AINumbers Fintech Intelligence Suite MCP Server

Validate an A2A agent-card.json against the v1.0 shape, check signatures, and confirm extension declarations. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("validate_a2a_agent_card").

Validate AP2 Policy Mandate JSON payloads against the Unified Build Contract v1.0 schema. Auto-generates MCP tool definitions from the mandate and simulates agent ingestion of agent_instructions. Use when authoring or testing AP2 agentic payment policies. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("validate_ap2_mcp_policy").

Canton DvP Atomicity Validator: OpenChainGraph compute node (settlement_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Output feeds: 505-tokenized-collateral-eligibility-checker. Open at: https://ainumbers.co/tools/507-canton-dvp-atomicity-validator.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Validate atomic DvP settlement on Canton — PFMI P12, finality model, unwind semantics. Outputs: verdict, atomicity_flag, finality_flag, herstatt_flag, compliance_flags, pacs008, execution_hash. Output schema: call describe_tool("validate_canton_dvp_atomicity").

Canton Party Allowlist Validator: OpenChainGraph compute node (compliance_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Open at: https://ainumbers.co/tools/509-canton-party-allowlist-validator.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Screen parties against FATF Travel Rule requirements and AML/KYA checks for Canton Network onboarding. Outputs: party_results, portfolio_verdict, compliance_flags, iso20022_party_identification. Output schema: call describe_tool("validate_canton_party_allowlist").

Collateral Swap Eligibility Validator: OpenChainGraph compute node (collateral_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Consumes upstream artifacts from: 505-tokenized-collateral-eligibility-checker, 507-canton-dvp-atomicity-validator. Open at: https://ainumbers.co/tools/515-collateral-swap-eligibility-validator.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Validate a collateral swap (upgrade/downgrade trade) under GMSLA, GMRA, and SFTR Art 15 collateral reuse rules. Outputs: eligibility, hqla_impact, hqla_tier_a, hqla_tier_b, net_economic_value, pacs008, value_a, value_b. Output schema: call describe_tool("validate_collateral_swap_eligibility").

Tokenized Fund Collateral Validator: OpenChainGraph compute node (collateral_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Consumes upstream artifacts from: 505-tokenized-collateral-eligibility-checker. Open at: https://ainumbers.co/tools/514-tokenized-fund-collateral-validator.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Validate tokenized MMF / CNAV / LVNAV / VNAV fund shares as collateral. Outputs: adjusted_collateral_value, eligibility, haircut_applied, hqla_tier. Output schema: call describe_tool("validate_fund_collateral").

Verify an artifact's input_attestations[] (ChainGraph Standard §23): per RFC 6901 pointer, checks the attested value resolves inside policy_parameters and its digest binding matches, then verifies each type along its own path -- vc-2.0 via the shipped §16/§13.11 Data Integrity proof, rfc3161-snapshot via the same §20 rfc3161-tst verifier (no second RFC 3161 implementation), c2pa-manifest structurally (hard-binding digest match), zktls structurally-only (reported verifiable:"external" -- OCG never treats it as confirmed). Returns one { pointer, type, structural, verifiable } record per entry. Pure client-safe compute, zero network.

Validate an MCP server.json against the 2025-12-11 schema and the official registry publishing rules; returns findings, a registry-readiness score, and an optional compliant skeleton. Use when a developer wants to check a server.json before publishing to the MCP Registry. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("validate_mcp_server_json").

Verify an artifact's private_inputs[] (ChainGraph Standard §25 ocg-private-input@1) WITHOUT ever seeing the plaintext witness: per RFC 6901 pointer, checks the pointed value inside policy_parameters IS the declared sha256-salted@1 commitment (never the plaintext, §25.2), that the commitment scheme is known, and -- when a §18 compute_proof is present -- that its journal commits the same commitment and binds output_payload. Optionally accepts an out-of-band {pointer, salt, input_value} disclosure package (authorized-verifier path) and recomputes sha256(salt || cgCanon(input_value)) to confirm it equals the commitment. Returns one {pointer, verifiable} record per entry: "proof-only" | "disclosed-verified" | "commitment-only" (no proof yet -- structural + plaintext-exclusion only) | "failed". Pure client-safe compute, zero network, never requires the plaintext.

Multi-Currency PvP Validator: OpenChainGraph compute node (settlement_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Consumes upstream artifacts from: 507-canton-dvp-atomicity-validator, 505-tokenized-collateral-eligibility-checker. Open at: https://ainumbers.co/tools/511-multi-currency-pvp-validator.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Validate atomic Payment-vs-Payment (PvP) settlement across currency pairs. Outputs: has_canton_leg. Output schema: call describe_tool("validate_pvp_settlement").

Tokenized Security Lifecycle Validator: OpenChainGraph compute node (compliance_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Consumes upstream artifacts from: 510-digital-asset-regulatory-classifier. Open at: https://ainumbers.co/tools/512-tokenized-security-lifecycle-validator.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Validate issuance, corporate actions, and maturity/redemption lifecycle for tokenized securities on Daml/Canton. Outputs: all_gaps, critical_gaps, event_matrix, verdict, verdict_badge. Output schema: call describe_tool("validate_tokenized_security_lifecycle").

