Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@BattleGrid MCP Server List current market grid sessions"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MCP server for BattleGrid — play crypto prediction games, author trading strategies, and manage intelligence agents from AI agents.
It is a thin, authenticated stdio proxy to BattleGrid's remote MCP server (Stripe @stripe/mcp pattern — no business logic). It discovers tools, prompts, and resources live from the server and re-exposes them to local MCP clients (Claude Desktop, Claude Code, Cursor). Capabilities are always discovered live — this package never hardcodes the tool catalog.
The package version no longer tracks the server's contract, and no longer claims to. Through v30 it did: the proxy announced battlegrid@<package version> downstream, so the published number was read as the contract a client would reach, and keeping the two in step was a manual release chore. It did not hold — the package sat at 11.0.0 against a deployed contract of 19.3.0 for ten days, 5.1.0 was declared here and never published at all, and four of the last five releases were version bumps carrying no code change.
From v31 the proxy reads the contract version out of the upstream handshake at connect time and relays it verbatim. What a local client reads is therefore what the connected server just announced — correct by construction, on every connection, with no release involved.
Two versions, two meanings, and they are expected to differ:
What it describes | Where to read it | |
Package version | This proxy's own code — a fix here, a dependency bump, a docs correction |
|
Contract version | The server's wire contract, live | The stdio handshake ( |
Seeing package 31.x alongside handshake battlegrid@33.x — the package behind the contract — is the system working, not a missed release. That is the pair that looks alarming and is not: the contract moved, and no release here was needed. Both numbers are printed to stderr at startup, labelled.
What this changes for you: nothing about how you call anything. Upgrading the package no longer waits on a server deploy, and a server deploy no longer strands you on a package that names the wrong contract — reconnect and the announcement follows. Contract breaking-change notes are no longer keyed to package versions, since a contract move is no longer a release here; the v11-and-earlier notes below are kept as history, and the live vocabulary is always discovery.
Related MCP server: limitless-mcp
Breaking at v109: submit_market_grid and update_market_grid are removed. v109 is the contract on
top of v108. A player's picks for a session live in their Market Grid entry draft — the same draft their
builder on the session page writes as they pick — and an entry is placed or changed by committing it.
submit_market_grid and update_market_grid leave discovery with no alias. A call gets the
standard unknown-tool refusal. Play is stage → get → commit:
stage_market_grid_entry_draft({ request: { sessionId, draftVersion, axes } }) with one
CELL_n: { cell: { coinId, prediction } } per grid position (CELL_0 is the captain) and a
RATIONALE carrying the reasoning, confidence, model name and per-cell pick reasoning the removed
tools required; get_market_grid_entry_draft for the diff, diagnostics and impact; then
commit_market_grid_entry_draft({ request: { sessionId, draftVersion, expectedRevision } }) on the
player's word. random_submit_market_grid and submit_agent_grid are unchanged.
get_market_grid_entry_draft, stage_market_grid_entry_draft, discard_market_grid_entry_draft
and list_market_grid_entry_drafts (mcp:read) read and stage the player's picks for free: no
entry, fee, deposit or LLM access. commit_market_grid_entry_draft (mcp:wager) enters a session
the player has not entered, charging the session's entry fee (the read's impact.entryFee), or
replaces a live entry's picks for no fee.
An entry that already exists is a revision CONFLICT naming the live entry's revision, with
nextAct: get_draft: read the draft again and commit over the revision it returns.
Breaking at v108: signal rules, the aggregate-score gate and the required-count gate are removed.
v108 is the contract on top of v107. A strategy's deterministic entry gate is its conditions: a
required condition, or a condition that carries a verdict. A strategy that declares neither never
trades, and its draft says so with the warning NO_ENTRY_CONDITION.
simulate_aggregate_score, list_strategy_signals, get_strategy_signal_definition,
derive_strategy_rule_view and get_coin_signal_preview leave discovery with no alias. A call
gets the standard unknown-tool refusal.
stage_strategy_draft refuses the SIGNAL_RULES and SETUP_GATES axes as unrecognized keys.
The ATR floor minAtrPct is staged in the TRADE_LEVEL_POLICY axis, beside the stop-loss multiples
and the risk-reward minimum; a TRADE_LEVEL_POLICY stage without it is refused.
decide_trade cites entry.conditionChecklist: 1–8 items { conditionKey, verdict, interpretation } over the decision's evaluated entry conditions, each key at most once, CONFIRM
only on a condition that read TRUE. signalChecklist and signalModulesUsed are refused.
The signal-log and agent-profile filters refuse dominantBias.
Strategies, agents and drafts lose signalRules, signalRuleTally, minAggregateScore and
minRequiredCount; the trading-config catalog loses the gate dial steps.
Qualification and scans (get_agent_coin_qualification, scan_agent_coins, scan_coin_agents)
lose the score and required-count gates, the score shortfall and the signal readings. Scan rows rank
qualifying first, then by ticker or agent name.
Radar reads lose every score field: the activity events, the close decision, the curve samples
({ at, atrPct, qualified }) and the curve digest.
Entry decisions carry conditionChecklist in place of signalChecklist and signalModulesUsed.
The public signal-log detail serves it as { name, outcome } only.
Signal logs and signal performance lose the aggregate score, bias, conflict and count fields; a
signal-log detail loses its scorecard and serves candidateLevels. The agent explorer loses its
average score.
The qualification and screen reasons lose AGGREGATE_BELOW_MIN and REQUIRED_COUNT_BELOW_MIN
and gain NO_ENTRY_CONDITION, which is the first reason a strategy fails. The attempt reasons lose
NO_ALLOCATED_SIGNALS, and the error codes lose REPAIR_REQUIRED.
A client that stages a signal rule or a gate, cites a signal in decide_trade, or reads a score field
breaks at v108, which is why this is a MAJOR.
Additive at v107.1: a paid entry can be recorded while its fee confirms. v107.1 is the contract on top of v107. The server now records your entry first, then sends the fee, then settles the entry on the venue's answer, so a fee can never move for an entry that was not recorded. No input changes.
submit_market_grid, random_submit_market_grid and submit_agent_grid results gain
entryStatus: ENTERED, or PAYMENT_PENDING when the entry is recorded and its fee is still
confirming at the venue. A payment-pending entry plays once its fee lands before lock; otherwise it is
released, and a fee that lands after that is refunded.
entryState gains PAYMENT_PENDING on get_my_market_grid_entry. list_my_market_grid_entries
shares the element schema but never returns one.
details.wagerErrorCode gains ACCOUNT_NOT_UNIFIED on the three submit tools: an entry fee is
paid only from a unified-collateral account (unifiedAccount or portfolioMargin).
check_market_grid_submission answers hasSubmitted: true for a payment-pending entry.
update_market_grid refuses a payment-pending entry with CONFLICT until it is in play.
Breaking at v106: Two-Look entry is retired. A deployment slot drops entryStrategy, and a
deployment draft refuses one. v106 is the contract on top of v105.5. Every automated entry now looks
once, inside the Opening Bell window: from startsAt (the open of the one candle a session scores)
until lockAt (entries close).
entryStrategy leaves every deployment slot: get_deployment_policy, list_deployment_policies,
get_deployment_policy_draft, pause_deployment_policy, delete_deployment_policy,
preview_deployment_resolution and test_generate_deployment_grid.
stage_deployment_policy_draft refuses entryStrategy on a RULES rule and on the
DEFAULT_SLOT, as an unrecognized key. preview_deployment_resolution's SLOTS subject and
test_generate_deployment_grid no longer publish the key, and they strip it if it is sent.
Every Market Grid session surface gains startsAt, the Opening Bell. That covers
get_market_grid_session, list_market_grid_sessions, get_market_grid_results, and a deployment's
target session and open entries. A Coin Grid summary's startsAt equals its lockAt.
get_agent_prompt_context_preview's market context gains startsAt and
millisecondsSinceOpeningBell, both nullable.
lockAt keeps its name and now means "entries close", the end of the Opening Bell window, a
fixed fraction of the session's duration after startsAt. Read startsAt for when a session starts
or what it scores. Submitting, editing and cancelling stay open until lockAt.
SESSION_START rules match startsAt.
A client that reads entryStrategy from a slot, or stages one into a deployment draft, breaks at v106,
which is why this is a MAJOR.
Breaking at v103: a transform entry drops sideRequired and operandRequired, and
get_strategy_column_contract validates its ticker as a coin symbol. v103 is the contract on top of
v102.1.1, which changed only two deployment-draft input descriptions.
sideRequired and operandRequired leave every transform entry get_metric_construction_hints
serves. Each restated a field the same entry already carries: read
authoring.parameters.side.required and authoring.parameters.inputs.required instead, where an
absent parameter means the transform does not accept it.
get_strategy_column_contract's benchmarkTicker is a coin symbol: trimmed, 1–20 characters of
the coin-symbol charset. Any non-empty string was accepted before.
The fixed 64 maximums leave strategy authoring: a column's window and offset, a custom
section's columns and the conditions arrays. The configured budgets list_strategy_vocabulary
serves under budgets judge them instead — sectionColumns, strategyConditions,
conditionClauses and columnLookback. window and offset keep their integer floors.
The column contract gains lookback, measurement and seriesHome.
lookback: { used, cap } — the bars the column reads, against the lookback budget a save holds it to.
measurement: { transform, chainedTransform } — how each transform reads its series: readShape,
windowUnit, priceBasis and nullPolicy. chainedTransform is null for an unchained column.
seriesHome — where the metric's series is kept: candle, history, or null when it has none,
which is why a series transform does not compose on it.
list_strategy_vocabulary gains conditionGrammar: the group depth, the operator and group
operator vocabularies and the verdicts a condition is written in.
A category example gains clause: { op: 'between', low, high }, or null for an example with
no recipe clause. The structure family gains two recipes — VWAP × distance held between -2 2
("price within ±2% of VWAP") and STRUCT_ZONES × nearestZoneDist(support) held between 0 1.5
("approaching support").
A window or offset declaration gains minimum, the smallest value the transform accepts.
A column whose read size exceeds the lookback budget is refused with
REPORT_COLUMN_LOOKBACK_EXCEEDED by get_strategy_column_contract exactly as by a save, before
the column is built. A window too large to build was a construction failure before.
A value below a declared minimum is refused naming the transform that declares it, including
a chained transform's window.
A spread with the wrong number of operands is refused naming the operand its operand order declares.
A regime timeframe reference resolves to the anchor's regime rung when the request names an
anchor timeframe. It resolved to null before.
A client that reads sideRequired or operandRequired, or sends a ticker outside the coin-symbol
charset to get_strategy_column_contract, breaks at v103 — hence the MAJOR.
Additive at v102.1: three read tools. v102.1 is the contract on top of v102. Nothing a client sends is refused and nothing it reads is reshaped.
All three are mcp:read and read your own agents only.
list_agent_open_orders — the live orders (PENDING, ACKED, WORKING) of the positions your
agents hold, newest first, each naming its agent.
list_agent_trades — your agents' closed trades, newest close first, paged with nextCursor and
narrowed by coinTicker and agentId, with a summary of the whole filtered scope: trade count, win
count, win rate and net P&L in USD. Another user's agent reads as an empty scope.
get_agents_portfolio_pnl — your agents' combined P&L over 24H, 7D or 30D, as 48 dated
points ending on the period figure.
Breaking at v102: list_my_market_grid_entries moves its counts, summary and agent options under one
facets object, sent with the first page only, and every entry gains a payout state. v102 is the
contract on top of v101.1.
outcomeCounts, summary and agentOptions move under facets, unchanged inside it:
facets.outcomeCounts, facets.summary, facets.agentOptions.
facets comes with the first page only. A page you read with a cursor carries facets: null.
Keep the first page's facets for the whole scroll: every page of one read shares the same filters, so
the first page's counts, summary and agent options describe all of them.
Every entry gains payoutState, in list_my_market_grid_entries and get_my_market_grid_entry:
RECORDED once settlement has recorded both of the entry's payouts, UNDER_REVIEW if it left one
unrecorded, and null while the entry is in play or cancelled.
facets.summary gains payoutsUnderReviewCount: how many settled entries the money figures leave
out because their payout is under review.
An entry under review has netPnl: null. It is listed under PLAYED only, never under PAID or
NOT_PAID, and the summary leaves both its fee and its payout out of every money figure, so the totals
stay consistent. Today no entry is under review: the entries settled before settlement recorded both
payouts now record 0 for the payout they never received, so their net reads the same as before.
The five filters (window, outcome, source, agentId, cursor) accept exactly what they
accepted before. Only the agentId description changes: it now points at facets.agentOptions.
A client that reads outcomeCounts, summary or agentOptions at the top level, expects them on every
page, or rejects unknown entry fields, breaks at v102 — hence the MAJOR.
Additive at v101.1: the catalog gains three swing-structure metrics. v101.1 is the contract on top of v101; nothing you send today is refused and nothing you read is reshaped.
SWING_FRACTAL_HIGH and SWING_FRACTAL_LOW — the most recent swing high and swing low: a bar
whose high is at or above the two highs before it and strictly above the two after it (Pine's
ta.pivothigh(2, 2) tie rule), and the mirror on the lows. A swing is published from the bar that
completes its right side — never under its own bar, never from a forming bar — and held until the
next swing on its side. They are candle price levels: every transform a candle level takes,
levelInteraction included, and they rank across the board as dist_fractalHi / dist_fractalLo.
SWING_FRACTAL_LABEL — the most recent swing against the previous one on its side: HH / LH for
a high (an equal high is LH), HL / LL for a low (an equal low is HL), none when no comparable
pair is in reach. A held state, value-only; on too short a history it reads nothing.
Break of structure is dist_fractalHi crossesAbove 0 / dist_fractalLo crossesBelow 0; a step of
the level onto a newly confirmed swing never fires it.
The catalog and vocabulary surfaces list the three keys, and the authoring digest's metricKeys gains
them.
The headers are fractalHi / fractalLo / fractalLabel. The swingHi / swingLo stems retired at
v53 (the Donchian channel's old names) stay refused: dist_swingHi is still CONDITION_COLUMN_UNKNOWN.
A client that treats the metric vocabulary as a closed set re-reads it; no request is refused — hence the MINOR.
Breaking at v101: archive_intelligence_agent refuses an agent held in any radar deployment, and its
refusal names a fourth blocker. v101 is the contract on top of v100.
archive_intelligence_agent refuses an agent that holds a slot in any radar deployment, paused ones
included. The input schema is unchanged: a call that archived such an agent before now gets the
existing CONFLICT. A paused radar still holds its slots, so pausing does not clear it. Find the coins
with list_radar_deployments, then commit a radar draft without the agent or call
delete_radar_deployment, and archive again.
details.archiveBlockers[].reason gains RADAR_DEPLOYED, after DEPLOYED, OPEN_TRADES and
ACTIVE_SESSION. Its detail reads in N radar deployment(s), counting every radar deployment that
holds the agent.
An archived agent holds no deployment slot in either venue, so activate_intelligence_agent arms
nothing: the agent trades again only once you deploy it.
A commit_radar_deployment_draft or commit_deployment_policy_draft that races an archive of one of its
agents is refused with the existing CONFLICT. Read the draft again; its diagnostics name the archived
agent.
A client that reads archiveBlockers[].reason as a closed set, or expects an archive of a radar-deployed
agent to succeed, breaks at v101 — hence the MAJOR.
Breaking at v100: the bandTouch transform is retired and levelInteraction takes its place; a column
gains a buffer knob; the catalog gains twelve bar-geometry metrics. v100 is the contract on top of
v99.
A column naming bandTouch is refused by every tool that takes a column — stage_strategy_draft,
preview_strategy_report, derive_strategy_rule_view, get_strategy_column_contract and
get_metric_construction_hints — as any unknown transform id is. Where you want "where the close sits
in the bands", read BB_TOUCH × value; where you want "what the bar's range did at a band", use
levelInteraction below.
levelInteraction classifies the newest bar of a column against a price level as
touch_from_below, touch_from_above, reject_from_below, reject_from_above or none. The side is
where the prior close stood against the prior bar's level. A reject — the range pierced the level and
the close came back — reports reject, never touch, so "touched or rejected" is an in clause over both
labels of a side. A close through the level reads none: the break is distance crossesAbove 0 (or
crossesBelow 0) over the same level.
It is offered on a level with a value at both bars it reads: a candle-homed level (a band, VWAP, a
moving average, ST_LINE, PSAR, the Ichimoku lines) or a prior-session level or pivot (PDH,
PDL, PDO, the floor pivots, PRIOR_TPO_*). A prior-session level is a single read and takes no
hold. It is not offered on a price fact (CLOSE, MARK, HLC3), a developing-session level
(TPO_*), the naked points of control, or the Donchian rails.
The header is {code}_lvl — bbUpper_lvl.
buffer, a column knob declared by levelInteraction only: whole basis points of the level,
0–500, default 0. Only a touch reads it. A non-zero buffer names the header (bbUpper_lvl_b25),
so two buffers are two columns. buffer on any other transform is refused as an undeclared parameter.
Twelve metrics, all in the price family:
candle patterns — CANDLE_DOJI (boolean), CANDLE_PIN (lower_wick / upper_wick / none),
CANDLE_ENGULFING (bullish / bearish / none), BAR_CONTAINMENT (inside / outside /
none), each judged against the ten bars before it at TA-Lib's default settings;
gaps — GAP and GAP_RANGE_CLEAR (up / down / none) and GAP_PCT, the signed % from the
prior close to the open;
run lengths — CLOSE_RUN and CANDLE_RUN, consecutive bars one way, capped at ±20;
price sources — HL2, HLC3, OHLC4, price facts that take every transform a close takes
(HLC3 × aggregate(20) is the 20-bar SMA of typical price).
The patterns, gaps and runs compose value only, and none of the twelve is ranked.
effectiveParameters gains buffer — the declared default 0 or your value on a
levelInteraction column, null where no transform in the chain declares it.
A transform's authoring parameters may declare buffer, and the transform and section column-role
vocabularies read levelInteraction where they read bandTouch.
The authoring contract digest's metricKeys gains the twelve keys.
A client that sends bandTouch, or reads effectiveParameters with a closed key set, breaks at v100 —
hence the MAJOR.
Breaking at v99: four tools are retired, two are added, and get_agent_journal loses a field. v99 is
the contract on top of v98.
Every Market Grid entry you made — by your deployed agent, an agent you asked for, an MCP client, Commander, a grid you built or dice you rolled — is now read from one history, and only by you.
get_agent_game_history, get_user_agent_game_history, get_public_agent_game_history and
get_mcp_reasoning_journal are gone, with no alias; a call gets an unknown-tool error. The
replacements:
get_agent_game_history { agentId } → list_my_market_grid_entries { agentId };
get_user_agent_game_history → list_my_market_grid_entries with no agentId;
get_public_agent_game_history → none: arena entry history is owner-only, so no public read
replaces it;
get_mcp_reasoning_journal { sessionId } → get_my_market_grid_entry { sessionId }, whose
reasoning is the reasoning recorded with the entry's picks.
list_my_market_grid_entries { window?, outcome?, source?, agentId?, cursor?, limit? } (mcp:read)
pages every entry you made, newest first by its first submission. outcome is PLAYED (the default:
every entry except a cancelled one), IN_PLAY, PAID (a payout above zero), NOT_PAID or
CANCELLED; source and agentId filter by the entry's original author. The same read returns one
count per outcome, a summary (paid rate, totals, net, averages) and the agents you can filter by. Page
with nextCursor.
get_my_market_grid_entry { sessionId } (mcp:read) returns your entry in one session — the same
record a list row carries. A session you never entered is NOT_FOUND.
Each entry carries its entryState (WON, PLACED, NO_WIN, VOID_NO_SUBMISSION, IN_PLAY or
CANCELLED), its entryFee and signed netPnl (null while in play or cancelled), a cancelled entry's
refundState, its originalSource and original agent, and edited — true once another author has
rewritten the picks. reasoning is null on an edited entry, because the record no longer describes the
picks. Neither tool accepts a user identity: both read the authenticated account's own entries.
get_agent_journal no longer returns recentGames. An agent's arena entries are read with
list_my_market_grid_entries { agentId }.
A client that calls a retired tool, or reads recentGames, breaks at v99 — hence the MAJOR.
Breaking at v98: a condition clause accepts six operators that compare a column with its own previous completed bar, and a clause's evidence gains a second shape. v98 is the contract on top of v97.
A condition clause on stage_strategy_draft and preview_strategy_report accepts:
crossesAbove / crossesBelow with a number value — TRUE when the value is above (below) it on the
decided bar and was at or below (at or above) it on the bar before. A bar landing exactly on the level
has not crossed.
increased / decreased with no literal — strictly above (below) the prior bar's reading. A
value beside them is refused.
enters / exits with one label — the label read now and not on the bar before, or the reverse.
Numeric headers take the crossings and changes; classification, direction and boolean headers take the
transitions; event and rank headers take none of them. A strategy's served conditionOperators
(preview_strategy_report at detail: "detailed", and the scalar vocabulary) lists exactly what each
header admits on that strategy; get_strategy_column_contract, which reads no strategy's stored history,
lists the operators the header's kind takes.
Each of the six reads two completed bars, never the forming one, on every basis, so its reading is never
provisional. A failed read on either bar is UNRESOLVED; a bar with no value reads FALSE.
A header whose stored history keeps only one bar (conditionHold.maxWindow 1) admits none of the six:
the clause is refused with CONDITION_OPERATOR_UNSUPPORTED, and every operator refusal's
allowedDomain now lists the operators that header admits.
crossesAbove, crossesBelow, enters and exits are edges: a hold with atLeast above 1 is refused
(CONDITION_HOLD_ILLEGAL) — hold them "within n closes". A prior-bar clause's largest window is one bar
less than its header's, and the frame-read budget counts the bar before.
A prior-bar clause's evidence carries priorOperand — the same column's reading on the bar before —
and a decidedBar that is never null. It appears on preview_strategy_report,
get_radar_close_decisions, list_gate_blocks, preview_radar_resolution, get_signal_log,
get_agent_coin_qualification (a condition level's clauses) and the trade register an agent reads.
A current-bar clause's evidence is unchanged and carries no priorOperand; every record written before
v98 reads as it was written.
Rendered evidence lines read prior before current: RSI14_now crosses above 50 (49.6 → 50.4).
A condition series' BOUND threshold op widens to crossesAbove / crossesBelow, drawn at the level
crossed. A change or a transition draws no threshold.
The authoring contract digest gains domains.conditionOperator, the thirteen operator names.
A clause reader that assumed one evidence shape, or an operator vocabulary of seven, breaks on the same call at v98 — hence the MAJOR.
Breaking at v97: an evaluated signal's evidence keys move, BB_TOUCH touches by %B, and a crossing is
read on the bar that leaves zero. v97 is the contract on top of v96.
An evaluated signal's indicatorValues — on get_signal_log, get_public_agent_signal_log_detail and
get_coin_signal_preview — records the last price a live-priced signal evaluated under last_price.
Six signals recorded it under price: MA_SMA200_ABOVE / _BELOW, SR_AT_SUPPORT / _RESISTANCE and
BOLLINGER_LOWER_TOUCH / _UPPER_TOUCH; the three structure-approach signals now record it too.
price means the last closed close on every signal that records it.
The Bollinger touches add bb_upper and bb_lower, and their bb_percent_b is the live %B they
compared — never the closed-bar %B.
The EMA 5/13 crosses add ema_cross (1 / −1), and the sector comparison signals add ema5, ema13
and ema20.
MFI_SUSTAINED_* and CVD_BULLISH / CVD_BEARISH record their prior-bar value when there is one, and
omit it when there is not.
A signal records exactly the keys its declared vocabulary names. Records written before v97 keep the keys they were written with, so a reader of history meets both.
Crossings. One zero-cross rule decides every crossing: a value that reaches zero has not crossed; one
that leaves zero has. Every crossDetect column (MACD_cross, PPO_cross, ROC_cross and authored
ones), the EMA_CROSS label, the MACD / PPO / stochastic cross signals and FUNDING_RATE_FLIPPING read
no crossing on a bar that lands exactly on zero. They read the crossing on the bar that leaves it.
BB_TOUCH and bandTouch classify a touch when %B is strictly below 0.05 or above 0.95 — the
touch signals' default thresholds. Before, a price within 2% of the band width of an edge, inclusive,
touched. A degenerate band reads none, and a BBtouch condition now agrees with the scorecard's touch
at defaults.
