. This is the activist lens 13F can't give: 13D/13G is event-driven (filed within 10 days of crossing 5%), carries stated intent, and tracks the amendment trail — vs. 13F's 45-day-lagged quarterly snapshot with no intent.
One row per (filer, subject company) current stake. Filter by any combination:
subject_ticker → 'who owns >5% of / is running a campaign on TICKER'
filer_cik → 'what is this filer accumulating / campaigning on' (resolve a name to a CIK first via thirteenf_search_managers)
is_engaged_activist=true → genuine activists only (smart_money's classifier, activist_type 'activist'; this is the flag Console counts). is_activist=true only means the 13D form family (has intent), which also covers control, founder / insider, venture / PE and strategic holders; false → 13G (passive)
Omit is_activist for the full ownership picture (both activist and passive >5% holders in one call — this is the right default for 'who owns X' questions because passive 13G holders like BlackRock/Vanguard are critical context for any activist research)
campaign_status (active/dormant), tone (hostile/friendly/neutral), intent_type (board_representation, proxy_fight, merger_acquisition, control_acquisition, governance, management_change, strategic_alternatives, capital_return, etc.)
When subject_ticker is set, the response also includes an ownership_summary block that stratifies 13D (n_activist_filers; n_engaged_activist_filers = genuine activists) vs passive (13G) totals over CURRENT stakes only (latest_percent >= 5, latest_shares > 0, last filed within 3 years of as_of); stale, exited and sub-5% filers are counted in n_dormant_filers, not summed. Render this to the user so they see the full picture not just the 13D filer.
Returns the accumulation trail (first_percent → latest_percent, n_filings amendments), intent classification, and the raw 'purpose of transaction' text (13D only). Each position carries accession + filing_url of its latest 13D/13G (as of as_of when given). Every position carries BOTH latest_shares (count) AND latest_percent (% of shares outstanding) — quote both when describing a position; 'Impactive owns 4,045,984 shares' is incomplete without '(11.7% of shares out)'. Swaps and derivatives (13D Item 6): swap_notional_shares, swap_pct, derivative_long_pct, derivative_short_pct and economic_pct (latest_percent + derivative_long_pct), cited by derivative_accession / derivative_filing_url / derivative_filing_date; null when the filer discloses none. Quote both numbers, e.g. '9.9% owned plus 13.0% in cash-settled swaps = 22.9% economic exposure'. The swaps carry no votes. ownership_summary adds total_derivative_long_pct and total_pct_economic. Most filings are passive boilerplate — the activist signal is is_engaged_activist=true (plus intent_type for what they want). Coverage: structured-XML era only (Dec 2024 → now); intent_* fields are LLM-classified weekly so a brand-new 13D may have raw purpose text before its intent_types/summary populate. Refreshed every ~4 hours.
Coverage gap: foreign private issuers (20-F filers — Indonesian / Chinese / Israeli ADRs etc.) typically don't file 13D/13G under US rules, so foreign-PI subjects return empty. Same applies to insider data (Form 4) — structural to how non-US issuers report.
Examples:
'Who's running an activist campaign on RPAY?' → subject_ticker='RPAY', is_engaged_activist=true
'Who owns >5% of VAC?' → subject_ticker='VAC' (omit is_activist for full picture)
'Active hostile campaigns demanding board seats' → campaign_status='active', tone='hostile', intent_type='board_representation'
'Everything Pershing Square holds >5% of' → filer_cik=1336528
Inputs: tone, as_of, limit, sort_by, +12 more.
search_classifiersFind Filing ThemeARead-onlyIdempotentInspect
Search for available binary classifiers by keyword. Searches classifier name, human-readable name, and description. There are 531 classifiers detecting events in SEC filings: FDA approvals, drill results, demand signals, risk indicators, etc. Use this tool first to discover classifier names before screening.
Inputs: source, keyword.
sec_comment_lettersSEC Comment Letters (staff reviews with the company's answers, counsel, market screen)ARead-onlyIdempotentInspect
SEC staff comment letters (UPLOAD) and the company's answers (CORRESP), as Kaleidoscope Console's Research report shows them.
With ticker_or_cik: the company's reviews, newest first. Each review covers one target filing (e.g. 10-K for FY2024, an S-1) with opened / closed dates and the staff letters in order. Each comment has seq, topic (non_gaap_measures, revenue_recognition, segment_reporting, ...), ask_type, the ask, the disclosure it points at, and the company's answer (revised / will_revise / declined / clarified / provided_info + summary). Company response letters carry date, signer (name, title, firm) and accession. no_staff reviews are answers whose staff letter isn't on EDGAR (comments given by phone, common for funds). Also answer_totals, topic_totals and counsel: outside law firms and auditors on the company's SEC correspondence (letters, first/last, lawyers) and the company's own signers.
Without a company it is a screen: staff comments filed in a window, filtered by topics / target_form, newest first, one row per letter with the matching comments — e.g. who got non-GAAP comments this quarter.
Every letter carries an accession to open with get_filing. Coverage: EDGAR comment letters as extracted by bios (≈200K staff letters, ≈630K answers); data_through is the latest staff letter loaded. The SEC releases letters ~20 business days after a review closes, so recent reviews appear late.
Inputs: as_of, limit, since, offset, +5 more.
sec_power_capacity_eventsPower Capacity Events (per-unit add/retire pipeline) [beta]ARead-onlyIdempotentInspect
Per-unit capacity add/retire pipeline from SEC disclosures. The row-level view that backs sec_power_supply_curve aggregation — useful when the question is about specific units ('what's NextEra's gas-build pipeline', 'every coal retirement announced in 2024') rather than aggregate trends.
Filter by filer (ticker_or_cik), event_type ('addition' / 'retirement'), fuel (coal / gas / gas_ccgt / gas_peaker / wind / solar / battery / nuclear / hydro / biomass / oil), status ('announced' / 'permitted' / 'under_construction' / 'completed' / 'cancelled'), ISO/RTO market, or target-year range.
Returns: filer_cik, unit_name_raw, event_type, fuel_type, mw, target_dt (COD), status, location_state, primary_iso, disclosed_capex_dollars, plus the source_accession + filed_at + snippet for citation. Deduped to the latest filing per (filer, unit, event_type).
Inputs: iso, fuel, as_of, limit, +5 more.
sec_power_demand_forwardPower Forward Demand (data-center/large-load wedge by year) [beta]ARead-onlyIdempotentInspect
Forward large-load demand from utility disclosures: data-center / hyperscaler / industrial / hydrogen / crypto MW commitments by year, firm vs pipeline. The answer to 'how much data-center load are utilities actually committing to' without rereading every PJM IRP.
Spec-validated headline: firm DC load 2.6 → 45 GW from 2024 to 2026 disclosures (the AI-buildout wedge). Live status enum (verified from data, the originally-spec'd values were renamed during extraction): firm side = 'contract', 'in_service', 'executed', 'firm', 'permitting'; pipeline side = 'mou', 'in_negotiation', 'announced', 'planned', 'pipeline'. Don't pass 'contracted' or 'signed_LOA' — those silently return 0 rows.
Filter by ticker_or_cik (per-utility), counterparty_type (data_center / industrial / crypto / hydrogen / ev_charging / mixed), iso, status, or year range. Returns one row per (filer, counterparty, start_year) deduped to latest filing, with the full source_accession + snippet.
Two caveats that bite per-ticker consumers hardest (power-sec 2026-07-23): (1) The firm-vs-speculative split is load-bearing, NOT cosmetic. The response separates firm_total_mw from pipeline_total_mw (and per-year in summary_by_year) — read the FIRM total, not a blended sum. Pipeline rows are interconnection-queue entries that double-count across overlapping territories and mostly never get built. Concrete trap: AEP's only demand rows are 2 pipeline entries totalling 200,000 MW — a blended read prints a 200 GW headline for one utility; its firm_total_mw is 0. (2) expected_load_mw is NULL on a meaningful share of FIRM rows (e.g. EVRG, ETR have firm events with no MW) — so sum(mw) under-represents coverage vs n_events. A null MW ≠ 'no commitment'; use n_events for count, the MW totals for magnitude, and don't conflate them.