Hash-aware sibling of build_workflow_links (ChainGraph Standard v0.1 §8.1). Returns an ordered, executable DAG over the ChainGraph suite's verifiable tools, with explicit parent_hash wiring: which upstream execution_hash each step must cite in its chain block. Pass target_tool_id to build the chain that produces that node (walks consumes-edges back to roots), or tool_ids for an explicit ordered list, or neither to list available ChainGraph nodes. Agent loop: run a node, capture its execution_hash, pass it as the parent_hash for each downstream node, then verify with verify_execution_hash.

Builds a Merkle-rooted disclosure manifest from a caller-supplied digest list (DATAROOM-1-BUILD-SPEC.md §DR-4) -- the agent hashes files itself and passes {path,size,digest,content_type} entries; the worker never sees file contents. Same leaf scheme as tools/546-disclosure-manifest-builder.html: sha256(path|digest|size), duplicate-last-leaf on an odd level. Entries are sorted by path before hashing so the root is order-independent. Returns the manifest object + merkle_root; sign and anchor it client-side (or via a separate §16 signing step) if a signed artifact is required.

Composes a session receipt (build_session_receipt logic), an optional §27.6 HA evidence bundle (ha_bundle_export logic, included only when ha_records is supplied and non-empty), and a disclosure manifest (build_disclosure_manifest logic) from one array of already-produced artifacts, all keyed to the same input hashes -- replacing what today takes 4-6 separate tool calls with hand-carried hashes. Calls the same in-process functions those standalone tools use; if any one section fails to build, the whole call fails isError:true with that section's own message -- no partial pack. Optionally carries a run_chain decision_trail (per-step reason codes: gate rule id, escalation rule id, input_required cause) as pack-level metadata so an auditor-facing pack answers "why did each step happen" without a second round-trip.

Build or validate a Google AP2 Checkout/Payment Mandate VDC (Open/Closed). Targets the external AP2 spec, not the AINumbers Policy Mandate. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("build_google_ap2_mandate").

Aggregates execution_hashes from N ChainGraph tool calls in one agent session into a single SHA-256 Merkle root (session_receipt_root). Returns a tamper-evident session receipt and a regulator-framed PTG-01 audit prompt. One receipt covers an entire agent session: supply all execution_hashes in call order. The Merkle root is deterministic — the same hashes in the same order always produce the same root. Compliant with EU AI Act Art. 12 (transparency) and DORA ICT audit-trail requirements.

Constructs an ordered set of ready-to-use deep-links for a named AINumbers workflow chain or an ad-hoc sequence of tools. Each link points directly to the browser tool; prefill-enabled steps accept #in=<base64url(JSON)> fragments so the tool opens pre-filled. Zero server-side execution -- all tool logic runs deterministically in the user's browser. Use this to hand a user a complete workflow: open step 1, run it, export its Policy Mandate, open step 2 (pre-filled from step 1 outputs), repeat. 416 named chains are available — enumerate them with find_chain.

Parses CSV text with the SAME strict RFC 4180 parser as the tools/554 browser workbook -- quoted fields, embedded quotes/commas/newlines, CRLF -- and REJECTS malformed input rather than repairing it (deterministic beats forgiving for hash-stable digests). Numeric-looking and TRUE/FALSE cells are type-coerced; everything else stays a string. Returns the resulting A1-style cells (raw + evaluated value) plus the sheet's row/column extent, ready for workbook_evaluate or workbook_range_digest -- unless as_artifact is true, in which case it returns a full OCG v0.4 artifact instead (see as_artifact).

Feeds a set of A1-style cells (formulas and/or literals) through the SAME headless evaluator the tools/554 browser workbook uses -- topo-sorted dependency graph, ~20 functions (SUM AVG MIN MAX COUNT COUNTIF IF AND OR NOT ROUND ABS CONCAT LEN LEFT RIGHT TRIM UPPER LOWER SUMIF), cycles resolve to "#CYCLE!" and any non-finite result to "#NUM!" rather than propagating. Returns the computed value of every supplied cell -- pure, no digest, no receipt -- unless as_artifact is true, in which case it returns a full OCG v0.4 artifact instead (see as_artifact).

Spreadsheet-range digest only; NOT the general artifact emitter (use emit_chaingraph_artifact to wrap arbitrary computation output — this tool's as_artifact mode covers whole-sheet workbook runs only). Evaluates the supplied cells (same evaluator as workbook_evaluate) and computes a canonical values_digest over one A1-style range (e.g. "B2:D9") -- the exact `executionHash(values, {})` path the rest of OCG uses, per chaingraph/workbook/INPUT-MANIFEST.md (WB-2). Returns one `ranges[]` fragment of the Spreadsheet Input Manifest schema; assemble the full manifest (source.csv_digest, produced_by/produced_at) around it. Same range digested from the same cells in the tools/554 browser workbook always produces the same values_digest -- unless as_artifact is true, in which case it returns a full OCG v0.4 artifact over the WHOLE sheet instead of a single-range fragment (see as_artifact).