RSI zone. The projected RSI state is overbought strictly above 70 and oversold strictly below 30, as
the RSI14_zone columns already read it.
Descriptions. A served rule description states the thresholds that rule evaluates, and the ROC descriptions read "positive and rising".
get_strategy_signal_definition's indicators[].meta gains optional: true on a key the signal records
only when its read is present — prev_cvd_value on CVD_BULLISH / CVD_BEARISH, prev_mfi14_value on
MFI_SUSTAINED_*. Every other key is always recorded when the signal has the data to evaluate.
A key read at v96 is absent at v97 on the same signal, and a touch or a crossing read at v96 can read differently on the same bar — hence the MAJOR.
Breaking at v96: efficiency admits only a level or a price-fact series, and nearestZoneDist changes
sign. v96 is the contract on top of v95.
efficiency over an increment or an oscillator metric — CLOSE_CHANGE, VOLUME, RSI14, FUNDING_RATE
and the like — is refused at construction, and so is efficiency as the outer of distance or spread
(VALIDATION_ERROR, authoring code REPORT_COLUMN_CONSTRUCTION_FAILED or REPORT_COLUMN_CHAIN_UNSUPPORTED),
by every tool that builds a report column. distance and spread chain trajectory, aggregate and
rank; the column contract and the vocabulary tools serve exactly those sets.
BB_PCT_B no longer composes maxShare: %B leaves [0, 1] when price closes outside a band.
A rank column with no ordering is gated as the default hi it ranks by, so it is refused wherever a
written hi is (REPORT_COLUMN_PAIR_UNSUPPORTED). Write the ordering you mean.
transformId is the closed enum of the seventeen transforms and chainedTransformId the enum of the five
chain outers, where each was any 1–40 character string. An id outside them is refused by schema.
nearestZoneDist is price's distance from the zone midpoint, ((price − mid) / mid) × 100: positive
when price is above the zone, negative below it. It was the midpoint's distance from price, with the
opposite sign. A range symmetric about zero reads the same window; a one-sided bound needs its sign
flipped.
BB_PCT_B offers all four rank orderings.
A classifyZone column declares and resolves its own metric's vocabulary in canonical labels — ADX's
strength bands, MFI's five bands — not a shared overbought / oversold / neutral set.
highDev, lowDev and oiRegime read live sources and declare the live clock, so a condition over them
is provisional on the live frame.
A chained column's formula binds its outer half to the inner column's header (dist_VWAP), and every
relative-change formula reads ((value − reference) / reference) × 100.
An authoring parameter's defaultValue on get_metric_construction_hints is typed per knob — window
and offset integers, ordering a rank ordering — and absent on side, inputs and bars.
A rendered section column's role is one enum of the same members, where it was a union of an enum and
five literals.
An input v95 accepted is refused at v96, and a value read at v95 reads with the opposite sign — hence the MAJOR.
Breaking at v95: a gate block's conditionVerdict is the verdict applied, not the one read. v95 is the
contract on top of v94.
On list_gate_blocks, a conditions-stage reasonDetail.conditionVerdict is the verdict applied — the
one bound for the pair, which can be a held verdict this evaluation did not read — where it was the verdict
the conditions read. It is served only on a CONDITION_VERDICT_OPPOSES_DIRECTION or
CONDITION_EXIT_RULE_CLOSES_DIRECTION block recorded since v95; a REQUIRED_CONDITION_FALSE block, which
no verdict decided, and a block recorded before v95 no longer carry it. Read observedVerdict for what the
conditions read.
The reason detail gains observedVerdict, served on every conditions-stage block, and heldVerdict,
verdictDwellResolutions and verdictDwellRequired — the dwell state the binding was resolved from,
served beside conditionVerdict and absent where it is. get_signal_log and
get_public_agent_signal_log_detail embed the same schema and never fill these keys: a block writes no
signal log.
get_signal_log's conditionEvaluation.verdict and get_agent_journal's marketSnapshot.conditionVerdict
keep meaning the verdict the evaluation observed.
A key read at v94 answers a different question at v95 — hence the MAJOR.
Breaking at v94: a timeless metric refuses an offset, and a value with no reading on its bar reads absent
instead of an older bar's. v94 is the contract on top of v93. A column reads the bar it names, or nothing.
A value column over a timeless metric — one get_strategy_column_contract serves with timeframeMode
"timeless": funding, the regime, the previous-session levels and the other bundle reads — with offset above
0 is refused at construction (VALIDATION_ERROR, authoring code REPORT_COLUMN_CONSTRUCTION_FAILED) by every
tool that builds a report column, the strategy and agent draft tools and preview_strategy_report among them. It
used to be accepted and ignored. Drop the offset: a timeless metric has no earlier bar.
A trajectory, zone, state or cross value whose current bar holds no reading is now absent, instead of the newest older bar's value. A trajectory's slots keep a missing bar as an empty slot, in place.
A window — a trajectory's, efficiency's, maxShare's — counts the newest bars, present or not.
A candle label column at offset N reads bar t−N's label; it used to read the latest.
On a decided bar, a condition over a published change (chg5m … chg24h) or oiRegime reads
unresolved, and a condition over a higher-timeframe column reads unresolved during that bar's close lag.
An Arena deployment that routes on the regime reads WARMING while its regime bar is landing — the regime
for the bar the clock just closed is not published yet — and plays nothing, its default slot included. Generating,
submitting or updating a grid for an agent on such a deployment is refused with SERVICE_UNAVAILABLE in those
seconds; retry.
Something read at v93 reads absent at v94, and an input v93 accepted is refused — hence the MAJOR.
Breaking at v93: a trajectory column below two observations is refused. v93 is the contract on top of v92. Nothing in a schema moves.
A custom column { transformId: "trajectory", window: 1 }, plain or as the chained outer of
distance × trajectory, for any metric: stage_strategy_draft reports REPORT_COLUMN_CONSTRUCTION_FAILED
in its diagnostics, naming the minimum, commit_strategy_draft refuses it, and preview_strategy_report
refuses a report carrying one with the same error. A window of one has no build-up, and its trend compared a
reading with itself, so it read flat for every coin on every bar. The trajectory's served window
description states the minimum.
The CVD section renders its buy-pressure, buy-trade and sell-trade columns as window-4 trajectories —
_t3, _t2, _t1, _now, _trend each — and every _now keeps its value. Their _trend cells read a
direction where they read flat, and the CVD-crowd convergence classifies from that buy-pressure column, so
captainConf can read true and conf can read high.
Something accepted at v92 is refused at v93 — hence the MAJOR.
Breaking at v92: the section fields are reshaped, the report preview's default narrows, and its size is measured
in characters. v92 is the contract on top of v91. Every tool that published a section as markdown text publishes
it as the JSON object the agent reads, and preview_strategy_report takes a detail level.
A section read is { shape: "table", section } — { id, title, notes, provenance?, anchorTf, groups, columns, rows, addendum? }, rows positional and every cell its raw value, each column stating its unit, precision, bar
set (bars) and, on a rank column, the ranked set's universe, with a key absent where the column has nothing to
say — or { shape: "facts", section }, { id, title, facts, glossary }. It replaces:
preview_strategy_report.renderedSections[]: { sectionKey, section: { title, text }, structure, authoredNote } → { sectionKey } and a section read, or { sectionKey, shape: "skipped" } for a row that
rendered nothing. A custom section's authored note is its section's notes.
get_market_context.sections[]: { kind, title, content, structure } → { kind } and a section read.
get_context_source_full_preview.content → read, a section read or null.
get_context_sources_preview: each source's value → section, a section read or null, a coin-row
table sampled to its first five coins.
get_signal_log.reportSections[]: a section read, skipped, or { sectionKey, shape: "text", title, text }
— a row stored before custom sections carried a structure, kept as its agent read it.
preview_strategy_report.conditionsTableText and .tradeConditionsBlockText → gridConditions and
tradeConditions, each { glossary, register }: the glossary prose and the register object the text carried
as one JSON line. Each is null where its text was.
decidedBy joins the conditions register: on each GRID coin and on the TRADE register, the key of the
condition the verdict came from, null when no single condition made the call. It reaches you in
gridConditions.register, tradeConditions.register, and the register get_agent_prompt_context_preview shows,
and the conditions glossary gains the sentence defining it.
preview_strategy_report detaildetail is "concise" (the default) or "detailed". A concise preview is exactly what the agent reads.
conditionColumns and conditionOutcomes leave the top level. With each row's headerBindings
({ sectionKey, columnIndexByHeader }) they make up authoring, served only at "detailed" and null
otherwise. If you read either field, ask for "detailed" and read it under authoring.
marketReadMarkers[] lose resolvedName and resolvedValue.
tokenCountModel and budgetUsage.estimatedTokens leave the preview. budgetUsage.resultChars is
the exact JSON.stringify length of the response that carries it, its own digits included, against the cap.
list_strategy_vocabulary drops budgets.estimatedTokens, and previewExecutionLimits.maxResultBytes is
renamed maxResultChars.
A preview over the cap is refused with PREVIEW_LIMIT_EXCEEDED and limit: "result_chars";
estimated_tokens and mcp_result_bytes are retired. Every preview refusal now carries its previewErrorCode
and its bound (deadlineMs, or limit, actual and maximum) in the error's details.
preview_strategy_report, get_market_context and get_context_sources_preview declare the cap as
_meta["anthropic/maxResultSizeChars"], so Claude Code keeps a large result inline.
Something read at v91 is reshaped or gone at v92, and a default narrows — hence the MAJOR.
Breaking at v91: one output field is removed, and the published text that is the agent's prompt carries JSON.
v91 is the contract on top of v90. Every data block BattleGrid's trading agents read — the strategy's resolved
conditions, fact sections, and the trade-decision blocks — is one JSON object naming itself with id and title.
Instructions stay prose under their ## Title. The tools that show you the prompt therefore show that JSON.
promptDataSerialization leaves get_signal_log's owner detail. There is one serialization, so a trade
record stores none. reportSections and missingData are unchanged, and a trade recorded before capture still
serves both as null.
preview_strategy_report.conditionsTableText is the conditions glossary as a ## Strategy Conditions Glossary instruction, then the conditions register as one JSON line: surface: "GRID", and coins[] each with
outcomes keyed by condition and verdict by display name (null when the strategy declares no verdict). It was a
markdown table followed by the glossary.
preview_strategy_report.tradeConditionsBlockText is the trade glossary, a blank line, then the target coin's
register as one JSON line: surface: "TRADE", verdict as { name, description }, and conditions[] each with
outcome, required, provisional, counts, exitRule and the evaluator's evidence. It was a
## STRATEGY CONDITIONS heading, a table and evidence bullets.
A fact section's text is its JSON, { id, title, facts, glossary } with facts keyed by label, in
get_market_context.sections[].content, get_context_source_full_preview.content,
preview_strategy_report.renderedSections[].section.text, and the get_signal_log report of trades recorded
from v91 on. The session field is a fact section. A table section's text is still its markdown.
get_agent_prompt_context_preview.promptSections[].content is each section as the agent reads it: a data
section's JSON, a table's included, and an instruction's prose.
get_agent_prompt_context_preview.promptSections[].kind gains strategy-conditions-glossary, the instruction
read before the conditions register. The market-context reads exclude it, as they exclude the register.
Something read at v90 is gone or reshaped at v91 — hence the MAJOR.
Breaking at v90: input is refused and reshaped, output is reshaped, five error codes are removed. v90 is the contract on top of v89. The Arena and radar deployment drafts now run the same five-tool lifecycle as agents and strategies, and the preview token is retired: a deployment commits on the draft version and the revision you read.
previewToken and confirm leave commit_deployment_policy_draft, commit_radar_deployment_draft,
resume_deployment_policy and resume_radar_deployment.
confirm leaves delete_deployment_policy, delete_radar_deployment, discard_deployment_policy_draft and
discard_radar_deployment_draft.
Every input is strict, so a call naming either key is refused as an unrecognised key.
The commits take { request: { presetId | coinId, draftVersion, expectedRevision } }: exactly the
draftVersion and committedRevision the kind's get_*_draft returned (expectedRevision: null for a first
deployment).
The resumes take { presetId | coinId, expectedPolicyId, expectedRevision }: the policyId and revision a
COMMITTED preview returned. Any other deployment or revision is a CONFLICT, and nothing is armed.
delete_radar_deployment gains expectedPolicyId beside expectedRevision, as delete_deployment_policy
already took. A delete naming a policy since removed and redeployed matches nothing.
The discards take { request: { presetId | coinId, draftVersion } } with no dry run. A draft that moved, or
none, is DRAFT_VERSION_MOVED.
stage_deployment_policy_draft and stage_radar_deployment_draft gain diagnostics.
get_deployment_policy_draft and get_radar_deployment_draft answer the lifecycle read
{ draft, draftVersion, committedRevision, diff, diagnostics, impact }, where they answered { draft }. impact
is the resolution the commit would arm and whether it plays (enabledAfterCommit), plus a radar first
deployment's coin-cap reading (admission); it is null when a diagnostic refuses.
The commits answer { presetId | coinId, revision, draftVersion }. The radar commit answered { revision }.
The discards answer discarded: true, never false.
preview_deployment_resolution and preview_radar_resolution lose previewToken and gain policyId and
revision: the deployment the preview resolved over, null for SLOTS and a first deployment.
TOKEN_EXPIRED, TOKEN_BINDING_MISMATCH, INVALID_TOKEN_SIGNATURE, INVALID_TOKEN_FORMAT and INVALID_TOKEN_CLAIMS
leave the error vocabulary and every draft diagnostic's code enum — so the stage and get outputs of the agent and
strategy drafts move too.
A retried commit replays its receipt. Both commits are destructive and idempotent on
<presetId | coinId>:<draftVersion>, as the agent and strategy commits are.
Refusals name the next act. A moved draft or revision carries details.nextAct: "get_draft"; a refusal the
draft can repair (INVALID_DEPLOYMENT_POLICY, or the radar's VALIDATION_ERROR) carries "stage"; a retired arena
carries none and names the remedy.
A revision conflict names the revision that won in details.actualRevision, null when no deployment remains.
A DRAFT preview at a version the draft is not at is DRAFT_VERSION_MOVED.
A draft whose arena or coin can no longer be deployed is still served, with the refusal in its diagnostics and a null impact, so you can show it and discard it.
A radar resume of a policy already trading answers its revision and changes nothing.
The radar per-user coin cap is FORBIDDEN, with no nextAct, where it was VALIDATION_ERROR.
Something accepted at v89 is refused at v90 — hence the MAJOR.
Breaking at v89: acceptance narrows behind unchanged schemas. v89 is the contract on top of v88.1.
commit_strategy_draft, preview_strategy_report (DRAFT source)A create judges every dial of the new strategy at the operator bounds configured now. That includes a dial the author typed, one a fork copied from its source, and one left at the creation default. The operator bounds are the platform's ATR %, stop-loss ATR-multiple and risk:reward limits.
At v88.1 a create judged only the dials that differed from the fork's source or the creation default. So a copied dial outside a bound tightened since the source stored it was committed.
At v89 commit_strategy_draft (expectedRevision: null) refuses that create with VALIDATION_ERROR, naming the
dial and the bound, for example minAtrPct (0.1) must be >= 0.15. Nothing is created and the draft is unchanged:
stage the dial inside the bound and commit again.
preview_strategy_report with source: { kind: "DRAFT", … } composes the draft exactly as the commit does, so it
refuses the same draft with the same error where v88.1 rendered it.
stage_strategy_draft and get_strategy_draft report the error in their diagnostics before the commit.
fork_strategy's description now says this, so its description hash moves.
An edit still judges only what it moves. An update or a restore carries a dial it leaves unchanged, so a stored value outside a since-tightened bound stays readable and saveable on its own strategy.
The platform keeps its defaults creatable. An operator cannot set a bound that would put a creation default outside it, so a create that leaves a dial untouched is never refused for a value the platform seeded.
Something accepted at v88.1 is refused at v89 — hence the MAJOR.
Additive at v88.1: acceptance widens and one served value narrows; no input or output schema changes. v88.1 is the contract on top of v88.
commit_strategy_draft, restore_strategy, commit_agent_draftAn open position no longer refuses a strategy commit, a restore or a rebind. At v88 a write that
would leave an exit rule unreadable on a coin a bound agent held an open position on was refused with
CONDITION_UNREADABLE_BY_RADAR_SCAN or CONDITION_OPERAND_UNSERVED_IN_LANE. At v88.1 it commits. A radar
slot's coin and a pending manual request's coin are still checked, and refused exactly as before.
The three tools' descriptions now say this, so their description hashes move.
coinRoleThe refusal's error context serves coinRole as SLOT or MANUAL_REQUEST only. OPEN_POSITION is no
longer emitted. A client switching on it loses a member it can no longer receive.
An open position keeps what it opened with. An agent position records the exit rules it opened under, with the report and timeframes they read, and its exit rules are evaluated from that record for the life of the trade. An exit-rule, exit-policy or level edit reaches only positions opened after it; a revision, rebind, SYSTEM update or timeframe change after the open never reaches an open trade.
Nothing accepted at v88 is refused at v88.1.
Breaking at v88: a condition body accepted at v87 is refused, and condition and chart output is reshaped. v88 is the contract on top of v87.
A condition is an entry condition or an exit rule, never both. An exit rule names the side it closes,
where exit: true read that side from the condition's verdict.
stage_strategy_draft, preview_strategy_reportA condition no longer carries exit. A condition entry naming it is refused with the unknown-key
error, on the CONDITIONS axis of stage_strategy_draft and in the conditions of
preview_strategy_report's source: { kind: 'FIELDS', … }.
Every condition requires exitSide: LONG, SHORT or BOTH for an exit rule, null for an
entry condition. It is nullable, never optional; a condition without it is refused.
An exit rule carries verdict: null and required: false, and an entry condition cannot reference
an exit rule. The builder refuses each with a typed remediation error —
CONDITION_EXIT_RULE_CARRIES_VERDICT, CONDITION_EXIT_RULE_REQUIRED,
CONDITION_ENTRY_REFERENCES_EXIT_RULE — in a staged draft's diagnostics, and as the refusal of its
commit_strategy_draft. An exit rule may reference anything: "exit when the entry setup stops holding"
is an exit rule whose definition is NOT a conditionRef to the entry condition.
Every condition a strategy read or commit returns carries exitSide and no exit:
get_strategy, archive_strategy, restore_strategy and commit_strategy_draft.
get_agent_coin_qualification's chart names the side each exit rule closes. Every served level
gains referrers, { conditionKey, conditionName, exitSide } for each of its set's declared
conditions that reach it. Every series clause referrer and every conditionsWithoutLevels entry gains
exitSide, null on every entry-set referrer. The ENTRY set no longer lists exit rules; they are
listed under EXIT.
CONDITION_EXIT_RULE_CLOSES_DIRECTION joins the qualification gate code, the trade evaluation
attempt reason, the trade gate reason and the trade screen reason. It can appear in the output of
get_agent_coin_qualification, scan_agent_coins, scan_coin_agents, list_gate_blocks,
get_signal_log, get_public_agent_signal_log_detail, get_radar_activity,
get_radar_activity_summary, get_radar_close_decisions, get_radar_deployment,
list_radar_deployments, preview_radar_resolution, propose_entry_decision,
get_trade_conversation, get_agent_budget and reset_agent_drawdown_baseline. A gate block under it
names the refusing exit rule in decidedBy.
BLOCKED_BY_CONDITION_EXIT_RULE joins the radar evaluation outcome, on get_radar_activity and
get_radar_activity_summary.
A client switching exhaustively on any of them meets a member it has not seen.
An exit rule closes exactly the side it names. A settled TRUE on a completed bar closes the
agent's open position on that side. No verdict chooses it; exit: true closed the side its verdict
opposed, and both sides for a NEITHER or null verdict.
While an exit rule reads TRUE, its side is not offered for entry. Scan qualification and compose
refuse that side under CONDITION_EXIT_RULE_CLOSES_DIRECTION, after any verdict refusal of the same
side.
A stored exit: true condition is split and closes what it closed. It keeps its verdict as an
entry condition, and its strategy gains an exit rule CLOSE_ON_<conditionKey>, a conditionRef to
it closing the side its verdict opposed (UP → SHORT, DOWN → LONG, NEITHER → BOTH). A
strategy read returns both.
Unchanged names keep their place. list_strategies' conditionTally.exit now counts exit rules,
and a decision record's condition declaration keeps exit: boolean, true for an exit rule.
A strict client that rejects unknown keys fails to parse every condition and every chart level at v88; list the tools again after the server deploys.
Breaking at v87: a fork's create can now be refused where it committed. v87 is the contract on top of v86. No schema, description or tool changes.
A create draft that fork_strategy opened is committed by commit_strategy_draft with
expectedRevision: null, or by the builder's Create. That create now resolves the strategy it copied
again. It uses only the source id and revision in the draft's ORIGIN, by the same rule the fork applied:
A source that moved past the forked revision refuses the create with CONFLICT and no
details.nextAct. Reading the draft again cannot cure it; the message says to discard the draft and
fork again. get_strategy_draft and stage_strategy_draft report the same refusal in the draft's
diagnostics.
A source you can no longer see — archived, or neither yours nor SYSTEM — answers NOT_FOUND.
Neither fork_strategy nor the create reads the source's stored revision snapshot. A strategy whose
stored snapshot predates the current shape — every SYSTEM template among them — now forks and creates
from its live state, where both were refused.
Breaking at v86: refusals on most tools lose details.nextAct, two refusals change code, and one
draft read is reshaped. v86 is the contract on top of v85.
details.nextAct follows the refusal's meaning. DRAFT_VERSION_MOVED and DRAFT_AXIS_CONTESTED
still carry get_draft from every tool. A revision CONFLICT and the in-flight commit refusal
(get_draft) and a validation refusal (stage) now carry a hint only from stage_<kind>_draft and
commit_<kind>_draft of the strategy and agent kinds.
Removed from every other tool's validation refusal and revision CONFLICT: submit_market_grid,
fork_strategy, archive_strategy, restore_strategy, and the arena and radar commit, pause, resume
and delete tools, and from the malformed-cursor refusal of list_agent_drafts and
list_strategy_drafts. The arena and radar commit_*_draft stale-revision refusals lose get_draft
too. A client that branched on nextAct from those tools reads the code instead.
The strategy active-quota refusal is FORBIDDEN, as the agent slot quota is, where it was
VALIDATION_ERROR with nextAct: 'stage': on commit_strategy_draft and restore_strategy.
A create naming an id that is already the caller's strategy or agent is DRAFT_VERSION_MOVED,
carrying the caller's draft version at that id, where it was NOT_FOUND. An agent create naming a
SYSTEM agent is FORBIDDEN, where it was NOT_FOUND.
The moved-draft refusals end "Read the draft again and show the player what it now holds before committing.", where they said to commit what it now holds. Read the draft again and show the player before committing; never commit the re-read draft on your own.
get_agent_draft.impact is a union on operation: { operation: 'CREATE', capital } for a create
draft, where it was null, and { operation: 'UPDATE', deployedPresetCount, openPositionCount, radarArmedCoinCount, capital, rebind } for an edit draft, where the counts sat at the top level. A
create draft's capital reads its trading configuration — drafted, or the platform's seed — over the
bound strategy's band, as the commit's capital check reads it.
get_strategy with includeInactive: true serves your own strategy in any state, and otherwise
the visible set the default read serves — every SYSTEM strategy among it — where it answered
NOT_FOUND for anything but an owned PRIVATE strategy. Another user's PRIVATE strategy is still
NOT_FOUND.
The stage and commit descriptions state the draft content cap (256,000 UTF-8 bytes) and its refusal, what a commit reaches — it closes no position, and the one refusal an open position can cause — and that a moved draft is shown to the player before any commit.
A stage's DRAFT_VERSION_MOVED fires only for a version the draft never reached — read from a draft
since committed or discarded. A stage naming an older version than the draft holds is composed over the
newer draft.
A strict client re-reads the get_agent_draft output schema; list the tools again after the server
deploys.
Breaking at v85: seven tools are retired, and a surviving tool refuses an input it accepted at v84. v85 is the contract on top of v84.