Inputs: iso, as_of, limit, status, +5 more.
sec_power_fundamentalsPower Fundamentals (per-ticker capacity/demand/outage panel + rate cases + coverage) [beta]ARead-onlyIdempotentInspect
Per-ticker power-sector fundamentals — a utility/IPP's year-by-year operating panel in one call: panel[] (capacity added/retired MW + net-by-fuel, firm data-center/large-load demand + YoY, outage count/MW, per year), identity (role / ISO / regulated-vs-merchant / is_active), rate_cases[] (commission rulings: allowed_roe, revenue increase, outcome), and coverage (a bool per dataset). Powers inferno's SIC-routed power section.
coverage is load-bearing, not decorative — a ticker can have real capacity + rate-case data but ZERO demand/outage extraction (AEP has no firm data-center commitments in the corpus). A naive zero would read as 'no data-center commitments' (false) vs 'none extracted' (the real state). Check coverage.demand before reading a 0 as a finding.
panel = last 10 years WITH DATA (not last 10 calendar years) — for a sparsely-covered ticker this can span decades (AEP's top-10 spans 1991-2025). Blank recent years are omitted rather than zero-filled.
Rate-case honesty: we surface allowed_roe (the OUTCOME ROE), revenue_increase_dollars, decided_date, status — NOT requested_roe, rate_base_usd, or the commission filed_date (not extracted; a v2 ask, not a silent null-map). allowed_roe hasn't had a two-model QA pass — rows outside a plausible utility band [3%, 20%] carry allowed_roe_suspect=true (rare mis-extraction, e.g. a rate adjustment read as the ROE).
Methodology: filer-disclosed SEC extractions only (not the full universe), per the sec_power_* family. Input: ticker_or_cik.
Inputs: ticker_or_cik (required).
sec_power_gas_viewsPower Gas Market Views (operator sentiment by market) [beta]ARead-onlyIdempotentInspect
Filer-stated forward views on gas markets, with per-market net sentiment summary. Each row is one operator's view of a specific market (Marcellus / Permian Waha / Henry Hub / LNG export / Haynesville / Appalachia / etc.) at a point in time — bullish / bearish / neutral with key_drivers, basis_levels, and the supply-demand-balance text. Aggregate view: net sentiment per market.
Use to answer 'what are utility / midstream operators saying about Permian basis through 2026' without rereading 50 earnings transcripts. Filer-disclosed (SEC filings) — not analyst-published forecasts.
Filter by market, date range, view_direction, or filer. Returns events sorted newest-first plus a net_sentiment_by_market summary (sum of +1 bullish / −1 bearish / 0 neutral).
Inputs: to, from, as_of, limit, +3 more.
sec_power_gazetteerPower Sector Entity Registry (ticker/name → entity, or list universe by role/ISO) [beta]ARead-onlyIdempotentInspect
Power-sector entity registry / lookup. Two modes:
Resolve a single entity — pass
ticker_or_cik(ticker like 'NEE' / CIK integer / legal-name substring). Returns one row with cik / primary_ticker / legal_name / role / primary_iso / gen capacity / regulated-vs-merchant mix / parent-child / active flag.List universe by filter — omit ticker_or_cik and pass
role('utility' | 'ipp' | 'lng_terminal' | 'midstream' | 'pipeline' | …) and/oriso('PJM' | 'ERCOT' | 'CAISO' | 'MISO' | 'NYISO' | 'ISO-NE' | 'SPP' | 'non-RTO'). Returns the matching subset of the 722-row universe.
Use BEFORE other sec_power_* tools to find the right CIK / ticker / role scope. Survivorship-complete (delisted / merged / acquired entities kept with successor_cik link).
Inputs: iso, role, limit, active_only, +5 more.
sec_power_outagesPower Outages (filer-disclosed outage events panel) [beta]ARead-onlyIdempotentInspect
Disclosed unit-level outage events from SEC filings — units down for planned maintenance, forced outage, weather, derate, or extended return. The events the filer thought were material enough to call out (lead-indicator vs the live ERCOT/PJM operator feeds).
Filter by filer (ticker_or_cik), outage_type (forced / planned / weather / derate / extended), fuel, or date range. Returns: unit_name_raw, fuel_type, mw_affected, outage_type, start_dt, expected_return_dt, cause_category, cause_description, plus source_accession + filed_at + snippet for citation.
Inputs: to, from, fuel, as_of, +3 more.
sec_power_rate_casesPower Rate Cases (rulings + ROE distribution by commission) [beta]ARead-onlyIdempotentInspect
Utility rate-case rulings panel + ROE-distribution-by-commission summary. Every row is one case: which utility, which commission (FERC / Texas PUC / state IOU commissions / etc.), ruling date, revenue increase, allowed ROE, effective date, outcome (approved / settled / denied / withdrawn / partial), summary text, source filing.
Returns a roe_distribution_by_commission summary: n_cases, mean / median / min / max allowed_ROE by commission_jurisdiction. The 'what ROE did the Texas PUC allow in recent cases' question answered in one call, plus the cross-commission comp set for utility-DCF work.
Filter by commission, outcome, filer (ticker_or_cik), or date range. Survivorship-complete back to 1994.
Inputs: to, from, as_of, limit, +3 more.
sec_power_supply_curvePower Forward Supply Curve (net MW adds/retires by fuel × year, PIT) [beta]ARead-onlyIdempotentInspect
Forward supply curve from SEC disclosures: net generation MW added vs retired by fuel × target year, as-of a date. The cross-sectional aggregate a frontier LLM can't reconstruct from reading filings — survivorship-complete, point-in-time, across the whole 722-filer power universe.
Validated headline (2026-06-03 data): coal retire 8,908 MW (8.9 GW) for target year 2027 across 12 events. Pass iso='PJM'/'ERCOT'/etc. to scope to one market, or fuel='coal'/'gas'/etc. to scope to one fuel.
PIT semantics: when as_of is provided, the underlying DISTINCT ON dedup is recomputed using only events filed on or before that date — so a 2024 supply curve doesn't include 2025 plan revisions. Always returns the as_of used.
Source: per-filer disclosures in power_capacity_events, deduped by (filer_cik, unit_name_raw, event_type) latest filing. Every aggregate row carries representative source_accession + filed_at + snippet so the LLM can cite. CAVEAT: filer-disclosed only — utility plans that didn't make it into a 10-K/8-K won't be here.
Inputs: iso, fuel, as_of, year_to, +1 more.
sections_filing_change_screenFiling Change Screen (cross-sectional Lazy-Prices ranking — who rewrote most this year)ARead-onlyIdempotentInspect
Cross-sectional Lazy-Prices screen: 'which filers rewrote their Risk Factors (or MD&A / Legal / Business) most this year?' — ranks the universe of paired (year, compare_year) filings by total_change_magnitude using the same algorithm as sections__filing_change (heading set-diff → topic set-diff → semantic match with cosine 0.97 noise gate). Sister to sections__filing_change (which answers the per-filer 'what changed for ticker X' version).
Live panels (more are added over time):
10-K Item 1A 2026 vs 2025 — 4,233 paired filers (Risk Factors, the marquee; mean mag 0.172).
10-K Item 7 (MD&A) 2026 vs 2025 — 4,967 paired filers (more stable YoY than 1A; mean mag 0.089).
10-Q Item 2 (quarterly MD&A) 2026 vs 2025 — 5,100 paired filers (mean mag 0.10). Cross-filer rankings mix fiscal quarters by design — each filer's diff is fiscal-quarter-aligned to its own prior year (NVDA Q1 FY27 pairs with NVDA Q1 FY26, not with NVDA's calendar-Q1 2025 filing). Top-N answers 'who rewrote their quarterly MD&A most YoY,' not 'who reported on the same calendar quarter.'