Compares a WB-2 Spreadsheet Input Manifest (expected digests) against pasted-back CSV/TSV text per manifest range (observed, e.g. after a recompute in Excel) and returns an XLR-1 round-trip receipt -- `result: "match"|"mismatch"` plus a `mismatches[]` cell list when expected_by_ref text is also supplied. SAME comparator module as the tools/ round-trip page (XLR-2/XLR-3) -- byte-identical receipt for the same inputs. Paste-intake is untrusted: finite-gate (#NUM! for NaN/Infinity) and CSV-injection sanitization apply identically to WB-1's CSV import. Verify-only -- never operates Excel, never ingests .xlsx.

BM25 search over all 416 AINumbers ChainGraph chains. Returns ranked chains with their full recipe: ordered node sequence, deep-links, composer URL, and entry tool mcp_name. Agent flow: find_chain(query) → read recipe → call the listed node MCP tools in order, passing parent_hashes between steps. Task-shaped queries work well: "reserve composition recompute", "FR 2052a classification", "CCP margin replication", "SR 26-2 model validation benchmark", "litigation damages recompute", "evidence pack assembly".

Executes every step of a named chain (list names with find_chain / build_workflow_links) and returns ONE composite artifact whose execution_hash anchors all step outputs. compute:"server"/"auto" (default) runs each kernel-backed step server-side, threading step N's execution_hash into step N+1's parent_hashes; compute:"browser" returns a zero-egress delegation bundle (composer URL + ordered deep-links) to run client-side instead — no data leaves the agent. Supply inputs as a map of step tool_id -> policy_parameters (field names per node manifest / build_chaingraph); a step whose kernel needs inputs you omit is reported per-step (status "input_required"), never failed silently. Steps that are browser-only (gpu:true or no registered kernel) are listed for browser delegation. Deterministic, zero PII, zero payload logging. Verify the result with verify_execution_hash. Runs with anything to explain carry decision_trail: per-step reason codes (gate rule id, escalation rule id, input_required cause), hash-excluded adjacent metadata recomputable from the hash-bound decisions[]. Each server-mode run also returns an OpenTelemetry GenAI span document as a resource link (one execute_tool span per executed step under an invoke_agent parent). Response includes a ledger_url fragment link for human verification at ledger.ainumbers.co. Server-mode responses also echo a stable `dedupe` object — `dedupe.input_hash` (JCS-SHA-256 over the effective run inputs: chain, per-step inputs after the caller/fixture fallback, and mandate_hash when a mandate governs) beside `dedupe.composite_execution_hash` — so a client can recognize and skip an exact re-run (idempotentHint) without recomputing anything. See docs/IDEMPOTENCY.md.

Batch over named chains in ONE round-trip with a PER-ROW terminal status — one row's failure never fails another, and every row's result rides the single response. mode:"run" (default) executes each row through the same engine as run_chain (composite artifact + execution_hash per row, deterministic); mode:"estimate" validates every row (chain exists, step counts, compute feasibility, inputs coverage, OCG §21.4 gate rule previews) WITHOUT executing anything. Capped at 8 rows and 11 total server-kernel steps per batch — caps MEASURED live on the deployed free-plan endpoint, where N kernel builds in one invocation share one CPU budget (an 11-step chain: green at 416 ms). Beyond the cap, use sequential run_chain calls — each call gets its own CPU budget (the blessed fan-out pattern on this tier; docs/client-fan-out-patterns.md). Retry = re-run the WHOLE batch: safe, because determinism reproduces byte-identical execution hashes. Mandates and input_required escalation are not accepted on batch rows; escalated rows carry the non-blocking resolve_handle record. Zero PII, zero payload logging.

Completes ONE step of a checklist/SOP run and returns its OCG v0.4 step receipt: execution_hash chains to prev_step_receipt_digest (pass the previous step's execution_hash, or omit for step 0). A blocking-gate step with evidence_requirement != "none" and no evidence supplied is refused (the caller enforces step order; this tool enforces the evidence requirement per step). Call once per step in order, then pass the full ordered list of returned receipts to checklist_verify_run to check the chain and mint the run receipt.

Validates a checklist or SOP definition JSON against the CHECKRUN-1 schema (definition_id, title, semver version, non-empty steps[] each with step_id/title/instruction/evidence_requirement (none|text|file-digest|attestation)/gate (blocking|advisory)). Returns valid:true/false plus a field-by-field error list. Pair with checklist_step_receipt to run the definition headlessly.

Recomputes and checks a checklist run: every step receipt's execution_hash, the hash-chain link between consecutive steps, and (if a run_receipt is supplied) the §20.1 RFC 6962 Merkle root over every step. Returns valid:true/false, per-step ok/hash_ok/link_ok, and broken_at (the zero-based index of the first broken step, or null). Same recompute a human gets from the browser Run Verifier.