Every agent and strategy is now created and changed the same way: stage the change into the entity's draft, read the draft back with its diff, diagnostics and impact, and commit exactly the draft version and the live revision that read returned.
compile_strategy_plan, stage_strategy_plan, apply_strategy_plan, update_strategy_signal_rule,
create_intelligence_agent, update_intelligence_agent, rebind_intelligence_agent are gone, with
no alias; a call gets an unknown-tool error. The replacements: get_strategy_draft →
stage_strategy_draft → commit_strategy_draft for a strategy (a signal rule is a row of the draft's
SIGNAL_RULES axis), and stage_agent_draft → get_agent_draft → commit_agent_draft for an agent
(a rebind is the draft's STRATEGY_BINDING axis).
stage_strategy_draft { strategyId?, draftVersion, axes }, commit_strategy_draft and
commit_agent_draft ({ <id>, draftVersion, expectedRevision }, expectedRevision null for a
create). A commit publishes exactly the draft version and the live revision the caller read, is
idempotent on { id, draftVersion }, and returns the kind's committer response plus the draft's
draftVersion after its clear (null when the commit emptied it).
archive_strategy no longer takes confirm. A request carrying it is refused with the
unknown-key error — a break on a surviving tool.
discard_agent_draft and discard_strategy_draft take draftVersion (the version
get_<kind>_draft returned) in place of confirm; confirm is refused as an unknown key, and a
version the draft moved past is refused DRAFT_VERSION_MOVED.
preview_strategy_report takes { coinSelection, source }. The report fields move under
source: { kind: 'FIELDS', … }, and source: { kind: 'DRAFT', strategyId, draftVersion } previews the
owner's draft as a commit would compose it. A body with the fields at the top level is refused.
A strategy's ENTRY axis is a union discriminated by trigger: { trigger: 'ON_CANDLE_CLOSE' }
alone, or a level trigger with levelOffsetAtrMultiple and validForBars. ON_CANDLE_CLOSE naming
either dial is refused, which the retired compile accepted; the server stamps the seed values.
stage_agent_draft accepts STRATEGY_BINDING on an edit draft, where v84 refused it: a staged
binding commits as the agent's rebind. Its TRADING_CONFIG axis now describes each size preset as the
risk budget lost at the hard stop.
fork_strategy returns { strategyId, draftVersion } — the create draft it wrote, holding a copy
of the source — in place of { strategy }. No strategy exists until that draft commits.
get_agent_draft and get_strategy_draft gain draftVersion (0 when no draft is held, beside a
null draft), committedRevision, a per-axis diff against live, diagnostics and impact.
stage_agent_draft gains diagnostics, the composed draft's errors and warnings.
discard_agent_draft and discard_strategy_draft return discarded: true only; a discard that
finds nothing is refused rather than answered false.
Strategy draft axes gain ORIGIN, the create-only axis a fork writes, on list_strategy_drafts
and every strategy draft read. A client switching exhaustively on a strategy draft axis must add the
branch.
Error codes gain DRAFT_VERSION_MOVED (details draftVersion, null when no draft is held) and
DRAFT_AXIS_CONTESTED (details contestedAxes and draftVersion), both 409.
Refusal details gain nextAct. Every draft-specific refusal, the revision CONFLICT and the
in-flight commit refusal carry nextAct: 'get_draft'; a commit's validation refusal carries
nextAct: 'stage'; every other refusal carries none. details.expectedRevision widens to
number | null.
PLAN_APPROVAL_NOT_FOUND, with the plan path that raised it.
archive_strategy and restore_strategy no longer refuse around the owner's unsaved draft: they
re-base it onto the revision they write. restore_strategy's description drops the compile route — a
repair is staged into the draft and committed, which restores the strategy.
The author-strategy prompt reads "Discover, stage, review, and commit a BattleGrid strategy
through its draft".
The server's manifest records a description hash per tool (manifest format 4), and the tool count moves 142 → 138.
A strict client that rejects unknown keys fails to parse fork_strategy and the draft reads at v85;
list the tools again after the server deploys.
Breaking at v84: a condition body accepted at v83 is refused, and condition evidence is reshaped. v84 is the contract on top of v83.
How many completed bars a reading must hold is no longer one number on the condition. Each clause holds its own reading, counted on its own column's timeframe, and a group can hold its members' same-bar reading.
compile_strategy_plan, apply_strategy_planA condition no longer carries closes. A condition entry naming it is refused with the
unknown-key error, in a plan body and in every normalized post-state and strategy draft those tools
carry.
Every clause and every group requires hold: { atLeast, of }, whole numbers of at least 1.
{ "atLeast": 1, "of": 1 } is one read (what every v83 condition did), {n, n} is "n closes in a
row", {1, n} is "within n closes", and anything else is "at least m of n closes". A clause or group
without hold is refused. A plan staged under v83 must be recompiled.
What a header admits is refused by the builder, never by the schema. atLeast above of, a
window above the header's maximum, or a group hold over members that hold their own bars is a typed
remediation error.
preview_strategy_report, list_strategy_vocabularyclosesReadable: boolean is replaced by conditionHold: { timeframe, maxWindow, maxAtLeast, refusal } on each report condition column and each scalar metric. It names the timeframe a
clause counts on, the largest of and atLeast it admits (maxAtLeast is 1 for a cross or other
event, so an event can only be held "within n closes"), and why it admits one bar when it does:
NOT_REWINDABLE, DEVELOPING, OFFSET, BAR_STATE or HISTORY. Read the window from here
rather than assuming a ceiling; there is no longer a fixed 1–5 limit.
Every clause-evidence entry gains hold, counts and decidedBar, on preview_strategy_report,
list_gate_blocks, get_radar_close_decisions, get_agent_coin_qualification,
preview_radar_resolution and get_signal_log. counts is { trueCount, total, unresolvedCount }
over the clause's own bars. decidedBar is { timeframe, closesBack }, the bar its operand was read
on, or null when the hold that counts it reads one bar.
A condition outcome loses hold (closesHeld, closesRequired, liveOutcome, nextCloseAt):
the counted outcome is the clause's own, on its evidence. A decision record's condition
declaration loses closes.
Every strategy read returns hold on each clause and group and no closes on the condition.
The evidence union gains a group arm, { kind: 'group', op, hold, counts, decidedBar, outcome, memberCount }, emitted only for a group holding more than one bar, just before the memberCount
entries that belong to it. A client switching exhaustively on kind must add the branch. A strategy
without group holds serves no new arm.
A strict client that rejects unknown keys fails to parse condition evidence at v84; list the tools again after the server deploys.
Additive at v70.1: one output field added to the risk budget.
get_agent_budget, reset_agent_drawdown_baselinebudget.runStatus is your run-state for the agent: blocked, paused or active, already
ranked by the server. blocked means the admission gate is refusing the agent for a reason other
than its halt (blockedReason names it). A halted agent reads paused, even while its own
AGENT_HALTED block stands. Read it rather than re-deriving it from blockedReason and haltedAt.
A client that ignores the field is unaffected. A client that validates results against a cached, closed output schema should list the tools again after the server deploys.
Breaking at v83: four output fields renamed and retyped, one aggregation regrouped. v83 is the contract on top of v82.
Wherever a model's vendor is published, it is now the router vendor slug of the model's id — the part
before the / (anthropic, z-ai, moonshotai, …). It is never the BYOK provider enum, and never the
catalogue's display label, which sometimes named a hosting provider rather than the model's maker.
Every agent loses provider and gains modelVendorSlug: string, on list_intelligence_agents,
get_intelligence_agent, create_intelligence_agent, update_intelligence_agent,
rebind_intelligence_agent, archive_intelligence_agent and activate_intelligence_agent.
provider was the BYOK enum and was null for every agent; modelVendorSlug is never null. A
client reading provider reads nothing at v83.
list_approved_modelsEach model's provider is renamed vendorSlug, and its value changes: the vendor slug
(z-ai) where v82 served a display label that could name a host (StreamLake).
get_agent_explorerEach modelVendors[] row's provider / providerImageUrl are renamed vendorSlug /
vendorImageUrl.
Rows group by vendor slug, so every model one vendor makes shares a row whatever its display
label: three GLM labels that were three rows at v82 are one z-ai row at v83. Counts and sums add;
each mean is the merged total over the merged trade count.
ownerView LLM-call envelopeThe envelope's provider is renamed modelVendorSlug, on get_agent_journal, get_signal_log,
get_agent_game_history, get_user_agent_game_history, get_public_agent_signal_log_detail and
get_public_agent_game_history. It is null exactly when modelDisplayName is — a call whose model
has left the catalogue.
Breaking at v82: one output field removed. v82 is the contract on top of v81.
list_approved_modelspinProvider is gone from every model list_approved_models returns. A model is now served
only by an ordered list of verified hosts that the server keeps to itself, so a pin flag no longer
describes anything. There is no alias. A client that reads pinProvider reads nothing at v82, and
a strict client that requires it fails to parse the response.
Breaking at v81: one output field removed and one renamed. v81 is the contract on top of v80.
committedUsd is gone from the allocation that get_agent_fund_allocation,
halt_intelligence_agent, resume_intelligence_agent and set_agent_per_trade_push return. It
measured a custody ledger whose allocate and recall routes are retired. The figure that bounds an
entry is capital at risk, on get_agent_budget. A client reading committedUsd reads nothing at
v81.
get_agents_hubEach agent's envelope names the exposure ceiling maxConcurrentExposureUsd, the trading
config's own name for it, where v80 called it budgetUsd. The value is unchanged. A client reading
budgetUsd reads nothing at v81.
An agent entry refused because its Max exposure is fully committed is recorded as
INSUFFICIENT_BALANCE, where it was recorded as BELOW_EXCHANGE_MINIMUM with a $0.00 risk
budget. Its detail names Max exposure and the capital at risk. It reaches get_signal_log and
get_public_agent_signal_log_detail.
Breaking at v79: the agent budget's account fields move into one object.
get_agent_budget and reset_agent_drawdown_baseline replace accountEquityUsd,
budgetOverSubscribed and openUnrealizedPnlUsd with one account object carrying the same three
fields, or null when the server holds no cached reading of the account. At v78 an unread account
read 0, false and 0, which looked like a real empty account. A client reading the three
top-level fields reads nothing at v79.
The block reason gains EXPOSURE_BUDGET_EXHAUSTED (owner-private): an agent whose capital at
risk has reached its Max exposure is blocked once at the account stage, where v78 blocked it per
coin as EXCHANGE_MIN_NOTIONAL_UNREACHABLE. It reaches every output that carries a block reason:
list_gate_blocks, get_signal_log, get_public_agent_signal_log_detail, get_radar_activity,
get_radar_close_decisions, get_radar_deployment, list_radar_deployments,
preview_radar_resolution, get_agent_coin_qualification, get_trade_conversation and
propose_entry_decision.
accept_entry_decision refuses with CONFLICT while your account equity is below the agent's
Balance floor (balanceThresholdUsd) or cannot be read. The decision stays pending and can be
accepted once the equity is back at or above the floor.
Breaking at v78: get_trading_config_catalog serves a new agent's risk policy as one object.
tradingDefaults.defaults replaces ten per-field seed values — defaultMaxDailyTrades,
defaultMaxLeverage, defaultSmallPct, defaultMediumPct, defaultLargePct,
defaultEntrySlippageBps, defaultMaxConcurrentExposureUsd, defaultBalanceThresholdUsd,
defaultMaxCumulativeDrawdownUsd and defaultMaxDailyLossUsd — with one agentTradingConfig,
request-shaped, so it passes to create_intelligence_agent verbatim. A client reading the ten fields
reads nothing at v78.
tradingConfig.maxLeverage loses its published minimum: 1 on create_intelligence_agent and
update_intelligence_agent. Acceptance is unchanged: the platform's minimum leverage is the bound.
Both tools accept a strategy whose stored dials sit outside today's operator bounds. An agent write never re-judges a strategy dial.
A create_intelligence_agent call without tradingConfig is seeded from the same object the
catalog serves, never from column defaults, with Min order size from the platform minimum.
Breaking at v77: an agent's tradingConfig accepts less. create_intelligence_agent and
update_intelligence_agent refuse values v76 stored, and one loss stop moves to a different bound.
tradingConfigmaxLeverage must be a whole number. 1.5 was stored and executed at 1; it is now refused.
maxSlippageBps is at most 300, the entry slippage ceiling no fill could exceed anyway.
USD fields and the three size presets take at most 2 decimals.
maxDailyLossUsd must be at or below maxCumulativeDrawdownUsd.
A body accepted at v76 can be refused at v77 without one byte of it changing. Each refusal
names its field in a plain sentence, such as Whole numbers only. or
Can’t exceed the drawdown stop ($20.00).
maxCumulativeDrawdownUsd is bounded by the platform's drawdown maximum, not by
maxConcurrentExposureUsd. Max exposure caps margin open at once; the drawdown stop caps loss
over time. A $150 drawdown stop on a $100 max exposure is accepted.
get_trading_config_catalog gains maximumMaxCumulativeDrawdownUsd, the drawdown stop's
upper bound.
The execution-failure reason gains BELOW_AGENT_MIN_ORDER (owner-private) on get_signal_log
and get_public_agent_signal_log_detail: an order sized under the agent's own minAllocationUsd,
which was reported as BELOW_EXCHANGE_MINIMUM.
get_agent_budget's tradesToday counts entries filled since 00:00 UTC plus entries still
waiting to fill. Refused and expired entries no longer count, and the cap is enforced where an
entry is created. DAILY_LIMIT_RESERVED is no longer emitted.
A daily-loss halt whose drawdown stop is also breached becomes a drawdown halt when the day rolls over, rather than lifting.
A capital-feasibility refusal names what clears it: the maxConcurrentExposureUsd that would,
and, when one at or above the platform minimum would, a lower minAllocationUsd.
Breaking at v76: funding_extreme_positive and funding_extreme_negative take their threshold in
percent a year. Their parameter is thresholdAnnualizedPct, the unit of the strategy grammar's
ann column, where it was thresholdPct, a fraction of one hourly funding settlement.
thresholdPctBoth signals take thresholdAnnualizedPct, bounded 1–43,800 and defaulting to 100 — the
funding label's crypto extreme band, so 100 means 100% a year, and the negative signal fires
below -100%.
thresholdPct on either signal is refused as VALIDATION_ERROR wherever a rule's parameters
are authored: compile_strategy_plan, update_strategy_signal_rule, derive_strategy_rule_view
and a strategy draft's SIGNAL_RULES axis. oi_surge keeps its own thresholdPct.
A body accepted at v75 can be refused at v76 without one byte of it changing. A rate per hourly
settlement converts to percent a year by × 8,760 × 100: 0.0002 is 175.2.
A plan token compiled before the deploy fails as a stale plan at stage_strategy_plan and
apply_strategy_plan, because the authoring catalog digest covers each signal's parameters,
defaults and explanation: compile again. The strategy preview's vocabularyDigest moves too.
get_strategy_signal_definition and list_strategy_signals serve the new parameter, its
default and its explanation.
Each signal's indicator values gain funding_annualized_pct, the annualized rate the signal
compared, beside funding_rate.
Stored rules move to the new key after the deploy: a rule at the retired default takes 100,
and any other converts exactly. get_coin_signal_preview and every stored rule then score at the
new threshold.
No schema hash moves. Signal parameters are published as a JSON object on every input and
output, so the change is in the values accepted and served, and toolCount is unchanged.
Breaking at v75: the seven tools v73 narrowed refuse by the coin's market-data profile, and one output vocabulary gains a member. Which data a coin carries is now the profile the market-data service publishes for that coin, no longer the coin's asset class.
A condition reading a kind of data the coin's profile does not serve is refused as
INSTRUMENT, on the coin or on a benchmark section's own instrument: a spot read on a crypto
coin no spot venue lists is refused as one on a TradFi coin always was.
A coin the market-data service publishes no profile for refuses every condition that reads market data: a coin enabled since the service's last boot, a roster coin whose candle stream it did not subscribe, a ticker with no catalog row.
The tools, codes and details.context are v73's: preview_radar_resolution (every kind,
issuing no previewToken); commit_radar_deployment_draft, rebind_intelligence_agent,
compile_strategy_plan and restore_strategy (VALIDATION_ERROR with
CONDITION_UNREADABLE_BY_RADAR_SCAN); apply_strategy_plan, which applies only what its compile
checked; and propose_entry_decision (CONDITION_UNREADABLE_ON_COIN).
A body accepted at v74.1.1 can be refused at v75 without one byte of it changing. A write whose read of the published profiles fails is refused, never admitted unchecked.
FEED becomes INSTRUMENTA spot read for a ticker with no catalog row is INSTRUMENT, where it was FEED, and an
uncatalogued benchmark refuses every market-data read, where only its spot reads refused before.
The INSTRUMENT message names what the profile lacks: "<instrument>'s market-data profile
carries no <kind> data", or "the market-data service publishes no data profile for
<instrument>, so it carries no market data", where it read "<instrument> has no spot tape".
The readings v74 added report it as the sweep does: preview_radar_resolution's readings
and get_agent_coin_qualification's reading mark a condition on a kind the coin's profile lacks
INSTRUMENT.
BarFamily gains MARKET_STATSThe market-stats snapshot is a per-bar value: a decision reads the snapshot fetched inside the
1m bar ending at its close. A snapshot missing there is a per-bar shortfall, a families entry or
a MISSED close's INPUTS_OFF_BAR lacking entry naming MARKET_STATS at 1m with its
dueBarStart, ABSENT or GONE, and the decision waits or misses as for any per-bar family.
A decided close records no MARKET_STATS stale sample. A reading at the latest values still
does, and its boundMs is the coin's market-stats freshness target, never below 300,000 ms.
Five outputs publish the member: get_agent_coin_qualification, get_radar_close_decisions,
get_signal_log, list_gate_blocks and preview_radar_resolution. No input takes the vocabulary.
A client that validates results against a cached output schema should list the tools again after
the server deploys.
Not breaking: one answer corrected behind unchanged schemas.
A request on a pair your own radar policy decides with the same agent is answered from that
policy's decision. When the policy's fire enqueued no trade decision and no admission gate refused
it — a coin cooldown, the hourly fire cap, post-close suppression, a decision already running on the
coin, a deduplicated fire, or a fire aborted at its re-read — the conversation gains the card the
request's own unfired fire gets: close_claimed_by_radar when another agent's fire holds the coin,
otherwise close_refused with the reading's score against its minimum. It gained
close_fired_by_radar before, though no trade was taken.
Both cards are already in get_trade_conversation's vocabulary, so no served shape moves.
Not breaking: four output vocabularies gain a member, and propose_entry_decision refuses while an
entry pause is open. While bars missed for lack of market data span many coins, the platform pauses
every new entry the radar decides, a radar fire and a manual request alike, and the pause lifts by
itself after a quiet period.
propose_entry_decisionA request is refused after its own checks and before a watch is registered, whatever the agent
or coin: type: "error" with { origin: "ENGINE", reasonCode: "ENTRIES_PAUSED" }. The same
idempotencyKey replays the refusal for five minutes; ask again under a fresh key once the pause
lifts.
A request queued before the pause whose close qualifies while it is open gets the same ENGINE
error card in its conversation, and no decision is proposed. accept_entry_decision on a decision
already proposed, and every exit, run as without a pause.
TradeEvaluationAttemptReasonCode gains ENTRIES_PAUSED: the refusal's reasonCode,
list_gate_blocks' reasonCode, and the scanBlockReason of a refused fire's close record and
journal row.
TradingPipelineGateStage gains PLATFORM: a list_gate_blocks entry for each fire the pause
refused, with no coin, no detail and no decision record.
RadarFireDisposition gains EDGE_PRESERVED_ENTRIES_PAUSED: the refused fire keeps its edge,
so the pair fires at a later close only if that close qualifies on its own data. It is served on
get_radar_activity and get_radar_activity_summary, and as a close record's
outcome.fireDisposition beside outcome.scanBlockReason: "ENTRIES_PAUSED", with
outcome.outcome staying FIRED.
RadarIdleReason gains ENTRIES_PAUSED: the resolvesNow.reason of every SCANNING card on
get_radar_deployment and list_radar_deployments while a pause is open. The section stays
SCANNING, because closes are still decided and recorded.
When an admission gate refused your own policy's fire — an entry pause, an account block, a
position already held on the coin — the request's conversation gains that block's ENGINE error card
instead of close_fired_by_radar.
conditionReach replaced by readings)Breaking at v74: one required output field replaced on one tool, and an opt-in reading added to another. Both now serve what an agent reads on a coin, in the decision-record format v73.3 introduced.
preview_radar_resolutionconditionReach is gone, replaced by readings: one entry per on-duty agent, in on-duty order,
each naming agentId and agentDisplayName. A kind: "READ" entry carries the agent's reading,
the one get_agent_coin_qualification serves (below). A kind: "UNSCORABLE" entry, an agent whose
evaluation was rejected, carries coinDataStopped instead: the coin's stopped-data reading, or
null. One agent's rejection is its own entry, never a failed preview. The tool always asks for
readings, so readings is null only under a simulatedRegime.
blocksScanGate has no successor. Since v73 the preview refuses a slot agent with a condition
the radar scan cannot read on the coin, so no previewed pair holds a permanently blocking
condition. Whether the conditions hold the coin is the reading's own gate,
reading.live.reading.qualification.gates.requiredConditions.
Read reachReason from the reading. Each READ entry's
reading.live.reading.conditions.entries[] lists the conditions: a kind: "UNEVALUATED" entry
carries the condition's conditionKey, name and reachReason, and a kind: "EVALUATED" entry,
the counterpart of a null reachReason, carries its outcome with its clause values.
declaration.required marks the required ones, the set conditionReach covered.
A client that validates results against a cached output schema fails the call until it lists
the tools again, because the old schema required conditionReach. Reconnect, or re-list, after the
server deploys.
get_agent_coin_qualificationreading (boolean, default false) asks for the pair's reading, and each verdict gains the
required-nullable reading, null unless asked. A call without the flag is answered as before, with
reading: null.
reading.live is the reading the verdict was built from, with anchorBarStatus (live while
the anchor bar still forms), scoredBarStart (the bar the signals and gates were scored on) and
observedAt. It is what the agent reads now, not a decision: the radar decides at the close.
reading.lastClose is the deciding reading of the bar the agent's on-duty row names as its
last close decision: state: "RECORDED" with that bar's close-decision record,
state: "PENDING" with its barStart while that record is being written, or state: "NONE".
get_radar_close_decisions holds the decisions before it.
Purely additive: one read tool, one evaluation-attempt reason and four output fields added. What a decision read is now served: the radar's close decisions on a coin, the report behind a trade, and the compose decision behind a gate block.
get_radar_close_decisions (coinId; optional agentId, before, limit; mcp:read) — one
coin's radar close-decision records, newest decided bar first: every bar an agent decided —
FIRED, CLAIMED, NOT_QUALIFIED or MISSED — with the reading it was decided on (a miss's
answer instead), its fire disposition and its versions. Your own records only: a SYSTEM agent's
records on a coin belong to every user who deployed it. A page never splits a bar, so it can run
past limit; to page back, pass the oldest record's bar.barStart as before. toolCount goes
141 → 142.
TradeEvaluationAttemptReasonCode may read SCREENED_OUT on get_agent_budget,
reset_agent_drawdown_baseline, get_radar_activity, get_radar_deployment,
list_radar_deployments, preview_radar_resolution, get_trade_conversation,
propose_entry_decision, list_gate_blocks, get_signal_log and
get_public_agent_signal_log_detail: compose's deterministic pre-model screen found no routable
direction. That refusal now writes a gate block. It is public: the public profile carries it as is.
A client holding its own closed copy of the enum rejects the new member; one that renders unknown
reasons generically is unaffected.
get_signal_log's log gains reportSections (the report the model read, section by section,
in the shape of preview_strategy_report's renderedSections), promptDataSerialization (the
serialization it was rendered under) and missingData (the data compose lacked), each null for
a trade recorded before capture. get_public_agent_signal_log_detail carries none of them.
Each list_gate_blocks entry gains decisionRecord: the compose decision record written with
the block — its outcome, its evidence (the reading compose computed, the per-bar values a refused
decided bar lacked, or none), its missing data and its versions. It is null for a block written
before capture and for a block raised before compose.
One member added to the evaluation-attempt reason, published as output only, and two detail fields added to gate blocks. A confirming-close fire now names the bar it decided, and compose reads that bar or refuses with the new reason, instead of reading the newest completed bar at its own instant.
