WHAT: answers a coding question (library/API usage, error diagnosis, best practices) with real citations, and ships a signed attestation that the citations were re-checked by an independent verification pipeline. WHEN: you need a grounded, cited answer mid-task and are willing to pay per call. WHEN NOT: for open-ended code generation, for questions needing private/proprietary docs, or when a free unverified answer is good enough (use web search). COST: $0.10 USDC per call (x402 on Base; 25-call free tier). LATENCY: ~2-6s typical (search + citation re-check). OUTPUT: {answer, citations: [{title, url, snippet}], provider, served_in_ms, citation_check: {verdict, confidence}, attestation, attestation_hash, signature}.

WHAT: independently checks another agent's work (citation support, code correctness via sandboxed pytest, or dataset/schema conformance) and returns a verdict plus an EIP-191 signed attestation verifiable offline. WHEN: before paying for subcontracted agent work, before merging agent-produced citations/code/data into a deliverable, or when you need portable proof a third party checked the work. WHEN NOT: for work you produced and already trust, for subjective judgments (style, product decisions), or for languages other than Python (code jobs). COST: $0.15 USDC per call (x402 on Base; 5-call free tier). LATENCY: ~1s (data) to ~30s (citation fetches / code sandbox). OUTPUT: {verdict: pass|fail|partial|abstain, confidence, evidence: [...], attestation, attestation_hash, signature, served_in_ms}. 'abstain' = too ambiguous to call: get a second opinion, do not treat as pass.

WHAT: returns the latest signed SLA report for a watched x402 service: uptime_pct, price_match_pct, and current verdict (honoring, price_mismatch, unexpected_status, invalid_payload, unreachable) from probes every 30 minutes, EIP-191 signed. WHEN: before routing paid traffic to an x402 service you do not control, or when checking whether a service still charges its advertised price. WHEN NOT: for real-time health (reports aggregate 24h), for services not yet watched (GET /watchtower/targets lists coverage), or for verifying work products (use oracle_verify_work). COST: $0.05 USDC per report (x402 on Base; 3 free reports). LATENCY: ~1s, reports are pre-computed. OUTPUT: {report: {subject, window, checks, uptime_pct, price_match_pct, current_verdict, timestamp}, attestation, attestation_hash, signature, generated_at}.

WHAT: pre-payment counterparty check. Given a 0x address or URL, returns a clear/caution/avoid verdict with per-source evidence (watchtower track record, CDP merchant catalog, address hygiene) and a signed attestation. WHEN: before sending any x402 payment to a service you have not paid before, or before raising a payment limit. WHEN NOT: for services you already trust, for judging delivered work (use oracle_verify_work), or as a guarantee. An unknown payee is always 'caution', never 'clear'. COST: $0.005 USDC per check (x402 on Base; 5 free checks). LATENCY: ~1-3s (catalog lookup + RPC call). OUTPUT: {verdict, confidence, evidence: [...], attestation, attestation_hash, signature, served_in_ms}.

WHAT: grades an agent output against your rubric and returns a 0-100 weighted score with per-criterion evidence and a signed attestation. Deterministic graders: structure coverage, citation presence, claim support against provided sources, length rules. WHEN: accepting or rejecting another agent's written deliverable, ranking candidate answers, or logging a signed QA receipt before downstream payment. WHEN NOT: for subjective taste calls (style, product direction), for citation truth-checking alone (use oracle_verify_work), or with a vague rubric: the judge is literal and grades exactly what you write. COST: $0.10 USDC per grading (x402 on Base; 3 free gradings). LATENCY: ~1s without source checks, up to ~30s when claim-support fetches sources. OUTPUT: {score, verdict: pass|abstain|fail, per_criterion: [{criterion, score, evidence, weight}], attestation, attestation_hash, signature, served_in_ms}. Bands: >=70 pass, 40-69 abstain, <40 fail.

WHAT: checks whether a package name really exists in PyPI or npm. On a miss, typosquat_adjacent flags real names within edit distance 2. WHEN: before installing or importing a package a model suggested; published studies measure 5-20% hallucination rates for suggested package names. WHEN NOT: for private packages outside the public registries. COST: free (25 calls/day per client IP shared across the four attester_* demo tools, reset 00:00 UTC; over quota the response points at the paid route). Paid route /v1/package/exists is $0.002 per call. OUTPUT: {exists, latest_version, typosquat_adjacent, adjacent_to, proof: {artifact_sha256, source_url}, attestation, attestation_hash, signature, served_in_ms}.

WHAT: checks whether a symbol (function, class, constant) exists in a package at the latest indexed version, with closest-match suggestions on misses. WHEN: before calling an API you are not sure exists; invented symbols are a leading agent compile error. WHEN NOT: for symbols created dynamically at runtime (static analysis ceiling). COST: free (25 calls/day per client IP shared across the four attester_* demo tools, reset 00:00 UTC; over quota the response points at the paid route). Paid route /v1/symbol/exists is $0.005 per call. OUTPUT: {exists, kind, deprecated, since_version, closest_match, version_resolved, proof, attestation, attestation_hash, signature, served_in_ms}.

WHAT: returns the exact signature, structured params, docstring summary, and deprecation status for one symbol, resolved at a pinned version when given. WHEN: before writing a call against an API you have not used at this version. WHEN NOT: PyPI type annotations are unavailable (the index stores names, not annotations); npm types come from .d.ts as written. COST: free (25 calls/day per client IP shared across the four attester_* demo tools, reset 00:00 UTC; over quota the response points at the paid route). Paid route /v1/symbol/signature is $0.01 per call. OUTPUT: {exists, symbol, kind, signature, params, docstring_first_line, docstring_summary, deprecated, deprecated_since, version_resolved, proof, attestation, attestation_hash, attestation_signature, served_in_ms}. Note: 'signature' is the function signature; the EIP-191 envelope signature ships as attestation_signature.

WHAT: symbol-level diff between two pinned versions of one package: added, removed, changed signatures, new deprecations, and a breaking/additive/neutral classification, signed. WHEN: before upgrading a dependency, or when reviewing what an upgrade breaks. WHEN NOT: unresolvable versions come back as a 422 error dict, never a partial diff. COST: free (25 calls/day per client IP shared across the four attester_* demo tools, reset 00:00 UTC; over quota the response points at the paid route). Paid route /v1/diff is $0.02 per call. OUTPUT: {classification, added, removed, changed, new_deprecations, counts, proof, attestation, attestation_hash, signature, served_in_ms}.