Borrower Escrow Auditor: OpenChainGraph compute node (consumer_finance). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Open at: https://ainumbers.co/chaingraph/732-borrower-escrow-auditor.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Audits a servicer escrow accounting statement from borrower-supplied values only: recomputes the aggregate accounting trial running balance month by month, compares the servicer stated end-of-month balances line by line, flags each discrepant line, checks the stated cushion against the one-sixth-... Outputs: verdict, account_status, periods_analyzed, recomputed_balances, recomputed_low_point, target_low_point, total_annual_disbursements, cushion_cap_amount, stated_cushion, cushion_within_cap, shortage_amount, surplus_amount, shortage_threshold_base, shortage_threshold_base_source, shortage_at_or_above_one_month_payment, shortage_permitted_courses, servicer_course_permitted, surplus_refund_required, surplus_refund_unverified, discrepancies, evidence, manual_review_required. Output schema: call describe_tool("audit_borrower_escrow_stmt").

Audit MCP OAuth 2.1 authorization: validate RFC 9728 protected-resource-metadata, check RFC 8707 audience binding, and assess token-passthrough / confused-deputy risk. Use when a developer is securing an MCP server's authorization. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("audit_mcp_oauth").

OTLP span/trace to receipt bundle; does NOT build a ChainGraph (use build_chaingraph for that). Three modes, chosen by which input is supplied. (1) Pass `trace` alone to GENERATE a receipt bundle: a per-span receipt (span digest, parent-span receipt digest, semconv_snapshot, eddsa-jcs-2022 signature over an ephemeral did:key) for every span, plus a Merkle-rooted (RFC 6962) trace receipt. (2) Pass `bundle` ({trace, span_receipts, trace_receipt, issuer_did}) to VERIFY it: recomputes every digest, walks the parent-receipt chain, rebuilds the Merkle root, and reports which spans are attested, unattested, or tampered. (3) Pass `run_chain_result` (the structuredContent from run_chain) instead of `trace` to bridge a ChainGraph chain run into an OTLP trace first -- one execute_tool span per successfully-executed step, each carrying its own execution_hash as a span attribute (ocg.execution_hash) -- then generates the same receipt bundle over it. Byte-identical receipt engine to tools/566-otel-span-receipt-verifier.html -- a human's browser-generated bundle and this tool's output interoperate on the exact same wire shape.

Structurally validates an OTLP/JSON trace (resourceSpans -> scopeSpans -> spans: hex traceId/spanId shape, quoted nanosecond timestamps, parent-span references, time ordering) and checks every gen_ai span's attributes against a pinned gen_ai semantic-convention attribute snapshot (the upstream open-telemetry/semantic-conventions-genai namespace ships zero tagged releases, so this is a dated pin, stamped as semconv_snapshot in the result -- expect it to drift as the upstream spec moves). Byte-identical lint engine to tools/556-otlp-genai-span-composer-linter.html.

An agent proposes an edit: diffs original text/markdown against its own revised version and returns the line-level hunks plus a diff receipt (original/revised digests, diff-algorithm declaration, per-hunk digests). Byte-identical to what the tools/552 browser workbench computes for the same inputs -- a human can paste the SAME original/revised text into that workbench, disposition each hunk (accept/reject/comment), and their resulting hunk/disposition receipts will reference this diff_receipt's execution_hash. Pass the whole bundle (this diff_receipt + the human's hunk_receipts + disposition_receipt) to redline_verify to check the interleaved chain end-to-end.

Recomputes a redline review from scratch: the diff (from the pasted original+revised text), the diff_receipt's execution_hash, every hunk_receipt's digest/execution_hash/chain-link (in the order supplied -- entries may be minted by an agent, a human, or both interleaved), the disposition_receipt's execution_hash, and the accepted-text digest. Returns valid:true/false, a per-check breakdown, and broken_hunk (the zero-based index of the first divergence, or null) -- never trusts a bundle claim without checking it against a fresh recomputation.

Issues a Selective Disclosure JWT (RFC 9901) from a claims object, marking selected top-level claim keys as selectively disclosable via salted-hash digests in the _sd array. Signed EdDSA over a fresh ephemeral did:key (generated per call, not reused). Pair with sdjwt_present to build a redacted presentation from the returned sd_jwt.

Builds a redacted presentation from an existing SD-JWT (as returned by sdjwt_issue), keeping only the listed disclosures. Pass aud to add a KB-JWT holder-binding (a fresh ephemeral holder key is generated per call). Pass issuer_did (the sd_jwt's "issuer" field) to also verify the JWS and return the verifier_view -- exactly the claim set a relying party would resolve -- plus an OCG receipt of the presentation activity.