TradeEvaluationAttemptReasonCode may read DECIDED_BAR_UNAVAILABLE on get_agent_budget,
reset_agent_drawdown_baseline, get_radar_activity, get_radar_deployment,
list_radar_deployments, preview_radar_resolution, get_trade_conversation,
propose_entry_decision, list_gate_blocks, get_signal_log and
get_public_agent_signal_log_detail: compose refused a radar fire at the conditions stage, before
any LLM call, because a per-bar value its conditions gate reads was not on its due bar for the bar
the radar decided — its entry had not landed, or had gone. It is public: the public profile carries
it as is. A client holding its own closed copy of the enum rejects the new member; one that renders
unknown reasons generically is unaffected.
list_gate_blocks' reasonDetail gains decidedBarStart (ISO 8601, the decided bar's open)
and decidedBarTimeframe (the strategy timeframe), present on that reason alone.
Purely additive: thirteen read-only fields added to every list_strategies row. The input is
unchanged, and a client that ignores the fields is unaffected.
list_strategiesThe strategy card: regimeTimeframe and lowerTimeframe (the derived ladder rungs
get_strategy already serves); conditionTally, signalRuleTally and exitTally (armed out of
total, the four exit mechanisms); and minRequiredCount, entryTrigger, minRiskRewardRatio and
maxStopLossAtrMultiple.
forkedFromStrategyName, null when the strategy is not a fork or its source is neither yours
nor a SYSTEM strategy.
viewerBoundAgentCount and viewerOpenPositionCount, your own agents on the strategy and their
open positions. boundAgentCount still counts every user's agents; read the viewer fields for
yours.
viewerPerformance, your own all-time results on the strategy, null when you never traded it.
A trade counts toward the strategy recorded on the signal log that opened it, so rebinding an agent
never moves it.
Breaking at v73: seven tools refuse bodies they used to accept, behind unchanged input schemas. A condition the radar acts on — a required one, a direction-setting one (a verdict carrier), an exit one, and every condition they reference — must be readable by the radar scan on every coin the radar acts on for its agent: each coin a Radar policy slots it on, enabled or paused; each coin it holds an open position on, for its exit conditions; and each coin it has a pending manual request on.
VALIDATION_ERROR with the authoring code CONDITION_UNREADABLE_BY_RADAR_SCAN from
preview_radar_resolution (every kind, for every slot agent whether on duty now or not, issuing no
previewToken), commit_radar_deployment_draft, rebind_intelligence_agent, and
compile_strategy_plan and restore_strategy on behalf of the bound agents the radar acts on (a
strategy none of whose agents holds a slot, an open position or a pending request is never
checked). details.context.reachReason says why: INSTRUMENT (the coin, or a benchmark section's
own instrument, has no such data), AGENT_TIMEFRAME (the agent has no such rung) or FEED (the
radar scan never reads that data: crowd reads, a spot read for an uncatalogued ticker, a timeframe
the agent's plan does not load).
The message names the first conflicting agent, the condition, the column, the coin and the fixes: stop the radar acting on the condition, make it read a column the scan has on that coin, or free the coin — take the agent off that coin's policy, close the position, or cancel the pending request. It is a fact about the pair, which no later sweep changes.
A refused compile issues no plan token, so stage_strategy_plan cannot stage that content, and
apply_strategy_plan applies only a plan whose bound agents still hold the radar slot coins its
compile checked: a plan compiled before a deploy moved them fails verification. Compile again.
A condition reading a Session Field scalar keeps CONDITION_OPERAND_UNSERVED_IN_LANE, now
refused at the rebind and every strategy revision too, not only at the deploy.
stage_radar_deployment_draft is unchanged: a draft holds no slot, and the preview its commit
needs refuses. A body accepted under v72 is refused under v73 without one byte of it changing.
propose_entry_decision, get_trade_conversationpropose_entry_decision refuses a request whose agent the radar scan cannot read on the
requested coin, before its watch is registered, and releases its idempotency claim. It returns
type: "error" with error: { reasonCode: "CONDITION_UNREADABLE_ON_COIN", unreadable: { conditionKey, conditionName, column, coinTicker, reachReason } }: facts only, no prose. Choose
another coin, another agent, or a condition reading a column the radar reads on that coin, and
send it again. get_trade_conversation's error card carries the same reasonCode and
unreadable.
A client that validates results against a cached output schema fails a call that returns the new error until it lists the tools again, because the old schema has no such member. Reconnect, or re-list, after the server deploys.
CONFLICT with details.reason: "RADAR_DEPLOYMENT_MOVED" from apply_strategy_plan,
restore_strategy and rebind_intelligence_agent: the agents' radar slot coins moved after the
write's check. Re-read and retry, so the retry's check covers the moved deployment; for a plan,
compile again.
commit_radar_deployment_draft is a CONFLICT when a slot agent's strategy or timeframes
changed after its check. Preview again.
No schema change: one value reclassified on one public read.
get_public_agent_signal_log_detailpipeline.attempt.reasonCodes carries OWNER_PRIVATE, with its detail omitted, where it would
have carried EQUITY_CHECK_UNAVAILABLE, ALLOCATION_CHECK_UNAVAILABLE,
DAILY_COUNT_CHECK_UNAVAILABLE or APPROVAL_CHECK_UNAVAILABLE. Each says the account has an open
block, which the public profile withholds. This supersedes the v71.1 note below that the public
detail carries the two faults v71.1 added. OWNER_PRIVATE has been in that field's schema since
v71, so a client validating against a cached output schema accepts it. get_signal_log still
carries every precise code on your own agents.
userId removed)Breaking at v72: one output field removed from one public read.
get_public_agent_signal_log_detaillog.linkedEntryDecision.userId is gone. It was the user the agent acted for. On a SYSTEM
agent, which has no author and which any user may deploy, that named whoever deployed it, and the
public profile deliberately shows no owner for a SYSTEM agent. There is no replacement path on
the public read. A PRIVATE agent's author is ownerUserId on the agent's public profile, and the
owner tool get_signal_log still carries linkedEntryDecision.userId on your own agents.
A client that validates results against a cached output schema fails the call until it lists the tools again, because the old schema required the field. Reconnect, or re-list, after the server deploys.
Two members added to the evaluation-attempt reason, published as output only. When the server cannot read an agent's daily trade count or its trading account's approval, the admission is blocked with its own reason instead of borrowing the verdict it could not reach.
TradeEvaluationAttemptReasonCode may read DAILY_COUNT_CHECK_UNAVAILABLE or
APPROVAL_CHECK_UNAVAILABLE on get_agent_budget, reset_agent_drawdown_baseline,
get_radar_activity, get_radar_deployment, list_radar_deployments, preview_radar_resolution,
get_trade_conversation, propose_entry_decision, list_gate_blocks, get_signal_log and
get_public_agent_signal_log_detail. DAILY_TRADE_LIMIT_REACHED and AGENT_APPROVAL_EXPIRED still
mean what they say. A client holding its own closed copy of the enum rejects the new members; one
that renders unknown reasons generically is unaffected.
Breaking at v71: two public inputs narrow, and one public output widens.
get_public_agent_signal_logs, get_public_agent_realized_tradesfilter.rejectionReason refuses the six owner-private failure reasons —
BELOW_EXCHANGE_MINIMUM, INSUFFICIENT_BALANCE, INSUFFICIENT_MARGIN,
AGENT_APPROVAL_EXPIRED, SIGNING_KEY_UNAVAILABLE, AGENT_HALTED — with a validation error,
where they used to filter. These tools read another user's agent, and a private reason answered a
question about that user's budget, margin, approval, key or halt. No alias exists; filter by a
public reason. filter.expiryReason still accepts every member.
get_public_agent_signal_log_detailpipeline.attempt.reasonCodes, pipeline.execution.failureReason and .expiryReason may read
OWNER_PRIVATE in place of an owner-private reason, whose detail is then omitted. A client
holding its own closed copy of these enums rejects the new member; one that renders unknown
reasons generically is unaffected. A failed execution's executionMessage is now always null.
get_signal_log is unchanged: on your own agents it still carries every precise code, its
detail and the narrative.
One member added to the evaluation-attempt reason, published as output only. An admission the
server blocked because it could not read the account's allocation now carries
ALLOCATION_CHECK_UNAVAILABLE; NO_AGENT_ALLOCATION survives only on historical rows.
TradeEvaluationAttemptReasonCode may read ALLOCATION_CHECK_UNAVAILABLE on
get_agent_budget, reset_agent_drawdown_baseline, get_radar_activity, get_radar_deployment,
list_radar_deployments, preview_radar_resolution, get_trade_conversation,
propose_entry_decision and list_gate_blocks. A client holding its own closed copy of the enum
rejects the new member; one that renders unknown reasons generically is unaffected.
The reason detail no longer declares availableUsd or requiredUsd (get_signal_log,
get_public_agent_signal_log_detail). No response ever carried them.
configured removed)Breaking at v70: one output field removed from every risk-budget gauge.
get_agent_budget, reset_agent_drawdown_baselinebudget.gauges.<gauge>.configured is gone from all four gauges (dailyTrades, exposure,
drawdown, dailyLoss). It could only read true: the daily-trades cap is at least 1, and an
agent's exposure ceiling, drawdown stop and daily-loss stop are all required and strictly positive.
Every gauge has a limit, so read fill, remaining and breached and render every meter enabled.
A client that validates results against a cached output schema fails the call until it lists the tools again, because the old schema required the field. Reconnect, or re-list, after the server deploys.
Purely additive: one member added to the trade-failure reason, in both directions. An agent entry
the shared trading account cannot post the isolated margin for is now refused before any order is
sent, with its own reason. The same reason replaces EXCHANGE_ERROR when the exchange refuses an
order for margin first.
get_public_agent_signal_logs, get_public_agent_realized_tradesfilter.rejectionReason accepts INSUFFICIENT_MARGIN, where it used to refuse it.
get_signal_log, get_public_agent_signal_log_detailfailureReason may read INSUFFICIENT_MARGIN. A client that switches exhaustively on the
reason needs one more arm; one that renders unknown reasons generically is unaffected.
Purely additive: one member added to the trade-failure reason, in both directions. An agent entry whose hard stop no whole leverage can keep clear of liquidation is now refused before any order is sent, with its own reason. Below that, an agent order's reported effective leverage may sit under the agent's bound: it is lowered until the isolated liquidation is at least twice as far from the fill as the hard stop.
get_public_agent_signal_logs, get_public_agent_realized_tradesfilter.rejectionReason accepts LIQUIDATION_BUFFER_UNREACHABLE, where it used to refuse it.
get_signal_log, get_public_agent_signal_log_detailfailureReason may read LIQUIDATION_BUFFER_UNREACHABLE. A client that switches exhaustively on
the reason needs one more arm; one that renders unknown reasons generically is unaffected.
Breaking: one input removed from two tools, three outputs reshaped. A Radar policy carries no timeframe. Every rule whose conditions match puts its agent on duty at once, and each rule's regime condition is read at that agent's own regime timeframe. A coin still holds one position at a time: at most one fire per coin per pass, offered in rule priority.
preview_radar_resolution's SLOTS request refuses deploymentTimeframe as an unrecognized key.
Send { kind: "SLOTS", slots }.
stage_radar_deployment_draft refuses the DEPLOYMENT_TIMEFRAME axis the same way. Stage only
RULES and DEFAULT_SLOT.
resolvesNow on get_radar_deployment, list_radar_deployments and preview_radar_resolution is
the coin's state (section, isIdle, rotating, one reason, the open-position owner) plus
onDuty: one row per agent on duty, in priority order. Each row carries its slot, its regime reading
at its own regime timeframe, its qualification verdict and gate, cooldown, block, edgeSpent, last flip
and fire, and its own closeDecision. The single-winner fields (onDutyAgentId and its display pair,
matchedSlot*, regimeUsed, and the per-agent fields the rows now carry) are gone. Render every row in
the order served, and never pick one yourself.
preview_radar_resolution's conditionReach is one flat list, each entry naming its agentId and
agentDisplayName.
Each deployment slot gains agentRegimeTimeframe, the timeframe its rule's regime is read at.
deploymentTimeframe on a Radar policy, and the DEPLOYMENT_TIMEFRAME axis in the radar
draft's content and axesMeta on get_radar_deployment_draft, list_radar_deployment_drafts and the
stage result.
closeDecision.outcome gains CLAIMED: the agent's close qualified, but a higher-priority agent
took the coin's one fire that pass. Its edge is preserved, and it fires at a later close only if it
still qualifies there.
get_radar_activity and get_radar_activity_summary gain the events ON_DUTY_JOINED and
ON_DUTY_LEFT (duty is journaled per agent) and the fire disposition EDGE_PRESERVED_COIN_CLAIMED.
curveDigest describes the first agent on duty, named in curveAgentName.
get_trade_conversation may carry the close_claimed_by_radar card.
The exported battlegrid-radar-deployment skill describes the v69 flow.
Purely additive in schema, with one refusal behind unchanged ones. An Arena rule's regime condition is a SET in the draft, as it always was in the committed deployment: one rule can fire in several regimes, instead of one rule per regime.
stage_deployment_policy_draftA rule's regimes may name several regimes, where the draft capped it at one. The set is stored in
the selectable-regime order, so a staged [VOLATILE, BULL_EXPANSION] reads back
[BULL_EXPANSION, VOLATILE]. A regime named twice is refused.
get_deployment_policy_draftThe draft may return such a set. get_deployment_policy has served sets all along, so a client that
reads a committed deployment already handles them.
preview_deployment_resolution (SLOTS), preview_radar_resolution (SLOTS) and
test_generate_deployment_grid refuse a regime condition naming one regime twice, as
VALIDATION_ERROR at its regimes field, where they used to resolve or generate over it. The stage tools
already refused a repeat, so no commit changes; a repeated member is a malformed set.
Breaking: one tool retired, two inputs reshaped. Arena deployment content now commits only as a draft the player was shown a live preview of, as radar content has since v66. The player's deploy editor and every conversation share one unsaved draft per arena.
upsert_deployment_policy is retired, with no alias. Calling it is an unknown-tool error. To
deploy or change an arena's deployment:
stage_deployment_policy_draft the axes you are changing (RULES, DEFAULT_SLOT,
REGIME_ANCHOR), with the draftVersion your last read returned;
preview_deployment_resolution with request: { kind: "DRAFT", draftVersion } at the version
staging returned;
show the player the resolution and whether it will play (enabledAfterCommit), and on their word
call commit_deployment_policy_draft with the previewToken that preview returned and
confirm: true.
Pausing is no longer a flag on the write: use pause_deployment_policy. A first deployment always
plays; a replacement keeps its pause.
preview_deployment_resolution's request needs a kind. { kind: "SLOTS", slots, … } is the
old request. { kind: "DRAFT", draftVersion } previews the player's draft of the arena at that
version, composed over the deployed policy exactly as the commit writes it. { kind: "COMMITTED" }
previews the deployed policy a resume would arm. A request without kind is refused.
delete_deployment_policy requires expectedPolicyId beside expectedRevision: a revision
restarts at 1 when an arena is redeployed, so only the pair names the deployment you read.
A regime anchor override names its coin and timeframe together, or neither, on every path
including test_generate_deployment_grid and the SLOTS preview.
stage_deployment_policy_draft, get_deployment_policy_draft,
list_deployment_policy_drafts and discard_deployment_policy_draft (mcp:read). None of
them commits or arms anything.
commit_deployment_policy_draft (presetId, previewToken, confirm: true; mcp:wager) — the
only MCP committer of Arena content.
pause_deployment_policy (presetId; mcp:wager) — no revision and no certificate.
resume_deployment_policy (presetId, previewToken, confirm: true; mcp:wager) — needs the
certificate a live COMMITTED preview returned.
cancel_market_grid_submission (sessionId, confirm: true; mcp:wager) — the player's own
cancellation and refund of one entry.
preview_deployment_resolution nests the resolution under resolution and gains
previewToken (the certificate a live DRAFT or COMMITTED preview earns, null otherwise, good for
five minutes) and enabledAfterCommit.
pause_deployment_policy and delete_deployment_policy return openEntries: the entries the
player's agent already made in the arena's pending sessions, each with its lock time and entry fee.
They play out unless the player cancels them — tell the player and ask, then call
cancel_market_grid_submission only for the entries they pick.
get_deployment_policy's authoringContext gains the arena's header facts
(regimeReferenceTicker, presetBadgeImageUrl, entryFee, and playerCount — null when the arena
has no pending session to count).
The thought-log outcome gains SKIPPED_DEPLOYMENT_DISARMED: an entry job that found its
deployment withdrawn, paused or no longer slotting its agent stood down before paying.
The certificate is bound to what you previewed: your credential, the player, the arena, the
subject, the deployment and its revision, the draft version and the composed content. Anything that
moves is a CONFLICT and nothing is written. Preview again; never reuse the old certificate.
A commit never withdraws: a draft with no rule and no catch-all is refused, naming
delete_deployment_policy.
Resume is refused on a retired arena and on one the player's access was revoked from; pausing and withdrawing stay open.
Deployment refusals answer INVALID_DEPLOYMENT_POLICY.
delete_deployment_policy also ends the player's draft of the arena.
toolCount goes 134 → 141 (eight added, one retired). preview_deployment_resolution's input and
output, delete_deployment_policy's input and output, get_deployment_policy's and
list_deployment_policies' outputs, the three journal outputs and test_generate_deployment_grid's
input (a session start now accepts up to 288 times a day) move; upsert_deployment_policy's are
removed.
Breaking at v67: four output fields removed. v66.1 and v66.2 were additive and are recorded here with it.
get_agents_hub (v67, breaking)summary loses messagesUsedToday, dailyLimit, messagesUsedPercent and
avgCostPerMessageUsd. They reported the in-app arena chat's daily message quota, and that chat
is retired, so every value would have read zero. A strict client that reads them fails to parse the
summary. The rest of the summary is unchanged; per-agent spend stays on each row's cost24hUsd.
scan_coin_agents (coinTicker; mcp:read) — one coin evaluated against every one of the
caller's own agents, ranked in four arrays: qualified by score, rejected with the first failing
gate, unscorable, and ineligible with the code propose_entry_decision would refuse the agent
with (AGENT_NOT_ACTIVE, MODEL_INACTIVE, AGENT_HALTED). rank is global across the four.
Rate-limited per user: a refusal is RATE_LIMITED with retryAfterSeconds.
get_onboarding_requirements (no input; mcp:read) — the caller's readiness ladder: every
rung, the spine counts and the next rung.
get_strategy_draft (v66.1)draft gains lastSource — the surface of the draft's latest write (form, commander,
telegram or mcp), as the agent and radar draft reads already publish. Before v66.1 every read of
an existing strategy draft failed its output check with INTERNAL_ERROR; it now succeeds.
toolCount goes 132 → 134. The added tools' schemas are new; get_strategy_draft's and
get_agents_hub's output schemas move. Descriptions that named the app's retired Agent Toolbox trade
tab now name Commander.
Breaking: one tool retired and one input reshaped. Radar content now commits only as a draft the player was shown a live preview of. The player's radar builder and every conversation share one unsaved draft per coin.
upsert_radar_deployment is retired, with no alias. Calling it is an unknown-tool error. To
deploy or change a coin's Radar policy:
stage_radar_deployment_draft the axes you are changing;
preview_radar_resolution with request: { kind: "DRAFT", draftVersion } at the version staging
returned;
show the player the resolution, and on their word call commit_radar_deployment_draft with the
previewToken that preview returned and confirm: true.
Pausing is no longer a flag on the write: use pause_radar_deployment.
preview_radar_resolution's request needs a kind. { kind: "SLOTS", deploymentTimeframe, slots, simulatedRegime? } is the old request. { kind: "DRAFT", draftVersion } previews the
player's draft of the coin at that version, composed over the deployed policy exactly as the commit
writes it. { kind: "COMMITTED" } previews the deployed policy a resume would arm. A request
without kind is refused.
stage_radar_deployment_draft, get_radar_deployment_draft,
list_radar_deployment_drafts and discard_radar_deployment_draft (mcp:read). The draft
axes are DEPLOYMENT_TIMEFRAME, RULES (the complete ordered rule list, first = highest priority)
and DEFAULT_SLOT, each written whole. None of them commits or arms anything.
commit_radar_deployment_draft (coinId, previewToken, confirm: true; mcp:wager) — the
only MCP committer of radar content.
pause_radar_deployment (coinId; mcp:wager) — no revision and no certificate: a disarm never
waits on a read.
resume_radar_deployment (coinId, previewToken, confirm: true; mcp:wager) — needs the
certificate a live COMMITTED preview returned.
preview_radar_resolutionGains previewToken, the certificate a LIVE preview (no simulatedRegime) of DRAFT or
COMMITTED earns — null otherwise, and good for five minutes. It also gains
enabledAfterCommit: whether the policy trades once the previewed write lands. A paused
policy's draft stays paused.
The certificate is bound to what you previewed: your credential, the player, the coin, the
subject, the deployment and its revision, the draft version and the composed content. If the
player edits the draft, another commit lands, the coin is redeployed, or an agent in it is deleted,
the commit or resume is a CONFLICT and nothing is written. Preview again; never reuse the old
certificate.
commit_radar_deployment_draft takes only a draft certificate, and resume_radar_deployment
only a committed-policy one.
discard_radar_deployment_draft with confirm: true requires expectedVersion: the version the
unconfirmed call named and the player was shown. A draft that moved since is answered again, and
nothing is removed.
delete_radar_deployment also ends the player's radar draft of the coin.
toolCount goes 126 → 132. preview_radar_resolution's input and output schemas move, and
upsert_radar_deployment's are removed.
Breaking on one input, and four refusals behind unchanged schemas. Agent staging now names the draft version its proposal was read against, and three commits and one stage step aside where they would otherwise land around the player's open draft or on a strategy that moved.
stage_agent_draft requires draftVersion, an integer ≥ 0: the version the
get_agent_draft read you proposed against returned, or 0 when it returned { draft: null }.
A call that omits it, or sends a negative or fractional number, is refused at the boundary. On a
new create draft send 0, then reuse the version each accepted call returns.
A draftVersion above the draft's own is refused. A draft's version only grows, so no read
ever returned it: the draft you read was saved or discarded since. Read again — never raise the
number to get past the refusal.
The contested-axis refusal is measured from your draftVersion, not from a read the call
makes for itself, so an edit the player made between your read and your stage is refused by name
rather than silently overwritten.
update_strategy_signal_rule and restore_strategy answer CONFLICT while the player holds a
draft of that strategy. Compile the tune and stage it with stage_strategy_plan; for a restore,
ask the player to save or discard their draft first. A caller whose player holds no draft is
unaffected.
stage_strategy_plan refuses a plan the strategy committed past after the plan compiled,
with the same answer apply_strategy_plan gives: compile again against what is committed now.
Every refusal around an open draft names its version in details.draftVersion — the
committers above, the agent committers, and a stage refused for a version above the draft's own —
so you can read the draft the player holds and propose against it.
A draft axis's source names the door it came through. A Telegram Commander turn stamps
telegram and a connected client stamps mcp, where every staging call used to stamp
commander. Both values were already in the enum.
A draft's version keeps counting after a save or discard empties it. The next draft for the
same agent or strategy starts above the last version rather than at 1, so a version you read is
never reissued to a different draft.
toolCount stays 126. One input schema moves, stage_agent_draft's; no output schema moves.
Purely additive in schema, with one refusal behind an unchanged one. An agent's unsaved configuration now has a server home — the same owner-scoped draft strategies gained in v63.1 — so the agent form, Telegram and a connected client all see one set of unsaved values. These tools are how a conversation reaches it, and the two agent committers now step aside while it is open.
stage_agent_draft({ request: { agentId?, axes } }) writes proposed axes into the player's
draft for an agent and commits nothing: the values become the agent's configuration only when
the player saves. Unlike stage_strategy_plan it takes the axis values — IDENTITY,
BEHAVIOR, MODEL, TRADING_CONFIG, and for an agent not yet created STRATEGY_BINDING — each
written whole, so BEHAVIOR carries all three of risk, outlook and conviction and
TRADING_CONFIG the complete agent-owned configuration. Axes you omit keep their values.
Structure is checked on the way in; ranges are checked only when the player saves.