10-K Item 3 (Legal Proceedings) 2026 vs 2025 — 4,317 paired filers (section-level grain, not sub-section). Mean mag is tiny (0.0018) because most filers carry stable boilerplate ('see Note X to Financial Statements'). The interesting cohort is the ~32 filers with mag ≥ 0.05 (the legal-disclosure tail — pharma Hatch-Waxman litigation, cannabis MSOs, CRE distress, environmental, etc.). min_change_magnitude=0.05 isolates the live legal stories. Per-filer depth via sections__filing_change(cik, item='3', form='10-K') returns the actual added/modified text. NB: heading / topic set-diff arrays are empty for section-level grain — that's by design, not a data gap.
Panel build pattern is offline; when a panel hasn't been built for the requested (form, item, year, compare_year), returns flag="panel_not_built" with empty rankings.
Filters:
min_change_magnitude — drop changes below floor.
sic_filter — exact SIC codes (array of strings).
sic_prefix — SIC prefix (e.g. "60" banks, "28" pharma).
cik_universe — rank a watchlist (array of CIK integers).
sort — magnitude_desc (default) / magnitude_asc / n_added_desc.
Magnitude distribution (10-K Item 1A 2026): 16% truly unchanged (<0.001), 28% light revision, 30% moderate, 15% substantial, 5% major, 5% saturated (50K+ chars rewritten = Lazy-Prices tail). Within the saturated cohort the secondary sort by raw_total_magnitude surfaces the actual biggest rewriters (electroCore 377K, Semnur Pharma 341K, FibroBiologics 319K — clinical-trial-restructuring biotechs, post-transformation issuers, distressed REITs).
Inputs: form, item, sort, year, +6 more.
sections_ipo_financialsIPO Financials (per-period financial statements from the IPO prospectus)ARead-onlyIdempotentInspect
Per-period financial statements extracted from a company's IPO prospectus — the revenue / operating-income / net-income / cash / balance-sheet time series from the 424B4, ordered oldest→newest. Fills the 'no 10-K history yet' gap for a company that JUST went public (or is about to price) and has no post-IPO annual filing.
financials[] — one row per reported period (annual / interim) with period / period_type / period_end + ~13 financial fields (revenue, operating_income, net_income, cash_and_equivalents, …). NULLs are preserved (a field the prospectus didn't report stays null — not zero).
status ∈ ok / no_financials_extracted (IPO in universe but no financials pulled) / not_in_universe (not a catalyst-window IPO).
Input: ticker_or_cik (CIK preferred). Pairs with sections__ipo_offering (deal terms + lockup), sections__ipo_risk_factors (risk factors), and sections__ipo_screen (find the IPO). Note: de-SPAC filers may currently return the pre-merger financials until the shared SPAC-aware resolver is applied here (sections follow-on).
Inputs: ticker_or_cik (required).
sections_ipo_offeringIPO Offering Card (deal terms + lockup per issuer, 424B4 + catalyst)ARead-onlyIdempotentInspect
One structured IPO offering + lockup card per issuer — the deal terms extracted from the 424B4 prospectus, JOINed to the catalyst lockup schedule. Covers 98.3% of the catalyst IPO universe (2,522 issuers).
offering: price_low/high, shares_offered, greenshoe_shares, exchange, ticker, deal_size_full, underwriters[], share_classes[], dual_class_flag, net_proceeds, use_of_proceeds[]. lockup: lockup_days, lockup_expiry, shares_locked, locked_pct_of_total, early_release_trigger, holder_types, tranche, is_confirmed. Plus per-source provenance (offering vs lockup source + extraction timestamp).
status ∈ ok (full card) / lockup_only (in catalyst but offering not extracted) / not_in_universe (not a catalyst-window IPO — e.g. a company that IPO'd long ago).
Input: ticker_or_cik (CIK preferred — digit string or int resolves directly; a ticker is resolved to its CIK). Pairs with sections__ipo_screen (find the IPO) and sections__ipo_risk_factors (its risk factors).
Inputs: ticker_or_cik (required).
sections_ipo_risk_factorsIPO Risk Factors (classified risk factors from an IPO prospectus)ARead-onlyIdempotentInspect
Classified risk factors from a company's IPO prospectus — prefers the 424B4 (final pricing prospectus), falls back to S-1 / S-1A if no 424B4 exists for the CIK in the corpus. Mirrors sections__risk_profile but on the PRE-IPO filing, so it works for names that just went public (or are about to price) and have no 10-K yet.
Returns a thematic + topic breakdown (counts per risk category) plus the individual risk factors (heading + primary_label + topics + verbatim text snippet). Pair with sections__ipo_screen — screen recent IPOs, then pull each one's risk-factor profile.
Perf caveat: risk-factor labels are precomputed (<100ms) for most filings. For 424B4s upserted before the 2026-07-21 labeler fix, the callable runs the labeler at query time (provenance.labels_source='runtime_fallback') — ~18s on the first such call (labeler warm-up), ~2-3s per new filer after.
Inputs: ticker_or_cik (required), year, accession, top_n_topics, +1 more.
sections_ipo_screenIPO Screen (recent IPOs from 424B4 pricing prospectuses)ARead-onlyIdempotentInspect
List recent IPOs in a date window, from 424B4 final-pricing prospectuses (the definitive 'this company went public' filing). Sections has 2020-2026 424B4 coverage (~99.9% of the catalyst_calendar IPO universe).
One row per company = its FIRST-time IPO — each CIK's EARLIEST-EVER non-SPAC 424B4 (scanned across all years, then filtered to the window). So a company that IPO'd in 2020 does NOT appear as a '2026 IPO' just because it filed a 2026 follow-on; renames of already-public companies aren't re-surfaced either. A renamed issuer carries current_company_name when it differs from the name at IPO (e.g. Classover → KIDZ AI), so it's findable both ways. For every 424B4 as a row (follow-ons + re-pricings) pass dedup_mode='per_accession'. For a CIK's full filing history use sections__get_filing.
SPACs excluded by default (SIC 6770) — pass exclude_spac=false to include them. Sorted newest-IPO-first (filing_date DESC), capped at top_n.
Filters: sector (SIC prefix, e.g. '28' pharma / '60' banks) OR sic_filter (array of exact SIC codes). Pair with sections__ipo_risk_factors to pull the classified risk factors from any of the returned IPOs' prospectuses.
Perf: ~3-4s for a ~2yr sector cut, ~17s for a full 3yr universe (per-year 424B4 metadata scroll). Narrow the window / sector to stay fast.
Inputs: since (required), top_n, until, sector, +3 more.
sector_detailSector Detail (one sector or SIC industry: KPIs, movers, flows, catalysts, companies)ARead-onlyIdempotentInspect
One sector page — or, with sic, one 4-digit SIC industry page — as in Console. Returns KPIs (cap-weighted returns, market cap, valuation medians, 13F / insider / Form D / activism flows), an industries table (sector only), movers up and down, largest notable-manager 13F buys and sells, insider buying by company, most shorted, recent activist 13Ds, upcoming catalysts, the largest private raises (Form D, sector only), cheapest vs richest P/E, and every company (largest first). Sector slugs: communication-services, consumer-discretionary, consumer-staples, energy, financials, health-care, industrials, information-technology, materials, real-estate, utilities. sector_monitor compares all sectors. Activist 13D, catalyst and private-raise rows carry accession + filing_url.
Inputs: sic, limit, top_n, sector, +1 more.
sector_monitorSector Monitor (all sectors: returns, valuation medians, 13F / insider / Form D / activism flows)ARead-onlyIdempotentInspect
All 11 sectors side by side — Console's Sector Monitor. Per sector: company count, market cap, cap-weighted total returns (1M / 3M / YTD / 1Y), notable-manager 13F net flow, insider buyers vs sellers and $ in the window, median short % of shares outstanding, activist 13D count, Form D $ raised, and valuation medians (P/E, EV/Sales, EV/EBIT, EV/EBITDA, P/B, revenue growth, margins, ROE, dividend yield). Same universe and numbers as Console (SIC-based sectors; GOOGL/META sit in information-technology). For one sector's industries, movers and companies use sector_detail.