Proves (or refutes) that a {path,digest} pair was in a disclosure manifest's room (DATAROOM-1-BUILD-SPEC.md §DR-4) -- worker-side mirror of tools/547-disclosure-manifest-verifier.html's single-file inclusion check. Recomputes the manifest's Merkle root over its entries[], builds the inclusion proof path for the target entry, and reapplies it to confirm it reduces to the claimed merkle_root. Absence is only provable against the exact manifest version passed in.

Independently verify a ChainGraph artifact (ChainGraph Standard v0.1 §6). Recomputes SHA-256 over the canonical (sorted-key, whitespace-stripped) JSON of policy_parameters + output_payload and compares it to the claimed execution_hash. A match proves integrity: policy_parameters and output_payload are byte-for-byte what was hashed. It does not by itself show the outputs were computed from the inputs; for that, re-run the stage and compare (SPEC §4), or check the artifact's compute_proof where one is present (SPEC §18). Pass either a full artifact object, or policy_parameters + output_payload + claimed_hash. Inputs must be I-JSON (RFC 7493): integers beyond 2^53 must be passed as JSON strings, and NaN/Infinity are rejected. Values outside I-JSON have no stable canonical form, so this tool returns a structured -32602 error (error.data.reason "ijson_violation") instead of a hash. Pure client-safe compute -- no data is stored. Use this to verify artifacts from any vendor that conforms to the ChainGraph Standard. Also verifies composite chain artifacts from run_chain (composite_artifact, ChainGraph SPEC §21.2/§21.4): hash-excluded adjacent metadata such as decision_trail, step_compliance_flags, non_live_steps and escalation_record is dropped before recompute, and the response names which rule applied.

Agent-facing mirror of the tools/553 browser workbench (VS-1) -- recomputes the self-addressing identifiers (SAIDs) of a pasted ACDC / vLEI credential JSON (top-level "d" plus any nested a/e/r blocks) under the KERI/CESR Blake3-256 ("E") or SHA2-256 ("I") derivation codes, and cross-checks the declared schema SAID ("s") against a pinned table of 6 official GLEIF vLEI schema SAIDs. Reuses the SAME vendored SAID-check module as T553 -- the same credential produces byte-identical findings in either surface. Returns per-field match/mismatch/unsupported/unknown findings plus a receipt. This is a STRUCTURAL check only -- NOT the KERI chain of trust, NOT issuer authority, NOT revocation state. No network egress.

Simulate agent payment policies for tokenized A2A corridors: set spend caps, MCC allowlists, velocity throttles, and approval thresholds; run synthetic transactions against the policy and export the result as a Policy Mandate. Browser-based, client-side only, zero PII. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("agentic_mandate_sandbox").

Search the AINumbers catalog (1230+ client-side fintech tools). Returns deep-links; prefill-enabled tools accept #in=<base64url(JSON of {element_id: value})>[&run=1] for one-click invocation.

Calls the anchor-suite MCP server (server-to-server, not the browser CORS-gated relay) to timestamp one execution_hash, or batch-anchor an array via anchor-suite's own anchor_batch (one upstream call per batch, RFC 6962 Merkle root anchored, per-hash merkle_inclusion returned), and shapes the result as literal OCG v0.7 section 20 anchor_bindings entries ready to append to an artifact. Success is only reported after this tool independently confirms anchored_hash (or, for a batch entry, merkle_inclusion.leaf) equals the caller-supplied hash -- it does not merely trust anchor-suite's response. On ANY failure -- anchor-suite unreachable, erroring, timing out, or an equality mismatch -- returns {ok:false, unanchored:true, reason} verbatim, never a fabricated binding. Caller-invoked only; not wired into any automatic per-artifact-emission path.

Anchor agentic tool for Cat-12. Translate AML/BSA program controls, TM rules, and customer risk policy into a structured Policy Mandate JSON for agentic payment sy Browser-based, client-side only. Zero PII. Link users to https://ainumbers.co/tools/131-ap2-aml-mandate-builder.html for interactive use. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("ap2_aml_mandate_builder").

Score and compare BaaS providers across 10 capability dimensions (regulatory standing, programme management, card issuance, rails, KYC/KYB, disputes, developer experience, pricing, FDIC pass-through, compliance tooling) with a user-adjustable 1-5 weighting matrix. Outputs a weighted comparison matrix and Markdown evaluation memo. Browser-based, client-side only, zero PII. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("baas_provider_comparator").

Assembles a §27.6 haEvidenceBundle from a subject's collected human_accountability_records[] (reviewers, approvers, annotations, exception rationale, timestamps) -- same algorithm as the browser verify.html HA export card (kernels/_haevidence.mjs). Set sd_jwt:true to also export it as a Selective Disclosure JWT (§13.12), signed EdDSA over a fresh ephemeral did:key (generated per call, never reused) -- subject_hash/verification_result/kernel_version/policy_version/timestamps/submission_receipt are always-disclosed; reviewers/approvers/annotations/exception_rationale/input_hashes are selectively disclosable.