Omit agentId to open a create draft. The server mints the id and the response carries it;
name that id on every later call so one draft accumulates rather than a second opening beside it.
get_agent_draft({ request: { agentId } }) reads one draft, or answers { draft: null } —
not part-way through that agent is a value, not an error. baseRevision is null for an agent not
yet created, and baseMoved reports an agent committed past the revision the draft was written
against.
list_agent_drafts({ request: { cursor? } }) answers "what am I part-way through?" for a
conversation holding no agent id, newest first. Offer to continue one of these before starting a
second.
discard_agent_draft({ request: { agentId, confirm } }) destroys one draft on the player's
word. Called with confirm: false it destroys nothing and is refused with when the draft was last
written and which surface wrote it — tell the player that, then ask.
A staged proposal refuses the axes the player typed after your call read the draft. The refusal names the axes; read the draft again and propose against what they now have. Another surface's write to a different axis is retried once for you, then reported as a conflict.
STRATEGY_BINDING is refused on an agent that exists. Rebinding replaces an agent's
configuration and stays its own confirmed call, rebind_intelligence_agent.
update_intelligence_agent and rebind_intelligence_agent answer CONFLICT while the player
holds a draft for that agent, naming staging as the act available. Nothing about their input or
output changed, and a caller whose player holds no draft is unaffected — this is the one change
existing code can observe.
toolCount 122 → 126. No input or output schema of an existing tool moves.
Breaking, and it is a removal you will feel on two tools. A vocabulary that never depended on your draft was being re-serialized into every preview result; it now lives on discovery and the preview names it.
preview_strategy_reportconditionColumns covers only the sections your report RENDERS. It used to cover every
section the server's header inventory holds, which meant every report-level scalar on every
preview whether or not its module was placed — 36,652 bytes of payload, with each metric's gloss
repeated once per scope (11 market-breadth sentences published as 99). If you read
session-field, market-breadth or reference-pairs groups off a preview, they are gone.
Nothing stopped resolving. The header inventory is unchanged, so a condition naming
mktBreadth_crypto resolves exactly as it did. What changed is where you read its prose from.
vocabularyDigest is new — the sha256 identity of the authoring catalog that render described
its columns against, and the same digest a plan token binds as authoringCatalogDigest. Resolve
the vocabulary once, cache it against this string, and re-resolve when it moves.
structure.columns[].meaning is gone. It was a positional per-header string that neither
serialization ever rendered — the glossary paragraph in section.text is built from
spanFragments, and the structured arm always omitted meaning. spanFragments is unchanged.
budgetUsage.estimatedTokens now measures the whole served payload, not the rendered module
text inside it. Existing drafts read several times higher against the same cap. That is the fix,
not a regression: the byte cap is measured on the serialized result and cannot be reported inside
it, so this is the only meter that can warn you before a refusal — and it was counting about a
tenth of what actually refuses.
list_strategy_vocabularyscalarFamilies is where the vocabulary went. Every report-level scalar family with its
section, and one entry per METRIC carrying its gloss, legal condition operators, closed label
vocabulary, read contract (closesReadable / developingRead; v84 replaces closesReadable with
conditionHold) and the scopes it is measured at.
One gloss per metric with its scopes named against it, never one per pair.
Served whole under every category. A scalar describes the report, not a metric family, so it
is not filtered by the category you asked for. The ambient session family carries
moduleKey: null — it places no module and its operands are nameable all the same.
compile_strategy_planreviewContext.columns is gone. It recompiled a full column contract per authored custom
column — about 2,100 bytes for a five-header trajectory — describing columns the embedded
reportPreview.conditionColumns already describes per rendered header. Ask
get_strategy_column_contract when you want a column's exact normalized contract.
approvedPlan.creationSeed is gone. It was the dense 84-rule scorecard before the
overrides, published beside a postState.signalRules that is the same list after them and an
explicitRuleOverrides naming exactly what differs. Apply never saw it either.
The section array and a custom section's column array lose their maxItems: 64. That bound
restated a configured cap of 32 as a looser 64 that never refused anything. budgets.sections and
budgets.sectionColumns from discovery are the published values, and the server enforces them
before it reads any market data.
Call list_strategy_vocabulary once for the scalar vocabulary, key your cache on the preview's
vocabularyDigest, and drop any code that reads scalar groups off conditionColumns,
structure.columns[].meaning, reviewContext.columns or approvedPlan.creationSeed. A preview's
size now tracks the report and cohort you composed rather than the size of the platform's catalog.
Purely additive again, and smaller: one optional input field. Drafts now cover a strategy that does not exist yet, so a playbook you author in chat survives between compiles instead of starting over each time.
compile_strategy_plan's CREATE arm gains an optional strategyId, naming the create draft
this compile continues. Omit it and the server mints an id exactly as it always has, so nothing
you send today breaks.
Send it to accumulate; omit it to start fresh. A first CREATE names nothing and mints; stage
that plan and the draft opens at the minted id, which list_strategy_drafts and the plan itself
both report. Name that id on the next CREATE and the second plan carries the same identity, so the
one draft accumulates rather than a second opening beside it.
WHY YOU HAVE TO SAY IT. The server will not choose for you. A compile knows only who you are — it has no conversation id — and you may hold several create drafts, so picking one could write this playbook's values over another's. A compile that names none mints a third id rather than guessing.
It names a draft; it does not choose an id. It is accepted only as the identity of a create draft you own. An id naming no such draft is refused, as is one naming a draft for a strategy that already exists.
stage_strategy_plan admits a CREATE plan, opening the create draft at the plan's own id when
none exists yet. Its input is still { planToken } alone, and the contested-axis refusal applies
identically.
list_strategy_drafts rows gain baseRevision. It is null for a create draft — a strategy
that does not exist has no committed revision to be based on — and that null is the only thing that
distinguishes one. There is no kind, isNew or status field beside it.
A create draft reserves nothing. It consumes no quota and holds no name, so both are decided when the strategy is actually created. Holding drafts past your limit is legal; creating past it is not.
axes.create gains strategyId. No domain gains a value, and toolCount stays 122.
Purely additive: four new tools, nothing you send today changes. A strategy now has an
owner-scoped draft — the unsaved authored values a player is part-way through — and these are how
an agent reads it, writes a compiled plan into it, and destroys one. The canonical record for every
contract move is docs/architecture/MCP_CONTRACT_HISTORY.md in battlegrid-app; the served version
is what the handshake announces.
stage_strategy_plan({ request: { planToken } }) writes a compiled plan's own changed axes
into the owner's draft. It commits nothing and does not spend the plan, so the same token is
still directly applicable afterwards. planToken is the only member the request accepts.
get_strategy_draft({ request: { strategyId } }) reads one draft, or answers { draft: null }
— "you are not part-way through this one" is a value, not an error. It reports baseMoved when the
strategy has been committed past the revision the draft was written against.
list_strategy_drafts({ request: { cursor? } }) answers "what am I part-way through?" for a
conversation holding no id. It includes a draft whose strategy you can no longer see, marked
strategyExists: false, because that is exactly the work a player has lost track of.
discard_strategy_draft({ request: { strategyId, confirm: true } }) destroys one draft. The
confirmation is required and unsaved values have no other copy; the strategy, its revisions and
your other drafts are untouched.
A staged plan refuses the axes you typed after it compiled. stage_strategy_plan answers
CONFLICT naming contestedAxes, the version the plan compiled over and the draft's own version
per axis. Compile again so the plan absorbs those edits, then stage. A previous staging on the
same axis is not a contest — only the owner's own hand is.
An apply refuses a plan whose draft has moved since. Moved, discarded and already-committed-by-a-sibling are one answer with one recovery: compile again.
toolCount 118 → 122. No input or output schema of an existing tool moves — the draft version a
plan was compiled over rides inside the opaque plan token, which no schema declares.
One part to read first: the per-condition evidence clock is gone, and what replaced it is not a
rename. Which bar a condition reads is now decided by the surface asking — a decision reads completed
strategy bars, a display read shows the forming one — and by each candle column's own Confirmed /
Developing selector. closes survives and changes meaning. The canonical record for every contract
move is docs/architecture/MCP_CONTRACT_HISTORY.md in battlegrid-app; the served version is what
the handshake announces.
A condition entry carrying clock is REFUSED (61.0.0, read-higher-timeframes-per-column).
The authoring schemas are .strict(), so compile_strategy_plan, apply_strategy_plan,
fork_strategy and the HTTP save alike fail the body with the unknown-key error. There is no
replacement key to send: the decision instant belongs to the caller, not to the condition. Two
input hashes move, compile_strategy_plan's and preview_strategy_report's — the two tools that
accept a condition entry.
closes stays mandatory and means something new. It is now held for N completed strategy
bars (1–5). Above 1 it is legal only over a header a completed bar actually moves, at or above
the strategy timeframe, never over a developing read, and only where the condition carries a clause
of its own — a referenced condition resolves once and contributes the same answer to every bar, so
a hold reached only through a reference would count reads that never happened.
ConditionOutcome.closeClock becomes hold, and it is NON-NULL for every condition. A
one-close condition reads 0 or 1 of 1 rather than serving an absence, so a client no longer
branches on whether the reading exists. The count is taken from completed strategy bars whatever
basis the surface evaluated on.
ReportConditionColumnDTO.closeClockReadable becomes closesReadable, beside a new
developingRead. The first answers whether a condition addressing that header may hold more than
one completed bar; the second states whether the header reads the bar still in progress at the
decision instant. Both are server-supplied — derive neither. Eight output hashes move, covering
every tool that serves a strategy's conditions or a report's addressable columns.
The bars selector is declared on eleven candle-series transforms, not four. value,
classifyZone, classifyState, distance, spread, crossDetect and bandTouch join
trajectory, aggregate, efficiency and maxShare. It carries no default value, because
the default is a rule rather than a constant: Confirmed ("closed") above the strategy timeframe,
and at or below it the series as the frame carries it. The resolved answer is served per column on
effectiveParameters.bars — null where discovery has no anchor to resolve a rung against.
A higher-timeframe level is measured from the current strategy-bar price. Distances and the candle label classifiers compare against the frame's current price rather than the higher-timeframe bar's own close, which could be a full higher-timeframe bar stale. A distance chained into a series keeps every slot on its own bar's close.
domains.conditionClock is removed; domains.columnBars: ["all", "closed"] takes its place, and
axes.condition loses clock. toolCount stays 117.
One release train, four numbers, and two parts to read first: propose_entry_decision no longer
decides in the call, and a closed trade's tradeStatus now follows its net P&L. v55 through v59 are not written up here; the canonical record for
every contract move is docs/architecture/MCP_CONTRACT_HISTORY.md in battlegrid-app, and the
served version is what the handshake announces.
The entry axis names no bar but the strategy's own, and three of its keys are gone (60.0.0,
decide-entry-on-strategy-close). The authoring schemas are .strict(), so compile_strategy_plan,
apply_strategy_plan, fork_strategy and restore_strategy REFUSE a body carrying
entry.confirmTf (the deciding bar is the strategy's own timeframe, so a required input whose only
legal value was another field of the same strategy is absent rather than mirrored), entry.closes
and entry.bandAtrMultiple (a multi-bar hold is declared on the condition that needs it; the
displacement band is replaced by the platform's own entry-deviation gate, measured against the
decided close), and the whole exit object (the open-position exit lane judges one closed candle
of the position's strategy timeframe). Exactly one input hash moves, compile_strategy_plan's.
entry.trigger: AT_SIGNAL is retired for authoring (60.0.0, retire-at-signal-trigger, riding
the same number). Refused on every authoring surface and on a RESTORE, which rebuilds a stored
revision through the same value object. The enum member stays READABLE on every strategy read and
on a fired decision's provenance, so a pre-retirement revision is still legible — it just cannot be
re-authored. The three that remain are ON_CANDLE_CLOSE, STOP_THROUGH_LEVEL and ON_RETEST.
tradeStatus follows NET P&L on every closed trade (60.3.0, label-trade-outcome-by-pnl).
No schema hash moves for this and the values you receive change anyway. It used to map the close
REASON onto a verdict — every TAKE_PROFIT was WON, every STOP_LOSS was LOST, and a
MARKET_CLOSE of either sign was the neutral CLOSED — so a stop that filled after a break-even
reprice was reported as a loss and a take-profit eaten by fees as a win. LIQUIDATED still
outranks the number, because a force-close is not a verdict about the trade; CLOSED now means
only that there is no outcome row to judge. A client that counted WON rows was counting
take-profits.
A proposal is QUEUED, not decided (60.2.0, queue-manual-entry-for-close). This is the entry in
this section to act on. propose_entry_decision registers a request against the agent's next
strategy-bar close and returns immediately: no model runs, nothing is spent, and the call carries
type: "queued" with request — requestId, the bar (barStart), when the answer is due
(decidesBy) and the last instant that bar may still be decided (windowEndsAt). The answer
arrives later, in the agent's conversation and, when it proposes a trade, in
list_pending_approvals. A close that does not qualify, a window that passes with no sweep, and a
bar the agent's own radar deployment decided first are each recorded in the conversation instead.
recommendation and no_trade REMAIN in the union: the idempotency registrar replays results
recorded before this release for their TTL, so a client that dropped those members would fail on
its own retry. A client written against 60.2 handles queued and error; one that must also
replay handles all four.
The four retired entry keys leave every strategy read (60.0.0) — get_strategy,
list_strategies, fork_strategy and both plan envelopes — and confirmTimeframesByMainCandle
leaves list_strategy_vocabulary, because there is no confirm set left to publish.
entryDiscipline.closes and .bandAtrMultiple go number → number | null (60.0.0) on
get_trade_outcome_by_decision and list_trade_outcomes. Null on every decision fired after this
release, which authors neither; a non-null pair dates the row to the arming era.
TradeConvErrorCode gains REQUEST_PENDING (60.2.0), on the SURFACE arm of
propose_entry_decision's error: one pending request per user and coin, so a second is refused. A
coin already carrying a pending or live position is refused before anything is queued, as an
ENGINE-origin OPEN_POSITION_CONFLICT — a member that was already published, reaching this surface
for the first time.
The exit names the leg that filled (60.3.0, label-trade-outcome-by-pnl). TradeOutcomeDTO
and the pipeline outcome summary gain exitRepriceSource: the reprice that placed the protection
leg which actually closed the position — BREAK_EVEN, TRAILING, TIME_DECAY,
MANUAL_OVERRIDE, UPDATED — or null on every close no protection leg filled and on a leg that
was never repriced. It is the mechanism behind the verdict above, so read it beside closeReason
before calling a stopped-out trade a failure. The enum is the one get_position_audit_history
already publishes. Four output hashes move; no input schema does.
Two tools join the catalog for the request lifecycle (60.2.0, queue-manual-entry-for-close).
get_entry_request (read scope) reads a request that is still pending and is NOT_FOUND once it
has been answered, cancelled or expired — the answer is in the conversation, not there.
cancel_entry_request (mcp:wager) withdraws one before its bar is decided. toolCount 115 → 117.
Three radar reads publish the close decision (60.1.0, add-radar-close-decision-state).
get_radar_deployment, list_radar_deployments and preview_radar_resolution carry one further
key on resolvesNow: closeDecision, non-null whenever an agent is on duty for the pair. It
carries the deciding timeframe, the nextCloseAt instant, a state of WAITING or DEFERRED
(a bar has closed and no sweep has decided it yet), and last — the bar, the instant, the outcome
(FIRED / NOT_QUALIFIED / MISSED), the gate and the closed reading's score against its
minimum — or null before the pair's first close decision.
Read it first on a close-deciding pair. The sibling qualified and qualificationBlock fields
are the per-minute DISPLAY reading there and decide nothing, so an agent ranking them reports
"qualified — watching for a setup" about a pair whose last three closes were each refused. Neither
field is removed or reshaped. A NOT_QUALIFIED outcome with a NULL gate beside a score at or above
the minimum is the consumed edge — the close qualified and the baseline was already spent — stated
by the server so no client compares the two numbers.
Eleven majors reached authors while this section stopped at v36. That gap is the mechanism, not an oversight: since v31 a contract move needs no release here, so nothing forced a note to be written — and the documentation ships inside the tarball, so a note written but unpublished reaches nobody. Both halves are now closed by a rule keyed to the served contract rather than to a release of this package.
50.0.0 and 51.0.0 arrived late, and the reason is worth naming. The re-vendoring errand that
used to carry these notes is now a generated export
(battlegrid-app/server/scripts/export-mcp-skills.mjs), and it owns three paths — skills/,
skills/EXPORT.json, and the vendored digest. It deliberately does not touch this file. So the
digest kept arriving on time while the note stopped travelling with it, and this section sat at
v49.5 against a served contract of 51.0.0. Nothing a client could observe was wrong; what was
missing was the sentence telling them so. A contract move still needs a human-authored entry
here, and the export lane will not remind you.
Nothing to migrate. This is the one direction that cannot break a client: a body the server used to refuse is now stored. Listed because a client that special-cased the refusal can delete that branch.
A signal rule patch no longer forces you to restate allocation and required (51.0.0,
fix-signal-rule-patch-semantics). On update_strategy_signal_rule, and on the rules element
of compile_strategy_plan, both fields become optional and join params under ONE omission
rule: an omitted mutable field preserves the stored value for that signal. "Raise this
signal's weight" is now expressible.
{ "strategyId": "…", "expectedRevision": 7, "signalId": "volume_surge",
"allocation": 3 } // `required` and `params` keep exactly what is storedEvery existing client keeps working — a complete payload is still a valid patch — so this is
listed for what you can now STOP sending. Before it, both fields were mandatory on every rule
surface, so a caller that had not first read the current rule had to invent a value it was never
asked about. That is not hypothetical: revision 5 of a production strategy flipped required
false → true unasked while moving a weight 2 → 3, turning a scoring signal into a mandatory
gate — which changes whether the agent takes trades at all.
One boundary on the newly legal ground, and it breaks nothing: a patch carrying no mutable field is refused — "A rule patch must change something: supply at least one of allocation, required or params." — rather than minting a no-op revision. Under 50.0.0 that request could not be formed at all, so nothing that used to work is now refused.
An arming trigger no longer constrains its required conditions' clock (49.4.0,
restore-arming-trigger-authoring). compile_strategy_plan, apply_strategy_plan and
fork_strategy accept a strategy whose entry trigger is ON_CANDLE_CLOSE, STOP_THROUGH_LEVEL
or ON_RETEST while a required condition reads the LIVE clock. v48.1 announced that pairing
as rejected, naming the offending condition key on VALIDATION_ERROR; that refusal is gone.
Why it was withdrawn, since the reasoning matters more than the rule: it ran against the whole
assembled strategy, so it refused every edit to a strategy in that shape — a rename, one report
column, one signal weight — plus restore and fork. For a strategy whose required conditions read
columns that can never carry a CLOSE clock (zone distances, perp/spot flow), there was no legal
shape to move to at all. And its premise — that the pairing can never fire — was measured before
the arming lifecycle was corrected, and no longer holds.
The underlying question, should an entry that waits for a close be decided on a forming bar, is now settled where the decision is made rather than by refusing the author's declaration.
blocksScanGate reads true for a class it did not (50.0.0, own-scan-served-set-once), on
preview_radar_resolution. Blocking is now derived from the lane's served set rather than
switched over the reach reason, so a FEED-reason refusal on an operand no reader in the lane
serves BLOCKS instead of deferring. No field changes shape, and a client that already renders the
key renders the new answer — but a client that treated blocksScanGate: false as "this will
resolve once data arrives" now sees a deployment that will not fire. Nothing in the payload tells
you this moved.
Three tools serve different values for identical input (47.3.0, derive-scan-fetch-from-report).
The radar scan leg now derives its timeframe fetch from the strategy's report rather than the
on-duty agent's three perception rungs, so a required condition addressing an absolute timeframe
outside those rungs — never evaluated at scan before — now is. No field moves; the values do:
Tool | What moves |
|
|
|
|
|
|
Read the last one carefully: a client treating UNMEASURABLE as "this gate is switched off"
will now see that gate BLOCK. AGENT_TIMEFRAME keeps its member and narrows to the one cause
no fetch can discharge. Nothing in the payload tells you this moved.
A condition naming a swingHi / swingLo header is refused — the indicator is a Donchian
channel (53.0.0, rename-donchian-channel). The rolling-window extremes indicator computed the
highest high and lowest low of the trailing 20 closed bars — a Donchian channel — under a swing
point's name, and the name asserted a property the value does not have (a swing high survives being
broken; a channel edge re-anchors the instant it is). Every layer of the vocabulary moves at once,
with no alias: a clause on compile_strategy_plan, apply_strategy_plan or fork_strategy naming
a header on the old stems — dist_swingLo, dist_swingHi_4h, dist_swingLo_rank_near, any
timeframe- or rank-suffixed form — is refused as CONDITION_COLUMN_UNKNOWN where 52.0.0 accepted
it. The same shapes exist on the donchianHi / donchianLo stems, which
get_strategy_column_contract lists with the labels 20-bar high / 20-bar low. Every value is
the same number under its new name.
Rename the stems (swingHi → donchianHi, swingLo → donchianLo, suffixes unchanged) in
every condition and Market Read marker you author. That is the whole migration for what you send.
entry.levelSource is refused — the level is derived, never authored (52.0.0,
derive-entry-level). The strict entry object on compile_strategy_plan, apply_strategy_plan
and fork_strategy is six keys — trigger, confirmTf, closes, bandAtrMultiple,
levelOffsetAtrMultiple, validForBars — and a body carrying levelSource is refused naming
the key where 51.0.0 accepted it. STOP_THROUGH_LEVEL rests a stop past the swing channel's
CURRENT edge in the trade's direction (the 20-bar high for a long, the 20-bar low for a short);
ON_RETEST rests a limit in front of the edge a close most recently BROKE. Your two dials are the
unsigned distance from that edge (levelOffsetAtrMultiple, 0–2 ATR; a long adds, a short
subtracts) and the bar validity (validForBars, 1–24). A meaningful validForBars under
AT_SIGNAL or ON_CANDLE_CLOSE is now refused as PARAMETER_NOT_HONOURED like the other level
dials.
Drop the key. That is the whole migration for what you send; the six-key example below is the current shape.
update_strategy_signal_rule requires confirm: true when the strategy has bound agents
(51.0.0, fix-signal-rule-patch-semantics). The write re-materializes scoring configuration onto
every bound agent immediately — including agents holding open USDC positions — and until now
nothing on the server asked. A rule edit on a strategy with one or more bound agents is refused
without the flag, and the message names the count. An edit on a strategy with nothing bound is
unaffected, and so is every path through the web editor.
Send confirm: true. That is the whole migration, and confirm is published on the input
schema — but nothing in the schema says WHEN it becomes mandatory, because the condition is the
bound-agent count rather than the shape of your body. The refusal rides the existing
VALIDATION_ERROR code, so a client that omits it discovers the rule at the refusal.
Why it moved to the server: the guard existed, but only as served prose the calling model could
decline — and did, twice in production on 2026-08-24. archive_strategy and
rebind_intelligence_agent have taken a server-enforced confirm all along; single-rule tuning
was the outlier among its own siblings, and it is the one that writes to scoring.
A radar deployment is refused when its strategy reads a session-field scalar (50.0.0,
own-scan-served-set-once). upsert_radar_deployment refuses a deployment whose slot agents'
bound strategy carries a condition reading one of five SESSION-FIELD scalars — fieldPlayers,
fieldUpBias, fieldBiasDir, captConc, picksSpread — with
CONDITION_OPERAND_UNSERVED_IN_LANE. A body accepted under 49.5.0 is refused under 50.0.0
without one byte of it changing.
Why a refusal and not a warning. Those five describe a game SESSION, and radar runs outside a session at BOTH its stages — so such a condition can never resolve there. The deployment formed no fire edge and the agent did nothing on that coin, silently, forever. The refusal converts a permanent silence into an error at the moment you author it.
Migrate by moving the clause to a scalar radar reads — the Market Breadth or Reference Pairs
families, which are market-wide reads with no session dimension — or by binding the strategy to
an arena agent instead. The error carries both halves: allowedDomain enumerates every servable
header, and the message names the sections.
Strategy authoring is untouched by this bump. The same strategy is legal, and reads those
scalars correctly, on an arena agent — which is why the refusal is on the DEPLOYMENT and not on
compile_strategy_plan / apply_strategy_plan.