Inputs: sort, window.
sedar_documentSEDAR+ Document Text (one filing by accession)ARead-onlyIdempotentInspect
The text of one SEDAR+ filing by its accession — the document Kaleidoscope Console opens from 'SEDAR+ Filings', the Canada Feed and Mining Projects (Console shows the PDF; this returns the text extracted from it). Accessions come from sedar_filings, canada_feed (material change reports) and mining_projects (technical reports), e.g. '06510971-00000001-000016960-20261107-PDF'. Only SEDAR+ accessions on record in MARS are served (SEC accessions: use get_filing).
Returns filer, filing date, SEDAR+ type, the PDF's availability and the text (max_chars, default 20,000; offset to read on; total_chars says how long it is). Long technical reports run to hundreds of thousands of characters: read in pieces or search with sedar_search first.
Inputs: accession (required), offset, max_chars.
sedar_filingsSEDAR+ Filings (an issuer's filings by Console family, newest first)ARead-onlyIdempotentInspect
An issuer's SEDAR+ filings, newest first — Kaleidoscope Console's 'SEDAR+ Filings' report on the company page. Each row: accession (pass it to sedar_document for the text), filed date, SEDAR+ type label (e.g. 'NI 43-101 technical report', 'Interim MD&A'), Console family, a readable title, filer name, French-copy flag. Also returns counts per family, per year, French copies, first / last filed.
family: key (default — Console's opening view: everything except shelf supplements and certificates & fees), all, or one of: fin Financial statements & MD&A; events Material changes & news; gov Proxy & meetings; bids Bids & early warning; tech Technical reports; offer Prospectuses & offerings; shelf Shelf supplements; cert Certificates & fees; other Other; untyped Not typed yet. q matches the document name (accession) or the SEDAR+ category ('annual information' finds an AIF). French copies of bilingual filings are hidden unless include_french. Pages of up to 100 (Console's page size); offset to page on.
Coverage: SEDAR+ filings 2010 on (filing_type is null on older rows: family 'untyped'). Big issuers file thousands of shelf supplements (TD ~34k), hence the 'key' default.
Inputs: q, cik, year, as_of, +7 more.
sedar_issuerSEDAR+ Issuer Profile (profile no., tickers, CIK, LEI, province, sector, cap)ARead-onlyIdempotentInspect
A Canadian issuer's SEDAR+ profile — the company header line of Kaleidoscope Console's SEDAR+ & Canada module. Returns the SEDAR+ issuer number(s) (profile), names, Canadian tickers, CIK, LEI, MARS company id, province, sector, market cap (US listing's cap, else the TSX / TSXV cap in C$ and US$), the TSX / TSXV listing's own cap (tsx_listing, as Console's Excel TSX_* fields), SEDAR+ filing counts (Console's 'key' view: no shelf supplements, certificates or French copies) and last filed date, material change reports in the last 12 months, and the number of NI 43-101 mining projects. Works for issuers with no CIK (pass SEDAR:, TSX:/TSXV: or the MARS id).
One CIK can have several SEDAR+ profiles (a parent and a trust, an old and a new profile): they are ordered as Console does (most MARS rows, then the newest profile number); the first is the main one.
Next: sedar_filings (the filing list), canada_feed (material changes), mining_projects (NI 43-101 projects), sedar_document (a filing's text).
Inputs: cik, issuer, profile_number, mars_company_id.
sedar_searchCanadian SEDAR+ Filing Search (semantic or keyword, 2026 on, with doc text)ARead-onlyIdempotentInspect
Search Canadian SEDAR+ filings (TSX / TSXV / CSE issuers and funds): news releases, prospectuses, financial statements, MD&A, material change reports, proxy circulars, technical reports and more, one result per filed document. Coverage: filings from 2026-01-01 on, updated as sections loads them.
mode='semantic' (default) ranks by meaning (learned-sparse, like semantic_search); mode='keyword' ranks by exact words. Either way the snippet shows matching text. Filters: ticker, company (filer name), issuer_no, filing_type (e.g. news_releases, material_change_report, annual_mda, interim_mda, annual_financial_statements, management_proxy_materials, shelf_prospectus_ni_44102), date_from / date_to. With filters and no query, the newest filings come first.
Each result cites its source: doc_id (one PDF), accession (the SEDAR+ document id) and package_id (the filing package). Pass doc_id to get that document's text (up to 20,000 characters, max_chars). Scanned PDFs have no text (text_quality='scanned_no_text').
Inputs: mode, limit, query, doc_id, +7 more.
semantic_searchSemantic Filing SearchARead-onlyIdempotentInspect
Find SEC filing SUB-SECTIONS semantically similar to a natural-language query (learned-sparse retrieval over the sub-section graph — Item 1A Risk Factors, Item 7 MD&A, Business, etc.). Unlike get_sentences (which filters by a known classifier name), this understands meaning: e.g. 'supply chain disruptions', 'interest rate sensitivity', 'workforce restructuring and severance'. Returns whole coherent sub-sections (more context than a single sentence). Optional filters: ticker, form ('10-K'/'10-Q'/'S-1'/'20-F'), sic, years, filed_before.
Inputs: query (required), sic, form, limit, +3 more.
short_interestShort Interest (FINRA biweekly panel — shares short, days to cover, squeeze/spike flags)ARead-onlyIdempotentInspect
FINRA short interest history for a ticker — biweekly settlement-date panel since 2021-06-15. Returns shares short, days to cover, and mars's own pre-computed signal flags (squeeze_setup, short_spike, short_covering, short_extreme, short_zscore_high) per settlement, newest first.
Ticker is the join key (CIK resolves on only ~37% of rows in the source panel) — the tool tries the ticker as given, then with punctuation stripped (BRK.B / BRK-B -> BRKB), since FINRA's own ticker spelling in this feed doesn't always match ours.
NOT YET computed: % of shares outstanding / % of float (the source panel has raw shares short + avg daily volume but not a shares-outstanding join — days_to_cover and the flag fields are the best available signal today; a %OS/%float enrichment is a follow-up, not silently guessed here).
Known limitation: ticker symbol reuse (e.g. a delisted company's old ticker reassigned to a new one, as happened with META) is not detected — a long history pull may silently splice two different companies' data if the ticker was reused. Cross-check against the company's known listing history if the early history looks inconsistent with the company you mean.
Inputs: ticker (required), limit, since.
short_volumeDaily Short Volume (FINRA Reg SHO off-exchange short-sale volume)ARead-onlyIdempotentInspect
Daily short-sale volume for a stock from FINRA's Reg SHO files (Console's daily short volume). Returns, per trading day, short volume, short-exempt volume, FINRA total volume and the short ratio (short / total), plus volume-weighted short ratios over the last 5 days, 20 days and 1 year (252 trading days), and the z-score of the last 5 days' average daily ratio against the past year.
What this is not: it covers off-exchange, FINRA-reported volume only, not total consolidated volume, and it is not short interest. A high short ratio is normal (market makers sell short to fill buy orders), so read it against the stock's own history (the z-score), not as a level. For shares held short, use short_interest.
Symbols follow FINRA: class shares use a slash (BRK/B); BRK.B and BRK-B are converted. About 400 days of history are kept. Some volumes in the source carry fractional shares (under review by FINRA's data team), so treat them as approximate.
Inputs: ticker (required), days.
thirteenf_cohort_flow13F Cohort Flow (net buys/sells across chosen managers)ARead-onlyIdempotentInspect
Net 13F flow per ticker across a SPECIFIC cohort of managers for one quarter — 'what is this set of funds collectively buying or selling.' Pass the manager CIKs (resolve names with thirteenf_search_managers). Ranks by ACTUAL share-based flow (Δshares × price), so price-only moves don't show as phantom buys/sells. Returns the top securities by net flow across the cohort, with how many of the cohort's funds moved each. Use for 'what are the activists / these 8 funds accumulating.' For the market-wide 'is smart money broadly buying X' consensus, that's a different question (all-notable precompute) — this tool is cohort-scoped on purpose. Distinct from thirteenf_screen (which is cluster/velocity signals).