On-Chain Repo Haircut Calculator: OpenChainGraph compute node (collateral_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Output feeds: 505-tokenized-collateral-eligibility-checker, 506-onchain-cash-leg-finality-checker. Open at: https://ainumbers.co/tools/508-repo-haircut-collateral-calculator.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Compute repo haircut, initial margin, and variation margin. Outputs: base_haircut_pct, total_haircut_pct, sft_floor_applied, initial_margin, vm_threshold, canton_haircut_pct, legacy_haircut_pct, weekend_saving_pct, flags. Output schema: call describe_tool("calculate_repo_haircut").

Runs ONE read-only AINumbers tool that is not in your tool list. tools/list is paginated (13 pages, 722 tools) and many hosts read only the first page; this is the door to the rest. Pass { name: "<exact mcp_name>", arguments: { ... } } — the target's own inputSchema is validated and its result is returned verbatim, including execution_hash, so a dispatched call and a direct call are byte-identical. Get a name from find_tool (single calculators), find_chain (workflows) or describe_tool (exact schema). Read-only tools only: anything that issues a credential, stamps an anchor or reaches the network is refused here and must be called directly so your host can approve it.

Parses a camt.053.001 bank-to-customer statement XML document with the same schema-subset structural and facet checks as the tools/565 browser reconciliation workbench (IBAN mod-97, BIC, currency, date/decimal facets), returning the extracted statement (message id, statement id, account IBAN/currency, balances, entries) on success or the structural error list on failure. Byte-identical extraction to the browser tool for the same input. Read-only parse -- feed the result to recon_match for reconciliation.

On-Chain Cash-Leg Finality Checker: OpenChainGraph compute node (attestation_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Consumes upstream artifacts from: 505-tokenized-collateral-eligibility-checker. Open at: https://ainumbers.co/tools/506-onchain-cash-leg-finality-checker.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Validate the cash leg of a Canton settlement: finality model (PFMI P8/P12), reserve attestation, GENIUS Act / MiCA regulatory classification, and depeg circuit-breaker. Outputs: verdict, finality_flag, genius_status, mica_status, compliance_flags, mandate_type. Output schema: call describe_tool("check_cash_leg_finality").

Digital Asset Regulatory Classifier: OpenChainGraph compute node (compliance_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Output feeds: 512-tokenized-security-lifecycle-validator. Open at: https://ainumbers.co/tools/510-digital-asset-regulatory-classifier.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Classify a tokenized asset under GENIUS Act, MiCA, MiFID II, and EU DLT Pilot Regime. Outputs: classification_results, compliance_flags, iso20022_party_identification. Output schema: call describe_tool("classify_digital_asset_regulatory").

Compare agentic payment protocols (AP2, ACP/Shared Payment Token, x402, Visa TAP, Mastercard Agent Pay) across credential, signing, scope, rail, identity, and audit dimensions; optionally recommend a fit for a scenario. Read-only local lookup and comparison — no state change, no keys held, nothing sent or registered anywhere. Use when a developer or strategist needs to orient across the fragmenting agentic-payments standards. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("compare_agentic_payment_protocols").

Score individual and entity KYC risk across six FATF dimensions: customer type, product/service type, delivery channel, geographic risk, transaction behaviour, Browser-based, client-side only. Zero PII. Link users to https://ainumbers.co/tools/110-customer-risk-rating.html for interactive use. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("customer_risk_rating").

Decode an x402 payment header or lint an exact-scheme PaymentPayload, and describe the HTTP-402 verify/settle flow. Use when a developer is integrating x402 and needs to inspect a header, check a payload shape, or understand the flow. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("decode_x402_payment").

Returns the full definition of exactly one AINumbers MCP tool: name, description, inputSchema, outputSchema (the JSON Schema its structuredContent validates against), annotations, and lifecycle_status. Use it after find_tool or find_chain hands you an mcp_name, or after tools/list whose entries omit outputSchema to keep the catalog reply small — the pointer sentence "Output schema: call describe_tool(\"<name>\")." marks exactly those tools. One call, no side effects, zero network on the server: the definition comes from the vendored catalog. Unknown names are rejected with -32602 plus the nearest registered names.

Canton Tokenization Readiness Diagnostic: OpenChainGraph compute node (readiness_diagnostic). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Output feeds: 504-settlement-risk-capital-optimizer. Open at: https://ainumbers.co/tools/503-canton-tokenization-readiness-diagnostic.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Output schema: call describe_tool("diagnose_canton_readiness").

Makes ChainGraph tools agent-callable (ChainGraph Standard v0.1 §3.1). Mode 1 — supply pre_computed_artifact (exported from the browser tool): validates §4 schema fields, recomputes execution_hash via SHA-256 over canonical {policy_parameters, output_payload}, returns verified structuredContent. Mode 2 — supply tool_id + policy_parameters: returns an artifact template envelope and browser prefill URL so an agent can hand the user a pre-filled link; GPU sims always delegate to the browser per §9.2. Mode 3 — supply tool_id only: returns node metadata and artifact schema scaffold. Mode 4 (Compute Binding, v0.4) — supply tool_id + policy_parameters + compute:"server" (or compute:"auto" for gpu:false nodes): runs the registered kernel server-side and returns a verified v0.4 artifact with execution_hash + output_payload in one round-trip. No browser required. gpu:true nodes always delegate to browser. Hashed input must be I-JSON (RFC 7493): integers beyond 2^53 must be passed as JSON strings, and NaN/Infinity are rejected — values outside I-JSON have no stable canonical form, so a structured -32602 error (error.data.reason "ijson_violation") is returned instead of an execution_hash. readOnlyHint: true. Zero PII, zero payload logging. Pair with verify_execution_hash (independent hash verification) and build_chaingraph (DAG wiring).