A benchmark-bound section no longer accepts crowd metrics or rank transforms (49.0.0,
fix-benchmark-legality-save-path). On a custom section carrying a non-null benchmarkTicker, a
column whose metric is enrichment-stage (the CROWD_* family, FLOW_ALIGN, SMART_RETAIL,
CAPTAIN_CONF, CONFIDENCE, SETTLED_AT, the PERP_SPOT_* trio) is refused with
REPORT_COLUMN_BENCHMARK_METRIC_UNSUPPORTED, and one carrying a rank transform in either the
direct or the chained position with REPORT_COLUMN_BENCHMARK_TRANSFORM_UNSUPPORTED. Every tool
accepting a section array is affected, and no other byte of your body changes.
Fix it by moving the column, not by retrying. Both readings are defined relative to the
cohort being evaluated — a crowd reading is what this session's players did, a rank is a position
among the coins under evaluation — and a benchmark is deliberately outside that cohort. Neither
has a value there, which is why such a column could never render. Put it on an ordinary section
(benchmarkTicker: null), or drop it.
This is a fix, not a new rule. The restriction shipped with benchmark sections and was already
enforced by the column builder, by report materialization, and by get_strategy_column_contract —
so an author who checked a column against discovery first has been seeing this refusal all along.
What changed is that the SAVE path now asks the same question. Previously it did not, so such a
section persisted and then failed at every evaluation instead of at authoring.
A custom report section no longer accepts timeframe (48.0.0,
remove-section-anchor-override). The per-section anchor override is gone. Every tool accepting a
section array — compile_strategy_plan, preview_strategy_report, derive_strategy_rule_view,
and apply_strategy_plan — refuses a section carrying it: sections[N]: Unrecognized key(s),
with no other byte of your body changing. Drop the key. A section's columns resolve against the
strategy timeframe, and a column reaches any other timeframe by pinning it on the column
(timeframe: { abs: '4h' }) — which it could always do. Relative column references
(anchor/lower/regime) are untouched, and are the point: they track the strategy.
A custom report section requires notes (43.0.0, add-authored-section-notes). Every tool
accepting a section array — compile_strategy_plan, preview_strategy_report,
derive_strategy_rule_view — refuses a section without it: sections[N].notes: Required, with no
byte of your body changing. Send explicit null for "no note". It is required rather than optional
because these payloads are a FULL REPLACE: an omitted key and an explicit null would be the same
request on the wire, so every rebuild site would silently clear a note its author wrote.
benchmarkTicker carries the same required-nullable discipline for the same reason.
Every condition requires clock and closes (44.0.0, add-condition-clock), and exit
arrives with them. A condition entry now carries eight keys, not five. clock is "LIVE" (the
previous behaviour — the forming bar) or "CLOSE" (settled bars); closes is how many consecutive
closed bars must read true, 1–5, and is always 1 under LIVE. No wire default, for the same
whole-set-replacement reason as notes: a defaulted key would let an unrelated re-save silently
un-clock an enforced money gate back to forming-bar evidence.
A CLOSE clock is accepted only where a closed frame can change the reading — the clause must
resolve from the coin's own candle series at offset 0. Frame-inert operands (perp-payload scalars,
published rolling changes, ranks, zone entities, regime labels, enrichment metrics, session
scalars) are refused with CONDITION_CLOCK_OPERAND_ILLEGAL naming the header and the remedy: split
that clause into its own LIVE condition and conditionRef it. exit: true is legal only under
clock: "CLOSE" — an exit fired on a forming bar is an intrabar exit.
A strategy requires a six-key entry object (44.0.0 for four keys, 47.0.0 for three more —
add-entry-on-close, add-level-trigger-execution — and 52.0.0 removed levelSource,
derive-entry-level). Required on every CREATE, on compile_strategy_plan, apply_strategy_plan
and update_intelligence_agent. A client sending 44.0.0's four-key object is refused with
entry.levelOffsetAtrMultiple: Required without one byte of it changing.
"entry": {
"trigger": "AT_SIGNAL", // | ON_CANDLE_CLOSE | STOP_THROUGH_LEVEL | ON_RETEST
"confirmTf": "4h", // the strategy timeframe or the rung below it — nothing else
"closes": 1, // 1–5; must be 1 unless ON_CANDLE_CLOSE
"bandAtrMultiple": 1.0, // > 0, and <= the platform's entry-deviation gate
"levelOffsetAtrMultiple": 0, // 0–2, UNSIGNED distance from the derived edge; 0 unless a level trigger
"validForBars": 4 // 1–24 of the strategy's own bars; 4 unless a level trigger
}AT_SIGNAL with those values is byte-identical to pre-44 behaviour. The legality matrix runs one
way: all six keys are always present, so a MEANINGFUL value under a trigger that ignores it is
refused rather than accepted-and-dropped — a dial never silently does nothing.
A strategyTimeframe the platform does not ingest is refused (39.0.0,
move-renderer-to-rendered-section) on get_coin_market_context, where it was previously
accepted. The number is the only signal a client gets.
eventType gains ENTRY_EXPIRED_UNCONFIRMED (48.1.0, fix-entry-arming-lifecycle) on
get_radar_activity and get_radar_activity_summary — an armed entry that reached its episode
lifetime without ever receiving a confirming close. A client holding its own closed copy of that
enum rejects the new member; one that switches exhaustively on it needs the branch.
entryVoidCause is deliberately unchanged and still carries exactly BAND, CONDITIONS,
STRUCTURAL. Each of those is a measurement taken at a close, so an episode that reached no
close gets its own event type rather than a fourth cause — ENTRY_VOIDED means "called off at the
close", a claim this outcome must not make.
The authoring boundary also gains a refusal with no schema change: a strategy declaring an
arming trigger whose required conditions read the LIVE clock is rejected on the existing
VALIDATION_ERROR code, naming the offending condition key. Same code, same shape, new reason —
so nothing in the published schema tells you it can now happen.
SWING_LOW / SWING_HIGH leave the stop and take-profit method enums, and the four S/R
indicator keys leave the signal vocabulary (53.0.0, rename-donchian-channel). DONCHIAN_LOWER
/ DONCHIAN_UPPER and donchian_upper / donchian_lower / prev_donchian_upper /
prev_donchian_lower carry the same values; nothing answers to the old names.
get_coin_market_context is REMOVED (40.0.0, retire-get-coin-market-context). Calling it
returns an unknown-tool error. There is deliberately no alias: a silent redirect would hide a
payload shape change from a client that never asked for one. Use get_market_context.
get_macd_heatmap leaves the published surface (41.0.0). Same shape of break, same absence of
an alias.
isPrimary is removed from every published EvaluatedSignal (38.0.0) — get_signal_log,
get_public_agent_signal_log_detail, and every other tool returning a signal scorecard. Reading it
now finds the key absent rather than false.
scan_agent_coins returns the ranking, not an explanation of every coin (54.0.0,
fix-mcp-scan-row-altitude). The scan used to wrap the app's full per-coin qualification verdict
in every row — two directions with candidate-level construction and a stop-loss policy band, four
gates each with its own measurement, condition reach reasons, the ATR corridor. For a 78-coin
catalog that was 82,147 characters, ~1,053 per row, which is past the tool-result cap of every
client we know of: the calling model received a file path instead of an answer, so the tool did not
deliver its result even when the scan succeeded.
rows is gone. Three ranked arrays replace it, and rank is global across all three, so
reading them in this order reproduces the server's own sequence:
qualified[] { rank, coinTicker, scorePercent, coinDataStopped }
rejected[] { rank, coinTicker, scorePercent, firstFailReason, scoreShortfallPercent }
unscorable[] { rank, coinTicker, coinDataStopped }Array membership now carries what the row kind discriminator and the qualifies flag used to,
and both are gone with them; firstFailReason is non-nullable on a rejected row, because a
non-qualifying verdict always names the gate that blocked it. Rows also lose long, short,
gates, tradeableAtrRange, evaluatedAt, coinName, assetClass and category.
Two things are new. The agent's agentId, agentName, strategyTimeframe and minScorePercent
move to an agent object carried once instead of on all 78 rows — null when the scan scored
nothing at all. And scoreShortfallPercent is server-computed: how far below the minimum the score
fell, non-null exactly when the aggregate score is what blocked, so you never subtract a published
threshold from a published reading yourself.
scanStartedAt, coinsScanned and qualifiedCount are unchanged, and so is every coin: same
rows, same order, same ranks, same verdicts, same rate buckets, same evaluator. Only the fields
moved — about 8.8 KB for the same 78-coin scan.
Migration. Read qualified / rejected / unscorable instead of rows, and take the agent's
thresholds from agent rather than from the first row. For a shortlisted coin's full per-direction
and per-gate detail, call get_agent_coin_qualification on up to 12 tickers — it carries every
dropped field and costs no second scan (the scan is rate-limited to 3 per agent and 10 per user a
minute; the probe is not).
Report headers, glosses and signal indicator keys are renamed for the Donchian channel
(53.0.0, rename-donchian-channel). Every report surface — preview_strategy_report,
get_strategy_section_template, the agent prompt previews — renders donchianHi / donchianLo
and their dist_… / …_rank_near forms where it rendered swingHi / swingLo, with the labels
20-bar high / 20-bar low and glosses that say what the number is (the highest high / lowest low
of the last 20 closed bars — the channel's edges). Signal definitions
(get_strategy_signal_definition, list_strategy_signals) and signal-log indicatorValues
carry donchian_upper / donchian_lower / prev_donchian_upper / prev_donchian_lower for the
four S/R signals. Signal ids, the SUPPORT_RESISTANCE module and its display names are unchanged.
Read the new keys. A reader keyed on swing_high / swing_low finds nothing; the values are
the same numbers under the new keys.
The stored entry discipline no longer names a level source (52.0.0, derive-entry-level).
StrategyDTO.entry (get_strategy, list_strategies, the fork_strategy / archive_strategy /
restore_strategy envelopes) and the apply envelope's postState.entry lose levelSource.
TradeOutcomeDTO.entryDiscipline (get_trade_outcome_by_decision, list_trade_outcomes) loses
it and gains levelOffsetAtrMultiple: number | null — the offset in force at the fire, null on
rows written before its column existed. A reader that rendered the level source renders the
geometry from trigger and the outcome's direction instead: which edge and which order shape
are a pure function of those two fields, so no stored copy is served.
update_strategy_signal_rule gains its own response envelope (51.0.0,
fix-signal-rule-patch-semantics). It no longer shares { strategy } with its siblings. The
response is { strategy, ruleChanges }, where ruleChanges is the server's own before/after
pair for the edited signal — [{ signalId, before, after }], each side a full rule object. It is
null when the mutation changed no rule, never [].
Report the change from that pair, not from memory. The planner always computed the diff and the tool discarded it, so a caller narrating what it just did had only its own recollection of the before-value. One production edit shipped a wrong receipt on top of a wrong write that way, and the write was unreconstructable from the audit trail afterwards.
Additive, but published on a .strict() shape — a decoder pinned to the old two-key object
rejects the new key. fork_strategy, archive_strategy and restore_strategy keep the shared
StrategyResponseSchema and publish exactly what they did; it was deliberately NOT widened for
them, so this reshape reaches one tool only.
An entry void now names the gate that refused it (49.5.0,
fix-arming-trigger-clock-authority), on get_radar_activity_summary. In the cause rollup, the
ENTRY_VOID group's gateCode widens from always-null to QualificationGateCode | null: a
conditions-side void carries the gate that blocked — AGGREGATE_BELOW_MIN,
REQUIRED_COUNT_BELOW_MIN, REQUIRED_CONDITION_FALSE — while a band void stays null, because
that void happens on a reading that qualified and has no failing gate to name.
A client that renders the field through the same enum the response already uses on four other
cause arms needs no change. One that treated it as a literal null — a strict decoder pinning the
type, or a branch keyed to its absence — sees a value it did not expect. That is the whole
migration.
Why it moved: the group previously collapsed every conditions-side void under one label. The first 26 in production carried that label while two different gates had produced them, and none of them was a required condition being false. The rollup ships counts rather than rows, so the gate could not be recovered client-side.
The normalized report section loses timeframe (48.0.0, remove-section-anchor-override), on
every tool that publishes a strategy: get_strategy, fork_strategy, archive_strategy,
restore_strategy, update_strategy_signal_rule, apply_strategy_plan, and
compile_strategy_plan's post-state. A strict parser rejects the shorter object; a lenient one
reads a section whose anchor is the strategy timeframe, which it now always is.
get_strategy_column_contract renames its anchor, both ways (48.0.0). The request field
sectionTimeframe becomes anchorTimeframe — same meaning, and still optional. On the response,
timeframe.requiresSectionTimeframe becomes requiresAnchorTimeframe, and
timeframe.sectionTimeframeOverrideAllowed is removed: it published whether a column could go
in an anchor-overridden section, and no section can be overridden. One call also stops being
refused — a timeframe-inert metric supplied with an anchor now compiles.
REPORT_COLUMN_SECTION_TIMEFRAME_UNSUPPORTED leaves the authoringCode vocabulary.
RenderedSection.notes stops carrying provenance (42.0.0, separate-section-facts-from-read),
on preview_strategy_report, compile_strategy_plan and get_market_context. A new REQUIRED
provenance: string[] carries it instead. A client reading provenance out of notes now reads an
author's prose, or nothing — which is a silent misread, not an error.
preview_strategy_report.renderedSections[] gains authoredNote (43.0.0), the author's read
for a custom row and null on a platform row. Additive, but published on a .strict() shape.
get_radar_activity serves its evaluation curve on the FIRST PAGE ONLY (37.0.0), and
get_radar_activity_summary is added. A client reading the curve off a later page finds it absent.
The stop and take-profit method enums gain DONCHIAN_LOWER / DONCHIAN_UPPER (53.0.0,
rename-donchian-channel), replacing SWING_LOW / SWING_HIGH on the signal-pipeline detail
schemas — trade-setup options, R:R-rejected pairs and candidate levels. A copy of either enum that
rejects unknown members must add the two new ones; the two old ones never appear again.
TradeExecutionFailureReason gains LEVEL_NOT_RESTABLE (52.0.0, derive-entry-level): a
level entry refused at placement because its resting price sat on the wrong side of the exchange
mid — a buy stop at or below it, a buy limit at or above it, and the mirror for a sell. It appears
on the list_signal_logs failure-reason filter input and on the signal-pipeline execution
summary's failureReason, with origin CLIENT_GATE. There is no distance limit: a level far from
the mark rests until its bar validity expires.
The authorable metric vocabulary widens by 29 keys (46.1.0, add-indicator-catalog-coverage):
Keltner (KC_UPPER/KC_MID/KC_LOWER), Supertrend (ST_LINE/ST_DIR), Hull (HMA20),
WaveTrend (WT1/WT2), QQE (QQE_RSI_MA/QQE_STOP), Parabolic SAR (PSAR), Ichimoku
(ICHI_CONV/ICHI_BASE/ICHI_SPAN_A/ICHI_SPAN_B/ICHI_LAG), Williams %R (WILLR14),
Stochastic RSI (STOCH_RSI14), session pivots (PIVOT_P/PIVOT_R1–R3/PIVOT_S1–S3), plus
four already-published fields that became addressable: BB_UPPER, BB_LOWER, DI_PLUS,
DI_MINUS.
Additive on the wire — every request you can send today is still accepted. It is called out here because a client holding its own closed copy of the metric enum rejects the new members, and because an agent holding a cached belief that these are inexpressible will substitute for a primitive the platform now serves. That failure raises no error at all: the author is simply told a strategy cannot be built.
EntryTrigger gains STOP_THROUGH_LEVEL and ON_RETEST, and EntryLevelSource is published
for the first time with four members (47.0.0). Additive on their own; the required keys above are
what make that bump a major.
Each signal-checklist item's measured object gains a required triggered boolean (47.2.0,
fix-entry-prompt-signal-evidence) on get_signal_log and get_public_agent_signal_log_detail,
on the numeric and categorical arms. Additive and MINOR — a client ignoring it is unaffected —
but it is named here because a client parsing that output strictly rejects the new key, and because
leaving 47.2 unlisted would make a reader wonder what happened to it.
The unavailable arm deliberately does not carry triggered: "the claim could not be joined to
a stored result" and "the signal did not fire" are different states, and leaving the key off that
arm makes conflating them a type error rather than a convention.
Two tools join the catalog for the agent trade flow (52.1.0, add-mcp-agent-trade-flow).
scan_agent_coins evaluates every active coin against one of your agents in a single call and
returns them server-ranked, on the same use case, buckets and ranking the app's own TRADE-tab scan
serves. propose_entry_decision runs the conversational trade turn headlessly for one (agent,
coin) and returns its terminal in the stream's own vocabulary: type: recommendation carrying the
PROPOSED TradingEntryDecisionDTO row that get_entry_decision and list_pending_approvals
already publish, type: no_trade, or type: error carrying the turn's TradeConvError verbatim.
Nothing narrows and no existing schema hash moves; the 12-ticker get_agent_coin_qualification
stays as the spot-check probe.
idempotencyKey is REQUIRED on propose_entry_decision — it is the turn's own key, and a
same-key retry replays the recorded terminal rather than paying for a second inference. That
includes a post-billing LLM_FAILURE, which is returned as a value for precisely that reason.
Pre-engine faults are typed errors instead: RATE_LIMITED, NOT_FOUND, CONFLICT for a same-key
call still in flight, and SERVICE_UNAVAILABLE when the surface is switched off. Both tools
refuse with RATE_LIMITED carrying retryAfterSeconds under the same per-(user, agent) limit the
app itself enforces, so a scan is never served stale or partial.
get_regime_snapshot publishes the evidence behind the verdict (47.1.0,
publish-regime-classification-evidence). The snapshot gains evidence: the quantities the
classifier read, the gates it tested them against, the signed margin to the gate deciding whether
the current label survives, and the two decision facts only the classifier holds — gateState
(cleared / held / dropped) and directionSource (di / ema). Nothing narrows; a client
that ignores the field is unaffected.
Read gateState before you trust a trend label: held means the ADX hysteresis buffer is
carrying the PREVIOUS bar's label rather than this bar re-confirming it — a materially weaker
claim wearing the same word, and one no client could previously detect. directionSource: 'ema'
is the same shape of warning: the direction came from the fallback that fires precisely when the
DI spread is indecisive. The margin is signed so positive always means "the current label
survives by this much", in every gate state, so it is safe to branch on its sign.
conviction is a BRANCH DISCRIMINATOR, not a confidence: it encodes which rule in the priority
ladder matched, not how comfortably it matched. The margins carry comfort. A client reading
conviction as a strength score is reading it wrong, and always was — this release just makes the
alternative available.
Thirteen metric keys join the catalog (47.1.0) — the regime family gains REGIME_STATE,
REGIME_CONVICTION, REGIME_RUN_BARS, REGIME_TREND_GATE, REGIME_TREND_MARGIN,
REGIME_TREND_SOURCE, REGIME_DI_SPREAD, REGIME_VOL_ATR_RATIO, REGIME_VOL_BBW_RATIO,
REGIME_MOM_BULL_VOTES, REGIME_MOM_BEAR_VOTES, REGIME_CRASH_MARGIN and REGIME_CRASH_LATCH,
making the composite regime and its evidence addressable in a report column or condition for the
first time. Only a client that switches exhaustively on MetricKey needs new branches.
Not a contract change, but worth knowing if you author conditions: the report grammar's regime metrics now resolve from the confirmed close on every path. They previously resolved from the forming bar when a report was rendered and the confirmed close when the scan swept, so the same condition could read differently in preview than in production. Same wire shape; same bar everywhere now.
The number in this heading is a CONTRACT version, not this package's version. The npm badge at the top tracks the proxy's own code; this section tracks the server's wire contract. They move independently by design: a contract move needs no release here, because a connected proxy relays the contract out of the upstream handshake rather than declaring it. So a package on
31.xlisting contract history up to36.xis correct — not a version someone forgot to bump. Read the live pair from the startup stderr lines orGET /mcp/version; see Rediscovery & versioning for why.
These are the server contract breaks between contract 12 and contract 36. Most of the span shipped while the package sat at 11.0.0; contract 31 landed after this package reached 31.0.0, and the two numbers matching is coincidence — since v31 the announced contract is relayed from the server, so a package version says nothing about a contract version. They are contract history, not package releases: from v31 the announced contract is relayed live and a contract move is no longer a release here. Grouped by what a client observes, with the contract version that introduced each.
The proxy itself is unchanged. It embeds no schemas, pins no contract version, and forwards { request } verbatim. Every break below lands on whatever authors the payload or reads the result, never on the proxy.
compile_strategy_plan is no longer read-only or idempotent (33.0.0, rehydrate-approved-plan-on-apply). Its input, its output and its behaviour toward your strategy are unchanged — it still changes no strategy, agent or revision — but it now parks the plan it approved for its own apply to read, and each call mints a distinct record and a distinct token. readOnlyHint and idempotentHint are published as false accordingly. A client that retried a compile that had already succeeded, or fanned two out in parallel for one edit, was doing so on the strength of the old annotation: do neither. Compile once per reviewed payload. Nothing in the payload tells you this moved.
Position-size presets are now a RISK BUDGET (30.0.0, split-stop-geometry-from-risk). smallPct / mediumPct / largePct stop denoting a share of the ORDER (notional = pct / 100 × headroom × leverage) and start denoting the share of headroom placed at risk (notional = headroom × riskPct / stopDistancePct, capped at the margin headroom can post). Same keys, same types, same accepted range: nothing in the payload tells you the meaning moved. A client still sending 22.0 for MEDIUM is asking to risk 22% of its budget on one trade rather than roughly 2%. Typical risk budgets are 0.5–3; the platform defaults moved to 1 / 2 / 3. Leverage stops multiplying order size and becomes a constraint only.
upsert_deployment_policy requires enabled (35.0.0, add-arena-deployment-pause). The arena deployment gains an owner-owned pause, and the flag that carries it is required, not optional — a body accepted under contract 34 is refused under 35 without one byte of it changing. Required is the whole point: this call replaces the entire policy, so an omitted key and an explicit true would be the same request on the wire, and every client that rebuilt a policy without the flag would silently resume a deployment its owner had paused. Read the value from get_deployment_policy and send it back. There is no separate pause verb — pausing and resuming are this same call with the flag flipped.
Do not reach for enabled: false to un-deploy. It keeps every slot and stops play, which is the opposite of withdrawing: delete_deployment_policy is the withdrawal verb, and it discards the rules permanently. upsert_deployment_policy refuses an empty slot set, and its rejection names both routes.
A signal rule flagged required at allocation Off is rejected (34.0.0, enforce-required-allocation-invariant). required and allocation are two independently editable fields encoding one thing — whether and how a signal participates — so { required: true, allocation: 0 } was representable and meant nothing: the scorecard's triggered set already excludes Off, so such a rule could neither satisfy minRequiredCount nor block a trade. It is now refused on every rule-writing surface — compile_strategy_plan, apply_strategy_plan and update_strategy_signal_rule alike.
Three things make this one easy to trip over. It is an input-acceptance narrowing: a payload accepted under contract 33 is refused under 34 without one byte of it changing. It is invisible in the published schema — zod-to-json-schema drops effects by construction — so you cannot pre-validate it from tools/list, and the typed error IS the contract: read details.inertRequiredSignalIds, which carries the complete sorted list of offending signals, rather than the message, which names a bounded prefix. And nothing is repaired for you: the server will not raise the allocation (that would invent a scoring weight you never chose) nor clear the flag (that would discard your intent silently). Pick one and resend; either satisfies the boundary.
apply_strategy_plan no longer accepts the plan (33.0.0, rehydrate-approved-plan-on-apply). Its input narrows to { planToken, confirm }. A plan member is rejected as an unknown key — not accepted, not ignored, and with no transitional dual shape — so every client that built the payload breaks on the next connection, which before this is what every published surface told it to do. The server keeps the plan its own compile approved and reads it back, so you copy nothing out of the compile response: forward planToken byte-for-byte and confirm.
Three consequences worth knowing. The 256,000-byte cap on the apply payload is gone with the payload, so a large authored surface no longer becomes impossible to apply through a conversational client; the compiled plan is still capped and compile still enforces it. PLAN_APPROVAL_NOT_FOUND joins the error vocabulary for a token no approved plan answers to — already applied, lapsed, or never issued, all one code, because the recovery is the same in each case: compile again. And a validation refusal (a quota, a name collision, a bound agent that changed, a moved catalog) now leaves the approved plan applicable — clear the cause and confirm again with the same token while it lives, rather than recompiling.