Cohort: EITHER cohort_ciks (explicit list, max 50) OR a NAMED cohort built from the 13F manager classification — cohort_style (e.g. activist, value, long_short, quant, macro, distressed, event_driven), cohort_tier (top_50 / top_200 / long_tail), cohort_manager_type (hedge_fund, family_office, pension, sovereign_wealth, ...), and notable_only. Named filters AND together; the cohort is capped at the 1,000 largest managers by tier (response reports cohort_size and whether it was truncated).
Per security: net_flow_usd plus how many cohort funds opened (n_new), added (n_increased), trimmed (n_decreased), fully EXITED (n_exits), or held unchanged. Exits are included in net flow by default (an exit is real selling); include_exits=false drops full exits to show only funds still holding. Coverage artifacts (#2168, 2025Q4 on): when a security joins the SEC 13(f) list (or its old CUSIP leaves as the new one joins), holders that already owned it start or stop reporting it; those new / exit rows are not trades and are left out of new-position and exit counts, flows and cluster flags. Affected rows carry coverage_artifact.
Inputs: limit, direction, cohort_ciks, cohort_tier, +5 more.
thirteenf_holders13F Top Holders of a TickerARead-onlyIdempotentInspect
Get the institutional holders of a ticker from 13F filings (postgres-backed). Returns the honest FULL holder base (total_holders count, not a top-20 cap), each holder with rank, filer_name, shares, value_usd, qoq_action (new/add/trim/exit/unchanged/reporting_started), and manager classification (manager_type / tier / is_notable). Plus top-level v3 signals: consecutive_accum_qtrs (sustained accumulation), is_cluster_entry (≥3 notable funds entered this quarter), is_quiet_accumulation (stealth pattern), accel_4q (this Q net change vs trailing 4Q), corporate_action_suspect. Use for 'who owns ticker X' questions.
Pass is_notable_only=true to filter the holders list to notable (smart-money) managers only — cuts response size on mega-caps. Recommended for any workflow citing holders by name.
limit caps the returned holder rows (default 20); total_holders reports the true full count. For a manager+security multi-quarter trail use thirteenf_position_history. Data mirrors EDGAR 13F on a ~4h cycle; report_date defaults to the latest mature quarter. Coverage artifacts (#2168, 2025Q4 on): when a security joins the SEC 13(f) list (or its old CUSIP leaves as the new one joins), a holder that already owned it starts or stops reporting it; that reads reporting_started / reporting_stopped (cusip_changed when the old CUSIP left and the new one joined) with no flow, not new / exit.
Inputs: cusip, limit, ticker, report_date, +1 more.
thirteenf_holders_full13F Full Holder List (exhaustive, paginated)ARead-onlyIdempotentInspect
Exhaustive, paginated list of EVERY institutional 13F holder of a ticker, from the full holdings database — with an honest total count (e.g. 'showing 1-50 of 5,610 holders'). Use this when the user wants the complete holder base, a deep tail, or a specific holder rank — NOT the top-20 snapshot (use thirteenf_holders for that, it's faster and has flow/velocity signals). Each holder: filer name, manager type/tier/notable flag, is_biotech_specialist (smart_money's tag for about 100 filers whose book is mostly biotech, rule biotech_v1), position value, shares, and qoq_action ∈ {new, exit, add, trim, unchanged} mapped from the quarter's holdings_events row, and accessions (the holder's 13F filing(s) for the quarter, newest first, including sub-CIK and amendment filings). Response also carries the v3 Layer-3 signals at the top level (consecutive_accum_qtrs, consecutive_dist_qtrs, accel_4q, avg_position_value, is_cluster_entry, is_cluster_exit, is_quiet_accumulation, corporate_action_suspect) — same names as smart_money's Qdrant snapshot payload so PIT-replay reads uniformly. Sorted by position value, largest first. Defaults to common-stock long positions as of the latest reported quarter. Coverage artifacts (#2168, 2025Q4 on): when a security joins the SEC 13(f) list (or its old CUSIP leaves as the new one joins), a holder that already owned it starts or stops reporting it; that reads reporting_started / reporting_stopped (cusip_changed when the old CUSIP left and the new one joined) with no flow, not new / exit.
Inputs: cusip, limit, offset, ticker, +2 more.
thirteenf_manager_performance13F Manager Track Record (replication return vs S&P)ARead-onlyIdempotentInspect
Track record of a 13F manager over time — the value-weighted return of their disclosed long book vs the S&P (SPY), per quarter, with trailing 1-year and 2-year figures, excess-over-SPY, and hit rate. Use to rank/weight managers by skill or to contextualize a fund's moves ('have they actually been good?'). Resolve names with thirteenf_search_managers. IMPORTANT: this is a disclosed-long-book REPLICATION PROXY (a '13F clone'), NOT audited fund returns — excludes shorts/derivatives/fees/cash; always present it as such and check priced_coverage_pct. Returns are decimals (0.12 = 12%).
Inputs: cik (required), as_of, quarters.
thirteenf_overlap13F Portfolio OverlapARead-onlyIdempotentInspect
Compute portfolio overlap between two 13F managers from their FULL holdings (postgres-backed — no top-50 cap, so small-tail overlaps are included). Returns the shared CUSIPs with each manager's position (shares, value_dollars), plus overlap statistics: shared count, shared value, and overlap percent on each side (shared_value / manager's total_value). Useful for 'do BRK and Bridgewater agree on anything?' or 'how much do two activists overlap'. Common-stock-long convention (share_type='SH', put_call IS NULL). Data mirrors EDGAR 13F on a ~4h cycle; report_date defaults to the latest mature quarter. manager_a / manager_b each carry accession + filing_url (that manager's 13F for the quarter; open with get_filing) and accessions (every 13F/amendment behind it).
Inputs: cik_a (required), cik_b (required), report_date.
thirteenf_portfolio13F Manager PortfolioARead-onlyIdempotentInspect
Get the full portfolio for a 13F manager (hedge fund, asset manager, pension, etc.), postgres-backed. Returns the manager's positions (rank, ticker/cusip, shares, value_usd, qoq_action) — the exhaustive book, not a top-50 cap — plus portfolio aggregates and the v3 accumulation signals. Lookup by CIK (preferred) or filer_name (case-insensitive exact match). For Berkshire Hathaway use cik=1067983.
limit caps the returned position rows (default 50); the response reports the true full count. For a manager+security multi-quarter trail use thirteenf_position_history. Data mirrors EDGAR 13F on a ~4h cycle; report_date defaults to the latest mature quarter. The header carries accession + filing_url (the manager's 13F for the quarter; open with get_filing) and accessions (every 13F/amendment behind the quarter).
Inputs: cik, limit, filer_name, report_date.
thirteenf_portfolio_full13F Full Manager Portfolio (exhaustive, paginated)ARead-onlyIdempotentInspect
Exhaustive, paginated list of EVERY position a 13F manager holds, by CIK, from the full holdings database — with an honest total count (e.g. 'showing 1-50 of 1,240 positions'). Use this for the complete book, the long tail, or a specific position's rank — NOT the top-50 snapshot (use thirteenf_portfolio for that; it's faster and has flow/velocity signals). Resolve a manager name to a CIK first with thirteenf_search_managers. Each position: ticker, security name, value and shares. Sorted by position value, largest first. Defaults to common-stock long positions as of the latest reported quarter; pass report_date for a past quarter. The header carries accession + filing_url (the manager's 13F for the quarter; open with get_filing) and accessions (every 13F/amendment behind the quarter).