Render a verified OpenChainGraph v0.4 artifact into a chaingraph_export profile (OCG Standard §13). Generated downstream of and EXCLUDED from the execution_hash preimage — the export is a view, not a fact; verification always routes back to the canonical JSON artifact. Pass the FULL artifact you received from a compute tool (the server is stateless — there is no hash cache). Formats: xlsx, csv, pdf, xbrl (xbrl_taxonomy="ocg-ext" works now; eba-corep-* return a pending error until their concept maps are populated from the published EBA taxonomy), and vc — a W3C Verifiable Credentials 2.0 rendering (OCG §13.11, application/vc+json) available on every node; it re-states the canonical execution_hash via ocg:hashAnchor and mints no new hash/proof. readOnlyHint: true; zero PII, zero payload logging.

Given a step's haGatePolicy and the human_accountability_records[] collected for a subject, returns whether the gate is satisfied, held, rejected, escalated, or overridden -- same algorithm as the browser verify.html HA gate card (kernels/_hagate.mjs, SPEC.md §27.4-27.5). A rejection record is terminal-blocking regardless of policy; an active time-boxed override (§27.5) takes precedence over the underlying policy. Pure function: caller supplies nowISO for determinism.

Inspect a Visa Trusted Agent Protocol HTTP Message Signature and score TAP readiness. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("inspect_visa_tap_signature").

Wraps the result of a run_chain call (server/auto compute mode) as standard in-toto links, one per successfully-executed step, plus a generated in-toto layout matching the chain's linear topology (each step MATCHes the previous step's product). Materials/products are keyed by the step's execution_hash; byproducts carries the raw execution_hash. All links and the layout are DSSE-signed with one ephemeral Ed25519 keypair scoped to this call (never persisted) — first shipping instance of "in-toto for MCP" per the 2026 research gap survey. Verify the bundle offline with the in-toto Link Builder & Verifier (chaingraph/intoto-link-builder.html verify tab) or a reference implementation. Steps that did not run (input_required, skipped_by_gate, gpu_browser_only, etc.) are listed in skipped[] rather than silently omitted.

Composes and signs a W3C Verifiable Credential 2.0 with an eddsa-jcs-2022 Data Integrity proof over a fresh ephemeral did:key (generated per call, not reused). Returns the signed credential, an OCG Standard §23 vc-2.0 input-attestation block ready to embed in a ChainGraph chain's policy_parameters at the given pointer, and an OCG receipt of the issuance activity. The signature proves the claims were not altered after signing and that the did:key holder produced it -- it is not, by itself, identity assurance.

Run a ChainGraph decision kernel's compute(policy_parameters) inside a sandboxed, deterministic, in-browser QuickJS-ng WebAssembly VM (ocg-deterministic-compute@2) and return its output_payload. Demo kernel set only -- for the full catalog, use the worker's compute kernels directly. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("run_kernel_vm").

Fetches an entity's live GLEIF LEI record (api.gleif.org -- the worker's only egress target for this tool, RULINGS-2026-07-19-EGRESS-APEX.md R1) and grades it across six GLEIF data-quality dimensions -- registration status, renewal timeliness, corroboration level, entity status, Level-2 parent disclosure, address completeness -- using the SAME deterministic grader as the tools/551 browser workbench (pasting the same record there reproduces identical grades). Returns per-dimension grades, a composite, and a receipt with source provenance (url, retrieved_at, response_digest). This is an ASSERTED FETCH, not a zkTLS proof of origin, and grades DATA QUALITY only -- never creditworthiness, sanctions status, or entity legitimacy.

Validate an MCP tool definition against JSON Schema 2020-12 and current naming, output-schema, and annotation rules; returns findings, a conformance score, and a recommended annotation set. Use when a developer wants to check an MCP tool definition before publishing. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("lint_mcp_tool_definition").

Margin Call Collateral Mobilizer: OpenChainGraph compute node (collateral_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Consumes upstream artifacts from: 505-tokenized-collateral-eligibility-checker. Output feeds: 506-onchain-cash-leg-finality-checker. Open at: https://ainumbers.co/tools/513-margin-call-collateral-mobilizer.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Compute initial and variation margin requirements and identify eligible collateral to mobilize. Outputs: collateral_detail, gap, im_call, shortfall, total_mobilizable, total_required, vm_call. Output schema: call describe_tool("mobilize_margin_collateral").