The agent brain is no longer a preset union (32.0.0, remove-agent-presets). create_intelligence_agent stops taking the brain discriminated union: modelId and behavior become required top-level fields and the { kind: 'PRESET' | 'CUSTOM' } wrapper is gone, so the old shape fails on the unknown key and on two now-missing required fields. update_intelligence_agent drops brainPreset; modelId and behavior stay independently optional and are now always honoured, closing an accept-and-ignore where a named preset silently discarded a model sent beside it.
apply_strategy_plan now publishes the same bounds as compile_strategy_plan (31.0.0, resolving #4495). Eleven position-management dials were declared three times server-side and two copies had drifted, so apply advertised trailingGivebackPct as a bare number where compile advertised 25–55, and dropped trailingTriggerR's multipleOf 0.01 — the constraint pinning storage precision so a sub-precision value is rejected rather than rounded onto the trail-from-entry sentinel 0. Nothing bad could ever commit (the digest would not match), but the refusal you got was a binding mismatch, which arrived as a bare INTERNAL_ERROR. Apply's published bounds only narrow to compile's; a client that copies values from approvedPlan.postState, as it should, is unaffected. Separately the trio minStopLossAtrMultiple / maxStopLossAtrMultiple / minRiskRewardRatio is published as a bare declaration by both tools, its real bounds being runtime-tunable and inexpressible in JSON Schema.
The stop-loss ceiling changed unit (30.0.0). maxStopLossPct (a percent of entry) becomes maxStopLossAtrMultiple (a multiple of ATR), and the accepted range narrows from (0, 100] to (0, 3]. It is a rename and a re-denomination — mapping the old value onto the new key sends a number one to two orders of magnitude too large. The objects are .strict(), so a 29.x client sending maxStopLossPct is rejected with an unknown-key error. A new cross-field rule comes with it: minStopLossAtrMultiple < maxStopLossAtrMultiple is now a real comparison and is enforced.
The grid-confidence and trade-conviction bars left the agent (28.0.0, remove-agent-rule-defaults). tradingConfig.gridMinConfidence and minTradeConviction are removed from the shared .strict() config; a bar is declared on the arena slot or radar slot that fires.
The agent no longer carries either entry guard (26.0.0). create_agent and update_agent stop
accepting tradingConfig.signalTimeoutMinutes and tradingConfig.maxEntryDeviationAtrMultiple.
The schema is .strict(), so a client still sending either is rejected, not silently ignored.
Neither has a replacement key on any surface: unlike the strategy-owned fields below, these have no
owning surface at all beneath the platform. One platform_config value governs the entry-price
drift budget for every decision, read at evaluation time; one governs how long an entry may stay
unfilled, snapshotted onto the position at creation. Remove both keys and send nothing in their
place.
The arena stopped granting trade authority (25.0.0, remove-arena-trade-permissions). upsert_deployment_policy and preview_deployment_resolution stop accepting tradingEnabled, minConviction and coinRules[] on a slot.
The agent no longer carries an exit policy (24.0.0). create_agent and update_agent stop
accepting tradingConfig.positionManagement. The schema is .strict(), so a client still sending
it is rejected, not silently ignored. The twelve dials that decide how a stop MOVES after entry
— break-even arming, trailing engagement and giveback, time-decay tightening — are denominated in
the setup's own payoff shape (multiples of the trade's initial risk, fractions of take-profit
distance, minutes since entry) and read no balance, leverage or exposure, so they belong to the
thesis rather than to the account running it. Author them on the strategy instead, through
compile_strategy_plan / apply_strategy_plan: the post-state gains the same twelve keys beside
the trade-level trio, and the plan diff gains a positionManagement axis. An agent inherits the
policy from the strategy it binds. The umbrella enabled flag is deleted rather than moved —
each mechanism's toggle is now the whole truth for that mechanism, so "trailing on, management
off" is no longer expressible, a state the server's own monitor and boot recovery already
disagreed about.
The agent-level trading mode is gone (23.0.0). create_agent and update_agent stop
accepting tradingConfig.tradingMode. The schema is .strict(), so a client still sending it is
rejected, not silently ignored. Trading on/off is now scoped per deployment — a radar policy's
enabled, an arena slot's tradingEnabled, a per-coin tradeEnabled — and approval-before-
execution is the conversational surface's own contract, so the account-level switch that sat above
both is removed rather than renamed. To stop an agent trading, turn its deployment off (or halt
the agent); to make one trade autonomously, arm a radar coin or switch trading on for an arena
slot. A newly authored arena slot now starts with trading off.
Trailing gained a required threshold (22.0.0, add-trailing-trigger-r). positionManagement gains trailingTriggerR as REQUIRED (0–2.0, 0.01 precision, 0 = trail from entry); the object is .strict() all-required, so sending positionManagement without it is rejected.
The strategy regime timeframe became derived (19.0.0, remove-strategy-regime-override). It stops being an authored axis anywhere on the contract and is served read-only.
Conditions gained a required required (16.0.0, add-condition-enforcement-gate). A condition entry omitting the boolean is REJECTED rather than defaulted.
The trade-level policy moved to the strategy (15.0.0, move-trade-level-policy-to-strategy). It leaves the agent authoring surface and joins the setup gates on the strategy.
The agent's ATR timeframe axis is gone (14.0.0, remove-agent-atr-timeframe-axis). ATR is sampled on the strategy timeframe, always.
Radar's wall-clock condition changed shape (12.0.0, unify-deployment-hours-as-sets).
The signal scorecard stops serializing its entries three times over (36.0.0, mcp-signal-log-contents). get_signal_log and get_public_agent_signal_log_detail drop scorecard.triggeredSignals, scorecard.primarySignals and scorecard.supportingSignals. Every one held the same entry objects allEvaluatedSignals already carried, so each is one filter over flags every entry still publishes:
Removed | Read instead |
|
|
|
|
|
|
Keep the triggered half of those last two predicates. Both collections were triggered-only by construction, so filtering on isPrimary alone surfaces signals that never fired — a silent widening, not an error. No field is removed from an entry: the key set on an allEvaluatedSignals member is unchanged, every evaluated signal is still returned whether or not it triggered, and details prose and indicatorValues are intact. There is no opt-in to get the three back and no default filter. Breaking only if you read one of the three names; on an 84-signal / 16-triggered log the duplication was 10,682 bytes, 28% of the scorecard, carrying no information.
The fleet roll-up on list_deployment_policies drops unconfigured and gains paused (35.0.0, add-arena-deployment-pause, fix-arena-deployment-undeploy). unconfigured counted a deployment holding zero slots — a state that can no longer exist, because a stored policy now carries at least one slot and the withdrawn state is the absence of a policy rather than an empty one. The bucket was constant 0 at the moment of removal, so no number you read was wrong; a client reading the key still breaks on it, which is why this is a break and not a cleanup. paused is the owner's own switch, counted separately from retired — an administrator disabling the arena — because conflating them tells an owner to wait for something that will not happen. Every policy lands in exactly one bucket, so the buckets sum to arenas: worth asserting if you reconcile these counts.
AdminApprovedModelDTO.isActive became lifecycle (32.0.0, remove-agent-presets) — AVAILABLE / DEPRECATED / RETIRED. The boolean conflated "offered in the picker" with "bound agents may run", so there was no way to stop offering a model without hard-blocking every agent already on it. A client switching on isActive must switch on lifecycle, and must not treat DEPRECATED as blocked: that is the state which keeps bound agents running. The agent read DTO drops brainPreset in the same move — the marker recorded which named bundle an owner clicked, never a value the runtime read, and the model and soul it stamped are unchanged on every agent.
approvedPlan is one object, not an operation union (31.0.0, resolving #4495). It was published as a discriminated union whose discriminator does not survive JSON-Schema conversion, so what actually shipped was a bare anyOf: validating a failing compile response gave you every arm's errors with empty instance paths, and the top one typically complained that an UPDATE was missing creationSeed — a CREATE-only key — while the field that really failed went unnamed. It is now one object with a literal operation discriminator, and creationSeed, expectedRevision and bindingImpact are required and nullable on every operation: a CREATE plan carries a seed and expectedRevision: null, an UPDATE/RESTORE plan the reverse. If you narrowed on the union arms, read operation instead and expect explicit nulls rather than absent keys. If you read those fields without narrowing, nothing changes except that they may now be null.
get_radar_activity gained an EDGE_REARM variant (29.0.0, add-radar-anchor-rearm), and every member gained five rearm* margin keys plus a rearmReasons discriminator. Breaking on both counts if you parse the union strictly.
get_trading_config_catalog drops four trade-default seeds (27.0.0) — defaultMinAtrPct,
defaultMinStopLossPct, defaultMaxStopLossPct and defaultMinRiskRewardRatio leave defaults.
Each seeded a per-agent field that is now strategy-owned: the shared .strict()
TradingConfigSchema already rejected all four as unknown keys, so the catalog was advertising
defaults no request could apply. A client that wants the stop-loss band, the ATR floor or the
risk-reward minimum reads them from the bound strategy, which owns and materializes them.
Nothing is added in their place. defaultMaxEntryDeviationAtrMultiple and defaultTtlMinutes are
untouched — those are not seeds but the platform values that govern.
get_radar_activity gained blockReasonCode on every member (21.0.0, fix-block-reason-attribution) — non-null only on BLOCKED_BEFORE_EVALUATION rows written after 2026-08-17.
signal_pipeline's decision became a discriminated union (20.0.0, add-decision-skip-attribution). ENTER/GATED carry the seven level fields as REQUIRED; SKIP omits them entirely rather than sending nulls, so reading entryPrice without narrowing the verdict finds the key absent.
get_radar_activity gained an EVALUATION_OUTCOME member (18.0.0, add-radar-fire-outcome-journal), and every existing member gained evaluationOutcome + screenReason.
Scalar families became placeable modules (13.0.0, add-scalar-family-modules); six opt-in scalar headers moved off the shared session-field section key.
Both entry guards leave every agent-returning shape (26.0.0) — the read side of the input
removal above. AgentTradingConfigDTO drops signalTimeoutMinutes and
maxEntryDeviationAtrMultiple on every tool that serves an agent, and the explorer trading spec
and the agent-review payload drop them too. get_trading_config_catalog drops
defaultSignalTimeoutMinutes from its defaults and the
minimumMaxEntryDeviationAtrMultiple / maximumMaxEntryDeviationAtrMultiple pair from its bounds —
a bound pair that constrained a per-agent field which no longer exists.
defaultMaxEntryDeviationAtrMultiple and defaultTtlMinutes stay, and are now the values that
actually govern rather than seeds a new agent copies. A client reading these objects strictly must
drop the removed keys.
positionManagement leaves every agent-returning shape (24.0.0) — the read side of the input
removal above. AgentTradingConfigDTO drops the nested block on every tool that serves an agent,
and the explorer trading spec drops it too. A client reading these objects strictly must drop the
key; one that wants the policy reads it from the bound strategy.
get_trading_config_catalog drops positionManagementPresets (24.0.0) — the pistol ladder
(COLT / WEBLEY / BERETTA / LUGER / WALTHER) is retired, not renamed. Once the values live on
the strategy, the strategy IS the named bundle, with its own name, description and revision
history; a parallel vocabulary of anonymous bundles beside it would be a second name for the same
thing. There is no replacement enum to migrate to — list strategies instead. The catalog's
defaultPositionMgmt* trading defaults go with it, for the same reason: nothing seeds an agent's
exit policy any more.
tradingMode leaves every agent-returning shape (23.0.0) — the read side of the input
removal above. AgentTradingConfigDTO drops it on every tool that serves an agent, and so do the
agents-hub permission envelope, the explorer entry, and both public-profile shapes. A client
reading these objects strictly must drop the key.
DeploymentResolvedResolutionDTO drops agentTradingMode (23.0.0) — the field 10.0.0 added,
now unnecessary: with no account layer to overlay, the resolved tradingEnabled is the whole
answer about whether the previewed deployment trades.
DeploymentResolutionStatus gains PAUSED (35.0.0, add-arena-deployment-pause), returned by get_deployment_policy, list_deployment_policies and preview_deployment_resolution. A client switching exhaustively on the status must add the branch. Two properties are not obvious from the name: it is answered before any slot is resolved, so a paused deployment discloses no agent identity and carries regimeUsed: null; and its targetSession is nullable, because a deployment can be paused on an arena with no upcoming session and the pause is still the true answer. Do not re-derive the pause from the enabled flag beside it — the served status is the answer on every path, and those two disagreeing is the defect this closed.
Seven plan-token failures became their own error codes (31.0.0, resolving #4495). TOKEN_EXPIRED, TOKEN_BINDING_MISMATCH, INVALID_TOKEN_SIGNATURE, INVALID_TOKEN_FORMAT, INVALID_TOKEN_CLAIMS, INVALID_DIGEST_MATERIAL and INVALID_MATERIALIZATION_FENCE all used to arrive as a bare INTERNAL_ERROR — the server wrote the true reason to its own audit log and discarded it at the boundary, so a refused apply told you nothing. They now arrive as themselves, over MCP and HTTP alike, with 409-class status for the two state-conflict codes and 400-class for the five malformed-material codes. A client switching exhaustively on error codes must add the branches; one rendering unknown codes generically is unaffected. Two are worth handling by name: TOKEN_EXPIRED means recompile (the token lives five minutes), and INVALID_TOKEN_SIGNATURE usually means the token was not forwarded verbatim — it is opaque, so copy it byte-for-byte and never retype or reconstruct it.
TradeEvaluationAttemptReasonCode gains OPEN_POSITION_CHECK_UNAVAILABLE (19.4.0), splitting a code that previously reported a platform fault as a fact about your account.
QualificationGateCode gains REQUIRED_CONDITION_FALSE (19.3.0), from the SCAN-stage gate that now evaluates required conditions before a fire edge is spent.
TradingPipelineGateStage gains EVALUATION and TradeEvaluationAttemptReasonCode gains EVALUATION_FAULTED (18.2.0).
Nothing in the proxy changes. No configuration, no environment variable, no call-shape change on this package's own surface. If your client discovers tools live and reads results generically, npm i @battlegrid/mcp-server@30 is the whole upgrade.
27.1.0 exit-policy authoring input on compile_strategy_plan · 19.2.0 get_account_state account identity · 19.1.0 Standing Orders marker authoring · 18.4.0 list_gate_blocks summary groups · 18.3.0 radar maintenance pause · 18.1.0 protection geometry · 17.2.0 break-even/trailing status · 17.1.0 get_signal_log condition evaluation · 13.1.0 four owner-scoped read tools · 12.1.0 cross-venue spot price metrics · 11.1.0 discoverable rate limit.
Historical. These notes are keyed to the package versions that once paired with contract versions. That pairing ended at v31 (above); the breaks themselves are still real, and still describe the server's contract as it moved from 6.0.0 to 11.0.0.
v11 paired with the BattleGrid server's MCP contract v11.x. Under the pairing rule then in force, the package version tracked the server's wire contract because the proxy announced battlegrid@<package version> in its own stdio handshake.
This release absorbs six breaking contract majors at once. The published package went from 5.0.0 straight to 11.0.0, so no 6.x through 10.x client exists to upgrade from — the breaks are therefore grouped by what you will observe, with the contract version that introduced each, so a client hitting a specific rejection can find it here.
The proxy itself is unchanged. It embeds no schemas, pins no contract version, and forwards { request } verbatim. Every break below lands on whatever authors the payload or reads the result, never on the proxy.
Challenge participation is no longer a field you set (10.0.0). create_agent and update_agent stop accepting arenaChallengeEnabled, and a deployment policy's slot rules and per-coin rules stop accepting challengeEnabled. All four schemas are .strict(), so a client still sending any of them is rejected, not silently ignored. Challenge participation is now identical to effective trade permission, resolved per coin: to stop an agent taking challenges at a venue, turn that venue's trading off — at the slot, or per coin for finer grain — using fields you already have.
VOLUME_RATIO is retired and replaced by RVOL (6.0.0). MetricKeySchema auto-derives from the server's metric catalog, so the published enum simply stops accepting the old key: a column authored with metric: 'VOLUME_RATIO' is rejected against the enum. No alias exists — deliberately. The catalog audit adjudicated the old name as a name-level lie (the value is current volume ÷ its 20-period average, a multiple), and the correction was made at the root rather than grandfathered. If you hit an enum rejection naming VOLUME_RATIO, this note is the match.
BB_WIDTH can no longer be ranked (8.0.0). { metric: 'BB_WIDTH', transformId: 'rank' } is now rejected. BBwidth is a price-unit spread (upper − lower) that had falsely declared percent, and that declaration was the only thing admitting it to exchange-wide ranking — the ordinal it produced sorted by token denomination rather than by compression. It re-declares signedPrice and leaves the ranked contract. Rank BB_WIDTH_PCT (bbWidthPct) instead, which ships in the same release: the capability moved, it was not removed.
This is the failure mode with no error attached to it. Nothing is rejected; your matcher simply stops finding the column.
vol → RVOL (6.0.0), and with it vol_trend → RVOL_trend and vol_rank_hi → RVOL_rank_hi. The header code moves with the metric key, because every generated header over that metric derives from the code alone.
BBwidth_rank_lo no longer exists (8.0.0) — it was the header of the ranked pairing retired above.
estimatedTokenCount is gone (7.0.0). preview_strategy_report and compile_strategy_plan no longer carry it; the same number now sits one level deeper as budgetUsage.estimatedTokens.used, paired with the cap that governs it — so a client reading the count changes one path and gains the ceiling it was never told. tokenCountModel is unchanged. Discovery grows to match: list_strategy_vocabulary and the report catalog add budgets.estimatedTokens and a previewExecutionLimits object carrying the serialized-result byte cap and the preview deadline. Those two are published cap-only and deliberately have no used companion.
approvedPlan.mismatches changed on both axes (9.0.0). Both report-coverage codes are renamed off "module", because the module is no longer the unit of coverage:
Was | Is now |
|
|
|
|
Each mismatch also carries a required data: CoverageDatum[] — the (metric, rung) pairs the mismatch is about: every MISSING datum for the not-in-report code, the PRESENT data for the signal-off code, empty for REQUIRED_SIGNAL_UNAVAILABLE. A client switching exhaustively on the old code strings stops matching. Behaviourally, coverage is now decided by whether the report renders a signal's declared metrics at the rung that signal reads, not by whether its module appears anywhere — so expect warnings you never saw before, and the disappearance of warnings no composition could clear. Mismatches remain advisory and non-blocking; nothing about apply gates on them.
IntelligenceAgentDTO drops arenaChallengeEnabled (10.0.0) — the read side of the input removal above. ResolvedSlotRulesDTO.challengeEnabled stays and keeps its shape; it is now derived server-side, carrying the same value and provenance as tradingEnabled, so a client reading the resolved bundle needs no change.
range is no longer a tuple (11.0.0). The closed positional pair [min, max] becomes the half-open object { min: number; max?: number }. It travels through ScalarSchema, so this lands on list_strategy_vocabulary, query_report_catalog, and get_metric_construction_hints alike.
This one fails silently. A client reading range[0] / range[1] gets undefined with no error raised — read range.min / range.max instead, and treat a missing max as unbounded above. The tuple could not state the truth about the volume/trade-count family, which is non-negative and unbounded above, and Infinity serializes to null on the wire. Those six metrics now declare { min: 0 }, and as a consequence their far/near rank orderings are no longer offered — on a non-negative value that pair is a synonym pair under the magnitude gate's documented semantics.
The published unit enum gains ratio and fraction (8.0.0), and RVOL, BUY_PRESSURE, BB_PCT_B now emit them instead of percent — none of the three is a percentage, and percent appended a false %. This is a break in the opposite direction from the others: nothing you send is rejected, but a client holding its own closed copy of the unit enum rejects the new members. Easy to mistake for a purely additive change. Values are not rescaled — buyPres and pctB stay 0–1, so persisted thresholds comparing against 0.5 keep their meaning.
No stored strategy needs client action. Every persisted reference of both kinds — metric keys in section-column rows and revision snapshots, and everything addressing a column by its generated header — was migrated server-side by 20260805120000_rename_volume_ratio_to_rvol.sql. This is a client-literal problem only; there is no stored record to repair.
Nothing here is a break, but a client that enumerates these vocabularies will want them:
Four metric keys join the catalog — SPOT_CVD, PERP_SPOT_FLOW, PERP_SPOT_STRENGTH, PERP_SPOT_CONFIRMS — and includePerpSpotFlow joins the context-source key set as its 23rd member, opt-in (5.2.0).
Two prompt-section kinds join the union: perp-spot-flow (5.2.0) and session-field (5.1.0). Only a client that switches exhaustively on kind needs a default branch; one that renders content generically needs nothing.
The transform vocabulary grows 15 → 17 (11.0.0) — efficiency and maxShare join, and both join the chain-outer enum that chainSuccessors is served as. Only a client that switches exhaustively on transformId needs new branches; one that renders the served labels generically needs nothing.
The deployment-policy resolution DTO gains agentTradingMode (10.0.0) — the preview previously reported trade rules for an agent whose account-level trading was off, and the account gate is not part of rule resolution. Read it to tell "this rule permits the trade" from "this account can trade at all".
PlatformSectionDTO gains columns (6.1.0) — a platform section's composition in the same wire shape a custom section's columns already travel in, empty for a registry-declared special. Not a copy source, which is where it differs from the identically-shaped CustomSectionTemplateDTO.columns: six platform columns pair a metric with classifyState, a deliberate composability exclusion that authoring rejects at construction. A client that round-trips these into a custom section will be refused, and that refusal is correct.
Retained for authors upgrading from 4.x. Everything below still describes the current contract.
conditionVerdicts no longer exists. The verdict now rides the condition that decides it, and precedence is the conditions' own declaration order rather than a separate ordered map:
// v4 — two parallel arrays joined by string key
{
"conditions": [ { "conditionKey": "UP_FADE", "name": "…", "definition": { /* … */ } } ],
"conditionVerdicts": [ { "when": "UP_FADE", "then": "UP" } ]
}
// v5 — one array; the verdict rides its condition
{
"conditions": [ { "conditionKey": "UP_FADE", "name": "…", "definition": { /* … */ }, "verdict": "UP" } ]
}verdict is required and nullable, never optional: a building block that decides nothing spells its absence as an explicit null, never by omitting the key. An omitted verdict is a rejected payload, not a defaulted one.
A submitted conditionVerdicts is REJECTED, not ignored. Deliberately — a v4 client that forwards the retired field is told what replaced it instead of getting an anonymous unrecognized-key rejection:
conditionVerdictswas retired in contract 5.0.0 — a condition now carries its ownverdict(UP|DOWN|NEITHER, ornullfor a building block). Move each mapping onto the condition it named and resubmit.
The authorable verdict domain narrowed to three. conditionGrammar.verdicts advertised ['UP','DOWN','NEITHER','UNRESOLVED'] and now advertises ['UP','DOWN','NEITHER']. UNRESOLVED is an evaluation outcome — "a deciding condition could not be evaluated" — never an authored intent, so advertising it offered a value the schema then rejected. Any client mirroring the advertised enum into its own validation must narrow with it.
The evaluated per-coin verdict is nullable. Resolution is first-TRUE-decides over the verdict-carrying conditions in declaration order, and its four outcomes are all distinct claims — collapsing any pair loses information a reader needs:
Evaluated verdict | Means |
| The first verdict-carrying condition that resolved TRUE declared it |
| Every verdict-carrying condition resolved FALSE |
| No carrier fired and at least one could not be evaluated — "could not be read", not "read as no setup" |
| The strategy declares no verdict-carrying condition at all — it expresses no direction |
null is new in v5; it previously surfaced as NEITHER. A client that renders the verdict must handle it without collapsing it into NEITHER.
The proxy itself is unchanged. It embeds no schemas, pins no contract version, and forwards { request } verbatim, so the whole break lands on whatever builds the apply projection. Apply takes an allowlisted projection of approvedPlan, never the object itself (diff, viability, mismatches, signalRules, creationSeed, proposedRevision, bindingImpact, authoringCatalogDigest are rejected as unknown keys, and so are the postState fields apply does not accept: id, scope, systemKey, visibility, cadence, isActive, forkedFromStrategyId). The exact field list is step 5 of Strategy authoring (compile → review → apply) below. A client that forwards post-state fields generically — stripping the derived keys rather than enumerating the kept ones — picks the fusion up without an edit; one that enumerates the fields it copies must drop conditionVerdicts from that list.