Inputs: cik (required), limit, offset, report_date, +1 more.
thirteenf_position_history13F Position History (quarter-by-quarter trail)ARead-onlyIdempotentInspect
Quarter-by-quarter history of ONE 13F manager's position in ONE security — the full entry → add/trim → exit trail, back to 2020. Use this to explain HOW a manager built or unwound a stake over time (not just the latest quarter). Returns each quarter's shares, marked value, the share count change, the actual dollar amount traded (flow_usd = Δshares × price, excludes price-only moves), and the event type (new/increase/decrease/exit/unchanged), plus a summary (first seen, latest, peak, net change). Resolve a manager name to a CIK first with thirteenf_search_managers. No other tool has multi-year position trails. Coverage artifacts (#2168, 2025Q4 on): when a security joins the SEC 13(f) list (or its old CUSIP leaves as the new one joins), a holder that already owned it starts or stops reporting it; that reads reporting_started / reporting_stopped (cusip_changed when the old CUSIP left and the new one joined) with no flow, not new / exit. Each quarter row carries accession + filing_url (the manager's 13F for that quarter; open with get_filing) and accessions (every 13F/amendment that quarter).
Inputs: cik (required), ticker (required).
thirteenf_screen13F Institutional Flow ScreenerARead-onlyIdempotentInspect
Cross-company 13F screener — find tickers matching institutional flow / velocity criteria. This is the cross-cutting view that single-company tools can't give you.
Key filters (the v3 differentiation):
is_cluster_entry=true → ≥3 notable funds entered this quarter (consensus buy)
is_cluster_exit=true → ≥3 notable funds exited (consensus sell)
is_quiet_accumulation=true → rising holder count + value with small avg positions (stealth)
consecutive_accum_qtrs_min=N → N+ consecutive quarters of net institutional inflow
consecutive_dist_qtrs_min=N → N+ consecutive quarters of net outflow
n_notable_new_positions_min=N → at least N notable funds opened a new position this Q
n_notable_exits_min=N → at least N notable funds fully exited this Q (the sell-side mirror)
exclude_funds=true → drop ETPs / open-end / closed-end funds / funds-of-funds
notable_holder_ciks=[...] → tickers held by any of these notable manager CIKs
Compound example: is_cluster_entry=true + consecutive_accum_qtrs_min=2 → sustained accumulation that just got a notable-fund cluster on top.
Default sort: net_change_value (this quarter's net institutional inflow in dollars, share-based). Each row carries breadth (n_increased / n_decreased / n_new_positions / n_exits across all 13F holders) and the notable-cohort counts (n_notable_new_positions / n_notable_exits / notable_net_value_usd).
Data is refreshed every ~4 hours from EDGAR 13F filings. Coverage artifacts (#2168, 2025Q4 on): when a security joins the SEC 13(f) list (or its old CUSIP leaves as the new one joins), holders that already owned it start or stop reporting it; those new / exit rows are not trades and are left out of new-position and exit counts, flows and cluster flags. Affected rows carry coverage_artifact.
Inputs: limit, sort_by, ticker_in, report_date, +13 more.
thirteenf_search_managersFind 13F Manager by NameARead-onlyIdempotentInspect
Find a 13F manager's CIK by name — substring / case-insensitive full-text search over filer names. Use this FIRST when you know a manager by name but not their CIK, then pass the resolved CIK to thirteenf_portfolio, thirteenf_overlap, thirteenf_screen, or search_activist_positions. Fixes the 'exact registered legal name required' problem.
Returns candidate managers (cik, filer_name, tier, is_notable, manager_type, manager_styles, current_total_value), notable + largest first. A query can match several entities (e.g. 'berkshire' → Berkshire Hathaway Inc AND Berkshire Asset Management) — pick the right CIK from the candidates.
Examples: 'berkshire' → Berkshire Hathaway Inc (cik 1067983); 'baupost' → BAUPOST GROUP LLC/MA; 'tiger global' → TIGER GLOBAL MANAGEMENT LLC. Note: only managers that file 13F are covered; a name with no 13F filings returns no hits even if it's a known fund. Each candidate carries accession + filing_url (its 13F for latest_report_date; open with get_filing) and accessions (every 13F/amendment that quarter).
Inputs: name_query (required), limit, is_notable_only.
thirteenf_smart_money_consensus13F Smart-Money Consensus (all-notable buys/sells)ARead-onlyIdempotentInspect
All-notable smart-money consensus for a quarter, from a precomputed table of ~760 notable funds. Two modes: (1) pass a ticker for that security's consensus — how many notable funds hold it, how many newly bought / exited / added / trimmed, and the net dollar flow ('is smart money broadly buying or selling X'); (2) omit ticker for the leaderboard of tickers notable funds are most buying (or selling). This is the WHOLE notable universe — for a SPECIFIC set of managers use thirteenf_cohort_flow; for cluster/velocity signals use thirteenf_screen.
Per-ticker responses also carry an all_holders block covering EVERY 13F filer, not just the notable funds: net_value_usd (share-based flow) plus breadth (n_holders, n_increased, n_decreased, n_new_positions, n_exits). Example: GOOGL 2026-Q2 all-holder net is about +$42.6B. Leaderboard universe: notable (default) ranks by notable-fund net flow, and all_holders ranks by all-filer net flow. Rows flagged corporate_action_suspect are excluded in both universes. Every leaderboard row carries both net figures and both breadth sets. Funds are excluded by default (smart_money's is_fund: ETPs, open-end, closed-end and funds-of-funds; untyped securities are kept), and n_funds_filtered says how many fund rows were dropped; include_funds=true ranks them too. New securities (is_new_security: IPOs, CUSIP successions; every holder is 'new', so the whole stake reads as buying) are excluded the same way: n_new_securities_filtered, include_new_securities=true. All net flow figures (notable_net_flow_usd, net_value_usd) are share-based (change in shares x price); per-ticker notable_value_change_usd is the separate price-inclusive change.
signal weighs flow AND breadth (adders = increased + new, reducers = decreased + exited): net_buying / net_selling when both agree, mixed_dollar_buying / mixed_dollar_selling when a few large trades go against most holders (e.g. GOOGL 2026-Q2: +$62.5B notable flow but 182 adders vs 226 reducers -> mixed_dollar_buying).
Look-ahead on past quarters: the notable funds are TODAY'S list, applied to every quarter, so a past quarter's notable figures use funds picked with hindsight, not the funds that were notable then. Don't treat them as point-in-time in a backtest. universe=all_holders and the all_holders figures don't use the list. Coverage artifacts (#2168, 2025Q4 on): when a security joins the SEC 13(f) list (or its old CUSIP leaves as the new one joins), holders that already owned it start or stop reporting it; those new / exit rows are not trades and are left out of new-position and exit counts, flows and cluster flags. Affected rows carry coverage_artifact.
Inputs: limit, ticker, universe, direction, +5 more.
thirteenf_who_holds13F Who Holds This TickerARead-onlyIdempotentInspect
Reverse 13F lookup — find institutional managers holding a given ticker (postgres-backed, the full 13F holder base — not a top-position cap), optionally filtered by manager classification. Returns each matching manager's position (shares, value_dollars) plus their classification (manager_type / tier / is_notable), largest first.
Filter combinations unlock targeted queries:
is_notable=true → only the ~760 smart-money managers ($58T AUM)
manager_styles='activist' → activist funds (Icahn, Pershing, ValueAct, Starboard, etc.)
manager_styles='quant' → quant funds
manager_type='hedge_fund' → hedge funds only
tier='top_50' → only the top 50 managers by AUM All filters are AND-combined (e.g. notable + activist = 8 funds).
Data is refreshed every ~4 hours from EDGAR 13F filings. Each manager row carries accession + filing_url (its 13F for the quarter; open with get_filing) and accessions (every 13F/amendment behind the position).
Inputs: ticker (required), tier, limit, is_notable, +3 more.
tranche_bond_cardCorp Bond Card (consensus price + breadth + identity, by CUSIP)ARead-onlyIdempotentInspect
Full consensus card for one corporate bond by CUSIP, as-of a date. Returns: cross-fund consensus price (mean / stddev / min / max), breadth (n_funds, n_marks), total fund-held par and value, the bond identity (issuer ticker/cik, coupon, maturity_date), and a yield analysis block (yield-to-maturity, 1-month and 3-month price deltas, a distressed flag). Source: N-PORT-P fund holdings panel.