Settlement-Risk Capital Efficiency Optimizer: OpenChainGraph compute node (capital_assessment). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Consumes upstream artifacts from: 503-canton-tokenization-readiness-diagnostic. Open at: https://ainumbers.co/tools/504-settlement-risk-capital-optimizer.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Quantify RWA and capital saving from Canton atomic DvP vs legacy settlement. Outputs: cet1_ratio, cost_of_capital, portfolio_bps, rows, total_annual_saving, total_capital_freed, total_notional_usd, total_rwa_delta_usd, verdict. Output schema: call describe_tool("optimize_settlement_capital").

Validates a pain.001.001.09 customer credit transfer initiation XML document against the same hand-derived schema-subset table as the tools/555 browser validator -- structural (mandatory fields), facet (IBAN mod-97, BIC format, ISO 4217 currency, decimal/date patterns), and batch-total cross-checks (NbOfTxs vs. actual transaction count, CtrlSum vs. sum of InstdAmt). Byte-identical error set to the browser tool for the same input. Subset validation, not full XSD conformance -- prepare/validate only, never transmits or generates a live payment message.

Parses a camt.053.001 statement (same schema-subset checks as camt053_parse) and a counterpart expectation set (CSV, header EndToEndId,Amount,Currency,Date -- same strict RFC 4180 parser as the WORKBOOK-1 CSV tools), then runs the SAME deterministic match engine as the tools/565 browser workbench: EndToEndId exact match first (statement order), then amount+date tolerance + currency match (first unmatched expectation in CSV order). Returns matches, exceptions (unmatched entries/expectations), and a reconciliation receipt (statement digest, expectation-set digest, match-rule declaration, counts, exception digests, execution_hash) -- byte-identical to the browser tool for the same inputs. Prepare/hash/receipt only; per-exception disposition receipts are the browser workbench's interactive follow-on.

Structural (non-cryptographic) §27.2 signed-named-human check: does this record carry a §16 whole-artifact proof (eddsa-jcs-2022) whose verificationMethod is bound to the record's own identity.id? Does NOT verify the signature bytes -- pair with verify_execution_hash / a proof verifier for that. Use before counting a record toward ha_gate_status.

Scan an MCP tool description/manifest for tool-poisoning and prompt-injection smells; returns a risk score and flagged patterns. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("scan_tool_poisoning").

Compute a composite MCP server ship-readiness score across tool definitions, server.json, OAuth, transport, tool poisoning, and spec compliance. Renders the interactive AINumbers tool as a widget; inputs are applied via the AIN Bridge and the tool runs client-side (zero PII, zero network). Output schema: call describe_tool("score_mcp_readiness").

Builds a prefilled GitHub Issue-Forms URL for proposing a new AINumbers fintech tool, OpenChainGraph compute node, or verifiable workflow. Returns the URL only -- it never posts to GitHub on the caller's behalf. Companion to the site's mailto suggestion form (suggest.html); this is the agent-callable path.

How to chain AINumbers tools into audited workflows: call with NO recipe_id for the compact recipe index, or with one recipe_id for that workflow's full step-by-step recipe. Every recipe carries the suite conventions end to end: Policy Mandate exports, execution_hash verification, and receipts.

Tokenized Collateral Eligibility Checker: OpenChainGraph compute node (collateral_mandate). Deterministic OpenChainGraph compute node. By default (compute:"auto") inputs are computed server-side on Cloudflare Workers for gpu:false nodes with a registered kernel; compute:"browser" forces client-side execution and returns a browser delegation URL instead. gpu:true nodes always delegate to the browser. Inputs are processed transiently to compute the response and are not stored, logged, or retained. Use synthetic or anonymised inputs only. Exports an AP2 artifact with execution_hash for chain provenance. Output feeds: 506-onchain-cash-leg-finality-checker, 513-margin-call-collateral-mobilizer, 514-tokenized-fund-collateral-validator. Open at: https://ainumbers.co/tools/505-tokenized-collateral-eligibility-checker.html FV-status (published/proven/still-trusted for this spec): /fv-status/ca4e1dca8343f788a296161f0e19712f7afb7e5eaec0431081e715a4cc34b784.json — a snapshot, not a subscription; this receipt verifies offline regardless of whether that file is ever fetched. Summary: Classify tokenized assets for DTC/Fed eligibility, Basel HQLA tier, and transfer restrictions. Outputs: dtc_status, hqla_tier, final_haircut_pct, adjusted_value, compliance_flags, mandate_type. Output schema: call describe_tool("check_tokenized_collateral_eligibility").

BM25 search over all 702 live AINumbers ChainGraph node tools. Returns ranked tools with mcp_name, URL, mandate type, and wave. Use to locate a specific computation node (e.g. "FRTB expected shortfall", "MiCA own funds", "XVA calculator", "reserve composition", "2052a", "initial margin", "model validation benchmark", "TVM / damages calculator") before calling it. Complements find_chain (chain-level) and list_ainumbers_tools (catalog-level).