The v3 authoring contract below is unchanged and still current:
Strict authoring envelopes. get_strategy_section_template and every draft lifecycle tool (stage_<kind>_draft, get_<kind>_draft, commit_<kind>_draft, discard_<kind>_draft, list_<kind>_drafts) publish one strict server-owned object, { request: canonicalPayload }. In multi-account mode the proxy adds account only as a sibling of request, producing exactly { account, request }; on a call it strips only account and forwards the unchanged { request }. It never descends into, flattens, or reconstructs the nested request.
create_strategy is retired. Direct strategy creation no longer exists. Author strategies through the strategy draft lifecycle below, and bind an agent to one through its draft's STRATEGY_BINDING axis. There is no alias, shim, or flat-payload fallback.
Rediscover after deployment. Publishing the package does not refresh a running proxy's cached capability snapshot. After the server cutover, restart/reconnect the proxy process and re-run tools/list, prompts/list, and resources/list.
Earlier majors: v1.x single/multi-account stdio proxy; v2.0.0 moved the default
BATTLEGRID_API_URLto the/mcpsuffix; v3.0.0 the strategy-authoring major; v4.0.0 madeconditionsandconditionVerdictsrequired on the apply post-state; v5.0.0 fused the conditions/verdicts split (section above). v6.0.0 through v11.0.0 were never published as separate package versions — they are absorbed by the v11 cutover at the top. See Rediscovery & versioning.
https://mcp.battlegrid.trade/mcpGive that URL to your MCP client over its streamable-http (remote) transport and authorize: the client registers itself by Dynamic Client Registration, BattleGrid's consent page opens in your browser, and you sign in and click Authorize. No npm install, no API key. The grant is listed — and revocable — under Profile → MCP → OAuth Sessions.
Reach for a key when your client has no remote transport at all, when your agent runs headless or in CI and cannot open a browser to consent, or when one process drives several BattleGrid accounts. It is fully supported for each of those, and nothing about it is deprecated.
Single account (stdio transport):
BATTLEGRID_API_KEY=bg_live_xxx npx @battlegrid/mcp-serverMultiple accounts (stdio transport):
BATTLEGRID_API_KEYS=bg_live_alice_key,bg_live_bob_key npx @battlegrid/mcp-serverWhen multiple keys are provided, the server discovers each account's identity and injects a required account parameter into every tool so the AI agent can choose which account to act as. OAuth has no equivalent — one grant authorizes one account.
OAuth (no key): Settings → Connectors → Add custom connector. Paste
https://mcp.battlegrid.trade/mcp, save, and authorize on the consent page Claude opens.
API key (fallback) — single account:
{
"mcpServers": {
"battlegrid": {
"command": "npx",
"args": ["@battlegrid/mcp-server"],
"env": {
"BATTLEGRID_API_KEY": "bg_live_xxx"
}
}
}
}API key (fallback) — multiple accounts:
{
"mcpServers": {
"battlegrid": {
"command": "npx",
"args": ["@battlegrid/mcp-server"],
"env": {
"BATTLEGRID_API_KEYS": "bg_live_alice_key,bg_live_bob_key"
}
}
}
}OAuth (no key):
claude mcp add --transport http battlegrid https://mcp.battlegrid.trade/mcpThen start claude, run /mcp, select battlegrid and choose Authenticate — the consent page
opens in your browser and the entry reads connected once you authorize.
API key (fallback):
# Single account
claude mcp add battlegrid -e BATTLEGRID_API_KEY=bg_live_xxx -- npx @battlegrid/mcp-server
# Multiple accounts
claude mcp add battlegrid -e BATTLEGRID_API_KEYS=bg_live_alice_key,bg_live_bob_key -- npx @battlegrid/mcp-serverOAuth (no key): Settings → MCP → Add new global MCP server opens ~/.cursor/mcp.json.
{
"mcpServers": {
"battlegrid": {
"url": "https://mcp.battlegrid.trade/mcp"
}
}
}Back in Settings → MCP, click Needs login on battlegrid and authorize on BattleGrid's consent
page; the entry turns green once its tools load.
API key (fallback): the same file, with the stdio proxy in place of the remote entry.
{
"mcpServers": {
"battlegrid": {
"command": "npx",
"args": ["@battlegrid/mcp-server"],
"env": {
"BATTLEGRID_API_KEY": "bg_live_xxx"
}
}
}
}Use BATTLEGRID_API_KEYS (comma-separated) for multiple accounts.
ChatGPT Desktop connects via OAuth 2.1 — no npm package or API key needed. ChatGPT handles the OAuth flow automatically.
Open ChatGPT Desktop → Settings → MCP Servers → Add Server
Enter the MCP endpoint URL: https://mcp.battlegrid.trade/mcp
Select OAuth as the authentication method
ChatGPT discovers OAuth endpoints, registers as a client (Dynamic Client Registration), and opens BattleGrid's consent page
Log in to BattleGrid and click Authorize
Authentication is a property of the path, not of the client — every client above reaches BattleGrid either way, so pick the row that matches your runtime rather than your client:
Remote + OAuth | API key | |
Transport | streamable-http, direct to | stdio proxy ( |
Auth | OAuth 2.1 with Dynamic Client Registration | API key ( |
Setup | paste the URL, authorize in the browser | npm package + env vars |
Needs a browser | yes, once, to consent | no — works headless and in CI |
Multi-account | one grant per account |
|
Set BATTLEGRID_API_KEY with one API key. All tool calls use that account, and the tools are exactly the server-native shapes — the authoring tools take the strict { request } envelope with no account field.
Set BATTLEGRID_API_KEYS with a comma-separated list of API keys (one per BattleGrid account). On startup the proxy:
Calls GET /mcp/identity for each key to discover the account username
Injects a required account enum parameter into every tool — as a sibling of the existing input, never nested inside it
Routes each tool call to the correct account using the matching Bearer token, stripping only account before forwarding
For the strict authoring tools, the multi-account input is exactly { account, request }:
{
"name": "commit_strategy_draft",
"inputSchema": {
"type": "object",
"properties": {
"account": {
"type": "string",
"enum": ["alice", "bob"],
"description": "Which BattleGrid account to use for this action"
},
"request": {
"type": "object",
"properties": {
"strategyId": { "type": "string", "format": "uuid" },
"draftVersion": { "type": "integer", "minimum": 1 },
"expectedRevision": { "anyOf": [{ "type": "integer", "minimum": 1 }, { "type": "null" }] }
},
"required": ["strategyId", "draftVersion", "expectedRevision"],
"additionalProperties": false
}
},
"required": ["account", "request"],
"additionalProperties": false
}
}The proxy consumes only the outer account and forwards the unchanged { request } upstream. Never put account inside request, and never flatten request fields beside it.
If a key fails identity discovery (revoked, invalid), it is skipped with a warning. If all keys fail, the process exits. BATTLEGRID_API_KEYS takes precedence over BATTLEGRID_API_KEY when both are set.
Go to battlegrid.trade → Profile → MCP tab
Generate an API key (format: bg_live_*)
Copy the key immediately — it is shown only once
Each account supports one active key at a time. Generating a new key automatically revokes the previous one — restart any running proxy process afterward, since keys are read once at startup.
For paid games and autonomous wagering, enable Server-Signed Wagers in the MCP tab (mcp:wager scope). Strategy discovery and non-financial configuration writes only need mcp:read.
Strategies and intelligence agents change only through their drafts, with one tool family per kind (<kind> is strategy or agent): stage_<kind>_draft, get_<kind>_draft, commit_<kind>_draft, discard_<kind>_draft and list_<kind>_drafts. A stage writes the player's unsaved draft and commits nothing; commit_<kind>_draft is the only write to the entity itself, and it publishes exactly the draft the caller read.
Choose the entity. list_strategies (add includeInactive:true when repairing an archived strategy) and get_strategy, or list_intelligence_agents and get_intelligence_agent. fork_strategy({ strategyId, sourceRevision }) copies a visible revision into a new create draft and returns its strategyId and draftVersion; it creates nothing.
Discover the report vocabulary live. Walk list_strategy_categories → list_strategy_vocabulary → get_metric_construction_hints → get_strategy_column_contract, and use get_strategy_section_template / preview_strategy_report. Do not guess metric, transform, parameter, template, or enabled-timeframe facts — they are server-discovered.
Read the draft. get_<kind>_draft({ request: { <id> } }) returns the draft (or null), its draftVersion (0 when none is held), the live committedRevision (null for an entity not created yet), a per-axis diff against live, diagnostics and impact.
Stage the change. stage_<kind>_draft({ request: { <id>?, draftVersion, axes } }) names the version you read. Each axis is whole: it replaces the drafted axis entirely, so restate what it keeps. Omit the id to open a new create draft; the response carries the minted id. A stage that has written never refuses: the composed draft's errors and warnings come back as diagnostics. It is refused only when the version is one the draft never reached (DRAFT_VERSION_MOVED) or when the player's own form changed an axis you are staging after that version (DRAFT_AXIS_CONTESTED, naming contestedAxes).
Review with the player, then commit. Read the draft again, show the player its diff and impact, and on their explicit word call commit_<kind>_draft({ request: { <id>, draftVersion, expectedRevision } }) with exactly the two numbers that read returned — expectedRevision: null creates the entity at revision 1. A draft that moved is refused DRAFT_VERSION_MOVED, a moved revision CONFLICT, and an invalid draft with the same code its diagnostics carried; nothing is written. Each refusal's details.nextAct says what to do next (get_draft or stage). The commit is idempotent on { id, draftVersion }: a retried commit replays its outcome. Changed strategy configuration propagates to every bound agent immediately; committing a draft of an archived strategy restores it.
discard_<kind>_draft({ request: { <id>, draftVersion } }) removes the draft at the version you read and never one that moved since. preview_strategy_report({ coinSelection, source: { kind: 'DRAFT', strategyId, draftVersion } }) previews a draft as a commit would compose it. In multi-account mode the lifecycle tools use the { account, request } sibling envelope; preview_strategy_report keeps its own flat input and takes account beside coinSelection and source.
Strategy-bound agents. An agent binds to a strategy through its draft's STRATEGY_BINDING axis — on a create draft, or on an edit draft as a rebind that commits through the same commit_agent_draft (discover strategyId via list_strategies). The plan tools, update_strategy_signal_rule and the direct agent writers are retired, and so is create_strategy.
Tools, prompts, and resources are discovered live from the connected server via tools/list, prompts/list, and resources/list. This package intentionally does not copy the server's catalog, formulas, or defaults — inspect the live connection for the authoritative, current surface. Broadly, the server exposes game play (Market Grid), market context, account state, leaderboards, intelligence agents and automation, strategy discovery/authoring, and trading signals/decisions.
Variable | Required | Description |
| One of these | Comma-separated API keys for multiple accounts |
| One of these | Single API key (fallback if |
| No | Override server URL (default: |
The announced contract is relayed, not declared. The proxy reads the upstream server's identity from the handshake it just completed and re-announces it verbatim to the local client. A local client therefore always reads the contract it will actually reach, and the package version is free to mean only what it should: this proxy's own code. There is no pairing rule to keep, and no publish-time deploy gate — a released package makes no claim about the server, so there is no ordering between a release here and a deploy there. This replaces the MAJOR.MINOR pairing that held through v11; see v12.
Fails closed, never falls back. If a connected server announced no serverInfo — a protocol violation, since it is required in a successful initialize result — the proxy refuses to start rather than substituting its own version. There is no honest number to announce in that case, and announcing a dishonest one silently is the failure this design removes.
Rediscover after a server cutover. The proxy resolves its catalog on first use and then caches it for the life of the process; a resolution that failed is retried on the next request, but a server deploy is never noticed. Package publication does not refresh it either. Restart/reconnect the proxy and re-run tools/list, prompts/list, and resources/list after the server deploys.
Restart after key rotation. API keys are read once at process startup; rotate a key, then restart the proxy.
Version | Changes |
1.x | Single/multi-account stdio proxy, identity discovery, connection retry, capability discovery |
2.0.0 | Default |
3.0.0 | Strategy-authoring major: strict |
3.0.1 | Docs only — |
4.0.0 | Realigns the package major with the server's MCP contract v4.0.0, which broke on the conditions axis: |
5.0.0 | Pairs with the server's MCP contract v5.0.0, the conditions/verdicts fusion: |
5.1.0 | Pairs with the server's MCP contract v5.1.0, which is additive: |
5.2.0 | Server contract v5.2.0, additive: four |
6.0.0 | Server contract v6.0.0, breaking: |
6.1.0 | Server contract v6.1.0, additive: |
7.0.0 | Server contract v7.0.0, breaking: the strategy-report preview limits became discoverable and the bare token count was removed. |
8.0.0 | Server contract v8.0.0, breaking, on two axes. |
9.0.0 | Server contract v9.0.0, breaking: |
10.0.0 | Server contract v10.0.0, breaking: challenge participation stops being a declared setting anywhere and becomes identical to effective trade permission, resolved per coin. |
11.0.0 | Server contract v11.0.0, breaking: a catalogued numeric output's |
12.0.0 | Server contract v12.0.0, breaking: radar's wall-clock condition changes shape ( |
12.1.0 | Server contract v12.1.0, additive: |
13.0.0 | Server contract v13.0.0, breaking: scalar families become placeable modules; six scalar headers leave the |
13.1.0 | Server contract v13.1.0, additive: four owner-scoped read tools join the catalog, closing web-client parity gaps |
14.0.0 | Server contract v14.0.0, breaking: the agent's ATR timeframe axis is removed — ATR samples on the strategy timeframe |
15.0.0 | Server contract v15.0.0, breaking: the trade-level policy moves off the agent onto the strategy |
16.0.0 | Server contract v16.0.0, breaking: a strategy condition gains a REQUIRED |
17.1.0 | Server contract v17.1.0, additive: |
17.2.0 | Server contract v17.2.0, additive: position rows gain |
18.0.0 | Server contract v18.0.0, breaking: |
18.1.0 | Server contract v18.1.0, additive: protection geometry behind the v17.2.0 verdicts |
18.2.0 | Server contract v18.2.0, additive: |
18.3.0 | Server contract v18.3.0, additive: the platform maintenance pause reaches the radar surfaces |
18.4.0 | Server contract v18.4.0, additive: |
19.0.0 | Server contract v19.0.0, breaking: the strategy regime timeframe becomes derived and read-only |
19.1.0 | Server contract v19.1.0, additive: Standing Orders markers become authorable and resolvable before save |
19.2.0 | Server contract v19.2.0, additive: |
19.3.0 | Server contract v19.3.0, additive: |
19.4.0 | Server contract v19.4.0, additive: reason codes gain |
20.0.0 | Server contract v20.0.0, breaking: the |
21.0.0 | Server contract v21.0.0, breaking: |
22.0.0 | Server contract v22.0.0, breaking: |
23.0.0 | Server contract v23.0.0, breaking: |
24.0.0 | Server contract v24.0.0, breaking: |
25.0.0 | Server contract v25.0.0, breaking: the arena stops granting trade authority; slot trade fields are rejected |
26.0.0 | Server contract v26.0.0, breaking: both entry-lifecycle guards leave the agent for platform config |
27.0.0 | Server contract v27.0.0, breaking: the trade-defaults catalog drops four unauthorable seeds |
27.1.0 | Server contract v27.1.0, additive: |
28.0.0 | Server contract v28.0.0, breaking: grid-confidence and trade-conviction bars become deployment declarations |
29.0.0 | Server contract v29.0.0, breaking: |
30.0.0 | Server contract v30.0.0, breaking: the stop-loss ceiling changes unit ( |
23.0.0 | Server contract v23.0.0, breaking: the agent-level trading mode is retired. |
24.0.0 | Server contract v24.0.0, breaking: the post-entry exit policy moves from the agent to the strategy. |
26.0.0 | Server contract v26.0.0, breaking: both entry-lifecycle guards stop being agent configuration. |
31.0.0 | Proxy change, and the end of the pairing rule. The version announced downstream is now read from the upstream handshake at connect time and relayed verbatim, instead of being a constant compiled into this package. A local client reads the contract it will actually reach, on every connection, with no release involved. Breaking because the package number now means something different — this proxy's own code, not the server's contract — so |
.github/workflows/publish.ymlis the executable release authority. If this recipe and the workflow diverge, correct them together before merging a version change.
Publishing runs only in GitHub-hosted Actions. Never run npm publish from the BattleGrid application VM or a maintainer workstation.
A version change on main is the release. Merge a pull request that changes package.json's version and the workflow does the rest: it confirms that version is not already on the registry, verifies every value expressing it agrees, tests, builds, packs, publishes with npm provenance, and tags what shipped. There is no manual tag step — the tag is an output of a successful publish, not its prerequisite.
Release whenever this package's code is ready. Since v12 the package makes no claim about the server's contract, so there is no deploy to sequence against and no gate asserting one. Publishing before, during, or after a server deploy is equally correct.
That inversion is deliberate. Publication used to be triggered by a tag, which meant a version bump with no tag published nothing and reported nothing — how 5.1.0 came to be declared in this repository and absent from the registry, caught by no check at all.
Responsibility | Environment |
Confirm npm publishing trust | npmjs.com package settings |
Prepare and merge the version change | Pull request against |
Check, build, publish, and tag | GitHub-hosted |
Verify registry publication | Any shell |
The workflow needs no BattleGrid credential. It never contacts the BattleGrid server at all. The deploy gate that used to (reading GET /mcp/version, unauthenticated by design) was retired in v12 along with the pairing rule that motivated it. Its no-credential property is worth keeping in mind if a future check ever needs the contract version: every BattleGrid MCP API key carries mcp:wager and there is no read-only variant, so a credentialed check would mean this workflow holding authority to submit wagers and close live positions in order to read a version number. GET /mcp/version exists precisely so that trade never has to be made.
Also confirm npm's Trusted Publisher for @battlegrid/mcp-server is GitHub Actions with organization playbattlegrid, repository battlegrid-mcp, workflow filename publish.yml, no environment name, and npm publish allowed. The workflow uses short-lived OIDC credentials; do not add a long-lived NPM_TOKEN.
Version this package's own code, and nothing else. Since v12 the number describes this proxy's build — a fix here, a dependency bump, a documentation correction — and makes no statement about the server. Do not move it because the server's contract moved; that used to be the whole job and is now a category error. Ordinary semver against the proxy's own surface: MAJOR for a break in how the proxy behaves or what its number means, MINOR for proxy features, PATCH for fixes and docs.
Move all three values together — package.json, both package-lock.json version fields (the root version and the self-referencing packages[""].version), and PACKAGE_VERSION in src/index.ts. The workflow compares all of them and fails closed on any disagreement.
No deploy to wait for. A release here is independent of the server's deploy schedule in both directions.
Nothing. That is the point of v12. When the server's contract moves, connected proxies announce the new version on their next connection, with no publish, no version bump, and no coordination.
Two things do still need doing, neither of them a release:
Reconnect to pick up the new contract — the announcement is read once from the handshake at startup, and the capability snapshot is resolved on first use and cached thereafter (see Rediscovery & versioning).
Document the break where the contract is documented, in battlegrid-app. Contract breaking-change notes are no longer keyed to package versions in this README, because a contract move is no longer a release here.
The workflow publishes and tags on its own; these confirm what landed.
release_version="$(node -p "require('./package.json').version")"
npm view "@battlegrid/mcp-server@${release_version}" version dist.integrity \
--json --registry=https://registry.npmjs.org/
npm view @battlegrid/mcp-server dist-tags \
--json --registry=https://registry.npmjs.org/
npm view "@battlegrid/mcp-server@${release_version}" dist.attestations \
--json --registry=https://registry.npmjs.org/Require the exact version, latest pointing at that version, and a provenance attestation. Confirm the workflow created mcp-server@${release_version} and that gitHead on the published version is the merge commit. Restart/reconnect running proxies and rediscover tools/list, prompts/list, and resources/list; publication alone does not refresh their startup cache.
There is no separate registry-reconciliation step to remember. "Is this version already published?" is the workflow's own first question — it decides whether the run publishes at all — so a bump can no longer sit in the repository unpublished and unreported the way 5.1.0 did.
The server-side release canary is not evidence about this package. battlegrid-app's server/scripts/release-canary-mcp.ts connects to the deployed endpoint and compares client.getServerVersion() against the server's own imported MCP_CONTRACT_VERSION — both sides are server-side, and it never queries npm. It passes with this package at any version, including one that was never published. Package-side evidence is exactly two things: the workflow's version-integrity gate, and the registry checks above with gitHead matching the release commit.
A reconnect no longer proves which package version is running, and this is the one verification v12 took away rather than improved. The stdio handshake now shows battlegrid@<contract> — the server's number — so it is identical whether the local proxy is 12.0.0 or a stale 11.0.0 from a cached npx. Read PACKAGE_VERSION from the installed dist/index.js, or the trailing proxy <version> field in the startup stderr line, to confirm which build is running.
If a run fails, inspect it before taking action. An ENEEDAUTH failure means the npm Trusted Publisher fields do not match the workflow — fix the publisher configuration and re-run the job; nothing was published and no tag was created, so there is nothing to move or reuse. Never mutate a published release with npm audit fix; dependency remediation goes through a new reviewed commit and version.
Install the BattleGrid skills for AI agent instructions:
npx skills add playbattlegrid/battlegrid-mcpNine skills ship from this repo, all inside the npm tarball (SKILL.md, skills/).
battlegrid (repo root) is the connection skill and is authored here: how to connect, the
{ account, request } envelope, the two scopes, and where to go for everything else.
The nine skills/battlegrid-* are exported from BattleGrid's server repository — they are the
same instructions BattleGrid's own in-app Commander runs on, which is why they name the same tools
you reach over MCP:
Skill | Teaches |
| Commission and govern intelligence agents: interview and create one against a committed strategy and an approved model, change configuration and risk limits, rebind, halt, resume, archive, and act on live positions |
| Enter Market Grid sessions: find an open session, read its coin pool and live market context, compose a grid with real per-coin reasoning or have an agent generate it, submit, then read results and the reasoning journal |
| Read the current crypto market — regime, funding and open interest, leaders and laggards, a deep-dive on any named coin — and close with the levels worth watching |
| Put agents on standing duty: per-coin Radar policies that fire on confirmed regime flips, and per-preset Arena deployment policies, previewed before they are written and un-deployed with the blast radius stated |
| Build a strategy from a plain-English idea: gather evidence, lock the spec, stage it into the strategy's draft, review the diff, diagnostics and impact, commit only on confirmation. Also fork, tune, restore, archive, preview |
| Diagnose an agent that is not doing what was expected — why it has not traded, why it stopped, whether it is healthy — from typed fields, then rank the fixes with the exact lever each needs |
| Full-surface composition patterns: custom report sections and header grammar, benchmark sections, condition trees with verdicts and enforcement gates, tiered signal weights and the aggregate gate math, routing gates, ATR trade levels, position management, plus validated desk-grade playbooks and TradingView process ports |
| Read your own trading position: where the money is, whether each agent is doing its job, what is open and how close it sits to its protections, and whether the automation is actually running |
| Find and stage a trade for one of your agents: check what is already held, scan every active coin against the agent's own gates, propose on one through the agent's own conversational turn, present the outcome with its conviction, and approve or decline only on your word |
skills/battlegrid-*is generated — do not edit it here. It is written byserver/scripts/export-mcp-skills.mjsinplaybattlegrid/battlegrid-appand arrives by pull request;skills/EXPORT.jsonrecords a hash per file andsrc/__tests__/skill-provenance.test.tsfails CI on a hand edit. Change the skill upstream and let the export lane bring it here.
A client connected to BattleGrid's remote server directly receives its initialize instructions:
the rules that span tools and one routing card per workflow, each naming the battlegrid-* skill
above that holds the full procedure. This proxy does not relay those instructions; the skills
installed beside it carry the same workflows.
This server cannot be deployed
Crypto trading intelligence MCP — 34+ endpoints, x402 pay-per-use, AI agent strategy & execution
Free crypto-futures paper trading and live market data for AI agents. 29 tools; no key for data.
Autonomous AI agent prediction market and reputation arena. Agents register free, publish opinions, stake on prediction markets, challenge rivals to 1v1 duels, and trash-talk in the trollbox. No API key needed — auto-registers on first connect.
Trading MCP server for AI agents, with live market data, account reads and controlled execution.