IMPORTANT: 'fund-held' means SEC-registered funds (mutual funds + ETFs) reporting via N-PORT-P. Roughly $3T of the ~$11T US corporate bond complex. NOT total debt outstanding. Use this as a large point-in-time, identifier-exact sample. Historical depth: 2019-09-30 to current.
Price interpretation: tight stddev (<0.5) = strong cross-fund agreement; wide stddev = real signal of pricing uncertainty on an illiquid bond.
Yield analysis: YTM is semi-annual-compounding, computed from the consensus price assuming a standard semi-annual coupon schedule to the stated maturity_date — a clean-price approximation (no accrued-interest adjustment, since exact settlement-date accrual isn't tracked in this panel). This assumes a bullet-maturity bond; issuer_cat='CORP' in the source panel is known to also include some agency/structured instruments (e.g. GNMA, Freddie Mac REMICS) whose amortizing-principal cash-flow shape doesn't match this formula — YTM on those will be misleading, not just imprecise. Prices outside roughly 1-300 are excluded from YTM/distressed entirely (per-unit structured-note marks, not real bond prices — verified live: some CUSIPs carry negative price_mean). distressed is true when YTM >= 12% or price < 50 — a common industry threshold, not a prediction. null fields in yield_analysis mean the input needed to compute them was missing (no coupon, already matured, or no prior-month mark within tolerance for a delta), not zero.
Inputs: cusip (required), as_of.
tranche_bond_holdersCorp Bond Holders (which funds + crowding stats, by CUSIP)ARead-onlyIdempotentInspect
Funds holding a corporate bond by CUSIP, as-of a date, with crowding stats. Returns: ranked list of holding funds (series_id, series_name, registrant_cik, par, value_usd, implied_price) AND a concentration block (n_holders, total fund-held value, top-5 share of total). Source: N-PORT-P fund-holdings panel.
IMPORTANT: 'fund-held' means SEC-registered funds (mutual funds + ETFs) only. Insurers / pensions / foreign holders are NOT in this panel — so this is the fund slice of the ownership picture, not the full holder list. Use as a large point-in-time, identifier-exact sample. Historical depth: 2019-09-30 to current.
Identifiers: series_id (S000…) is the fund; registrant_cik is the fund family (one registrant has many series — don't confuse these). Each holder row carries accession + filing_url (the fund's N-PORT for that report date; open with get_filing).
Inputs: cusip (required), as_of, top_n.
tranche_bond_price_dispersion_screenCorp Bond Price-Dispersion Screen (cross-fund mark disagreement, distress proxy)ARead-onlyIdempotentInspect
Corporate bonds where N-PORT-P funds' marks DISAGREE most at an as-of date — a public pricing-uncertainty proxy for illiquid / distressed bonds that have no recent TRACE prints. Ranked by absolute price_stddev (mark dispersion in price points). Returns CUSIP, issuer ticker/name/CIK, n_funds, n_marks, price_mean, price_stddev, dispersion_pct (= coefficient of variation, stddev/mean ×100), price_min/price_max, coupon, maturity_date.
Two reads of dispersion: this ranks by ABSOLUTE stddev (price points). For a DISTRESS lens re-sort the returned rows by dispersion_pct (CV) client-side — low-priced bonds inflate CV, so absolute-stddev and CV rank different bonds to the top. Extreme rows are usually defaulted / distressed bonds where funds' recovery marks legitimately diverge (some near par, some near 0) — read price_min/price_max + n_funds to see that spread, don't treat wide dispersion alone as an error.
Source: N-PORT-P registered-fund holdings (mutual funds + ETFs) — a SAMPLE of the corporate-bond complex (~$3T), NOT total marks outstanding; insurers / pensions / foreign holders excluded. Corporate-only (issuer_cat='CORP'). Historical depth 2019-09-30 → current.
Views (view, Console Credit Monitor): dispersion (default, ranked by stddev as above); disputed = price std dev >= 5 points across 3+ funds (securitisations / agencies and prices outside 1-2000% of par left out); distressed = est. yield >= 12% or price < 50, securitisations / agencies / mortgage & lease trusts excluded, ranked by fund-held value. disputed and distressed read MARS's company-linked bond panel (Console's source; bonds with no linked company are left out), match securitisations on the issuer name and the MARS company name, and default to the newest SETTLED month. Every row carries est_yield_pct, is_distressed, is_disputed, is_structured_issuer and listed_parent {ticker, cik, name, via}. Coupons above 20% are blanked and never used for a yield (structured notes).
Inputs: view, as_of, top_n, min_funds.
tranche_debt_flow_moversDebt Flow Movers (issuers funds are fleeing / piling into)ARead-onlyIdempotentInspect
Market-wide: the issuers whose corporate bonds funds are FLEEING (direction='outflow', the credit-distress feed) or PILING INTO ('inflow') in one month, ranked by net fund flow in USD. Console Credit Monitor rules: reads mars.tranche_issuer_flow_v2; defaults to the newest month whose issuer coverage is at least 85% of the prior month's (a month still filling in is skipped — month_choice says which month was used and why); only issuers with $50M+ of fund-held bonds (min_issuer_value_usd). Each row: issuer cik / ticker / name, listed_parent {ticker, cik, name, via}, net_flow_usd, net_flow_pct, n_continuing_funds, total fund-held value, small_basis (< 20 continuing funds). Rows marked 'warn' by tranche (tiny bases, median 2 funds) are left out unless include_warns=true.
How to read it: net_flow_usd is quarter-over-quarter (N-PORT holdings are quarterly per fund) and coverage-immune — it sums, over funds that held the issuer's bonds at both this report and that fund's OWN prior report, the change in value; which other funds filed doesn't matter, so it measures real trading, not the staggered-filing sawtooth. Negative = existing holders net-trimming (outflow), positive = net-adding. n_continuing_funds is the basis: it swings with the fiscal calendar (~600 funds at a calendar quarter-end, ~15 on odd fiscal dates), so prefer net_flow_pct for comparisons and distrust pct on small bases. Fund-held only (N-PORT-P, a ~$3T sample of registered funds), corporate bonds only.
Inputs: as_of, top_n, direction, min_funds, +2 more.
tranche_dispersion_trendCorp Bond Dispersion Trend (mark-stddev trajectory, credit-event lead)ARead-onlyIdempotentInspect
The cross-fund mark-dispersion (price_stddev) trajectory of a single corporate bond over time, newest-first. Rising dispersion is a mark-uncertainty regime change — funds' recovery/valuation marks start to disagree, which can lead a credit event (default / restructuring) before it resolves. Returns per report_date: n_funds, n_marks, price_mean, price_stddev, dispersion_pct (CV = stddev/mean ×100).
Watch n_funds alongside stddev — dispersion computed off very few funds (2-3) is noisy; a rise that coincides with more funds marking the bond is a stronger signal than one on a thinning holder base. Source: N-PORT-P registered-fund holdings (a ~$3T sample, not total marks outstanding). Pass the CUSIP from tranche__bond_price_dispersion_screen / tranche__widest_held / tranche__bond_card.
Inputs: cusip (required), n_months.
tranche_fund_portfolioFund Debt Book (every bond a fund holds, by series_id)ARead-onlyIdempotentInspect
Full debt-holdings book for one fund (by SEC series_id, S000…) at an as-of report_date. Returns every position: CUSIP, issuer ticker/name, par, value_usd, coupon, maturity_date, implied_price. Ordered by value descending. Source: N-PORT-P.
Every call also returns asset_classes: the fund's net assets and its mix across equity, debt, derivative, repo, cash and other (value_usd + share of net assets), plus positions_to_net_assets (about 1.0 when the book reconciles; above 1.0 is gross leverage from financed repo / TBA). include_noncusip=true adds the same report's positions without a CUSIP (loans, foreign bonds, repos with their terms, cash, other). For derivatives use fund_derivatives.
Identifier: series_id is the FUND (e.g. S000016548 = PIMCO Income Fund). registrant_cik is the fund family (one registrant has many series). Pass the series_id, not the registrant. If you only know the registrant or fund name, no resolver is built yet — query current_holdings by series_id once you have it.
Historical depth: 2019-09-30 to current. Uses the fund's latest report on or before as_of (funds file on staggered calendars, many quarterly).
Inputs: series_id (required), as_of, top_n, include_noncusip.
tranche_issuer_debtCorporate Issuer Debt Profile (card + maturity wall)ARead-onlyIdempotentInspect
Corporate issuer's fund-held debt profile by ticker or CIK. Returns the issuer summary card (n_bonds, n_funds reaching, weighted coupon, total fund-held value, next maturity, 1y/2y/3y maturing buckets) and — by default — the year-by-year maturity wall ladder. Source: N-PORT-P fund holdings panel.
IMPORTANT: CORP-only (issuer_cat='CORP') — munis, treasuries, agencies are filtered out at build time. Fund-held only (N-PORT-P universe — mutual funds + ETFs); insurers, pensions, and foreign holders are NOT in this panel, so these numbers are a sample of total debt outstanding, not the totals themselves.
Issuer resolution: prefer ticker for resolved equity issuers; for debt-only financing subs (Toyota Motor Credit, GE Capital, etc.) pass issuer_cik directly. Coverage ~65% of corporate fund-held par by ticker; the rest by cik.
Historical depth: 2019-09-30 to current.
Inputs: ticker_or_cik (required), as_of, include_maturity_wall.
tranche_issuer_debt_flowIssuer Debt Fund Flow (QoQ net adds / trims by holding funds)ARead-onlyIdempotentInspect
One issuer's quarter-by-quarter FUND FLOW in its corporate bonds, newest first: are the mutual funds and ETFs that hold its debt net-adding or net-trimming? Per report date: net_flow_usd, net_flow_pct, n_continuing_funds (the basis), the continuing funds' prior value, total fund-held value, n_funds and n_bonds.
How to read it: net_flow_usd is quarter-over-quarter (N-PORT holdings are quarterly per fund) and coverage-immune — it sums, over funds that held the issuer's bonds at both this report and that fund's OWN prior report, the change in value; which other funds filed doesn't matter, so it measures real trading, not the staggered-filing sawtooth. Negative = existing holders net-trimming (outflow), positive = net-adding. n_continuing_funds is the basis: it swings with the fiscal calendar (~600 funds at a calendar quarter-end, ~15 on odd fiscal dates), so prefer net_flow_pct for comparisons and distrust pct on small bases. Fund-held only (N-PORT-P, a ~$3T sample of registered funds), corporate bonds only. Pair with tranche_issuer_debt (maturity wall) and tranche_dispersion_trend (mark disagreement) for a credit read.
Inputs: ticker_or_cik (required), n_quarters.
tranche_refi_screenCorp Refi Screen (issuers ranked by debt maturing in year)ARead-onlyIdempotentInspect
Cross-sectional refi-pressure screener: corporate issuers ranked by fund-held debt value maturing in a target year, as-of a date. Returns issuer ticker/cik + fund_held_value + weighted_coupon + bond_count, descending. The 'who has the biggest 2027 wall' question. Source: N-PORT-P fund holdings panel.
IMPORTANT: This is the FUND slice of total maturing debt — N-PORT-P covers ~$3T of ~$11T US corp bonds. Honest as a RANK (within-panel ordering is load-bearing); do NOT present absolute $ as 'total debt maturing.' Corporate-only (issuer_cat='CORP'). Historical depth: 2019-09-30 to current — pin as_of to any quarter-end for backtests.
Console Credit Monitor rules: defaults to the newest SETTLED month (the newest month is still filling in); weighted_coupon above 20% is blanked (structured notes). Each issuer carries listed_parent {ticker, cik, name, via} — the listed company behind a financing sub, resolved via bond CIK, MARS company CIK, Ex-21 parent, then normalised name.
Inputs: maturity_year (required), as_of, top_n.
tranche_widest_heldWidest-Held Corp Bonds (most-owned by N-PORT-P funds)ARead-onlyIdempotentInspect
Most widely fund-held corporate bonds at an as-of date, ranked by number of distinct funds holding the bond. Returns CUSIP, issuer ticker, total fund-held value, n_funds, and consensus mean price. Source: N-PORT-P fund holdings panel.
IMPORTANT: 'Widely held' = held by the most N-PORT-P registered funds (mutual funds + ETFs). Insurers, pensions, foreign holders are NOT counted. Use as a sample of broad-market ownership, not absolute breadth. Corporate-only (issuer_cat='CORP'). Historical depth: 2019-09-30 to current.
Each bond also carries coupon (blanked above 20% — structured notes), maturity_date, est_yield_pct, is_distressed (est. yield >= 12% or price < 50, structured issuers excluded), is_disputed (price std dev >= 5 pts across 3+ funds) and listed_parent {ticker, cik, name, via} (Console Credit Monitor rules).
Inputs: as_of, top_n.
upcoming_earningsUpcoming Earnings (PIT-honest projected report dates with calibrated ranges)ARead-onlyIdempotentInspect
Which companies are projected to report earnings in a forward window. CRITICAL: dates are PROJECTIONS, NOT facts. EDGAR has no forward calendar; companies pre-announce dates on IR pages as unstructured text. Dates are projected from each filer's historical cadence (period-end clock + announcement-lag fit + seasonal refinement), each with a confidence-scored calibrated range. Always present BOTH projected_report_date AND report_date_range — never the point date alone as if it were known.
Validated: PIT backtest on 97k held-out predictions 2018-2026 returned medAE 1 day, 92.6% interval coverage out-of-sample (tightens with confidence: ±4d metronomic / ±6d tight / ±8d loose / ±12d irregular / ±13d unproven). The naive '≥90% within ±3d' bar is NOT met by cadence-only methods — companies re-pick their date each quarter (~several days irreducible). v2 will collapse ranges to ±0-1d when an 8-K Item 7.01 or a press release confirms a specific date.
Pairs with earnings_surprise_history — 'who reports → what to expect.' Optional triage enrichment per row (size_tier / dollar_volume / market_cap / last_rev_yoy / last_eps_yoy) so the LLM can prioritize without N follow-up calls.
av_calendar (present per row only when an AlphaVantage vendor calendar corroborates the projection within 45d; absent on historical as_of reads): av_report_date (vendor's date — a SECOND opinion, not ground truth; unvalidated + can be stale for low-coverage names), av_days_diff (vendor date − our projection; ≈0 ~75% of the time — a large value flags genuine disagreement worth a second look, NOT a tool error), av_estimate_eps (vendor consensus EPS — independent of our own last_eps_yoy), av_time_of_day (pre/post-market session timing — new info, not elsewhere), av_collected_at.
Filters: window (days_ahead), min_confidence_tier, min_dollar_volume, min_market_cap, size_tier_in, tickers, include_overdue, sort_by, limit.
Inputs: as_of, limit, sort_by, tickers, +7 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceRemote MCP server that lets AI assistants read tracked competitive intelligence — competitor news, reviews, website changes, social posts and ads — and act on it by managing projects, competitors, battle cards, reports and canvases. Every answer carries a source and a date, and write actions only publish when explicitly shared.MIT
- AlicenseCqualityCmaintenanceCompetitor Monitor AI - MCP server providing AI-powered tools and automation by MEOK AI Labs112 npm37 PyPIMIT
- AlicenseAqualityDmaintenanceEnables brand visibility monitoring across major AI platforms like ChatGPT, Claude, Gemini, and Perplexity. It allows users to track visibility scores, analyze competitor data, and receive actionable insights to improve AI-generated brand recommendations.1611 npm1MIT

industrylens-mcpofficial
AlicenseNot gradedqualityCmaintenanceBrowse IndustryLens's published competitive-intelligence reports and head-to-head competitor comparisons from any AI agent — real, source-backed data.MIT
Glama MCP Gateway
Add one secure layer between your agents and this server.