Model risk / EUC inventory
EUC register.
Every live tool on AINumbers.co is a spreadsheet a compliance officer would otherwise have to inventory by hand: name, kernel version and hash, declared inputs and outputs, a control description, a trust label, a data vintage, and a last-validated date. This page generates that register entry from the tool's own published metadata, not from a separately maintained log, so it can never drift from what is actually deployed.
Execution proofs cover the computation shown; they do not verify the regulatory interpretation beside them.
Banking & capital reporting
The BANKING-OCG program kernels: FR Y-14 roll-forward, IRRBB shock calculators, collateral haircut, credit concentration, counterparty limits, and securitization risk retention.
Rolls forward a caller-declared FR Y-14A/Q capital worksheet (CET1, additional Tier 1, Tier 2) from beginning balance through period additions and deductions to an ending balance, applies a caller-declared published-scenario adjustment (e.g. a Federal Reserve DFAST/CCAR severely-adverse published-scenario delta), and cross-checks the computed ending total capital against a caller-declared reported total-capital figure sourced from another schedule (e.g. FR Y-9C Schedule HC-R, art-436) within a caller-declared tolerance. Not a filer, produces evidence artifacts and worksheet totals only, never a submission. All roll-forward line items, the scenario adjustment amount, and the cross-check reference figure are caller-declared; this tool performs only roll-forward arithmetic (beginning + additions - deductions + scenario adjustment = ending) and a tolerance comparison, never capital-component classification, scenario modeling, or projection, firm capital-planning models (PPNR, loss forecasts, scenario translation into balance-sheet impact) stay strictly outside this boundary. Complements art-436 (BHC Schedule HC-R capital) as an independent roll-forward/cross-check, not a replacement for it.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:c6c4a22b5331...
Basel CRE22 comprehensive-approach collateral haircut engine for counterparty credit risk: applies a caller-supplied, versioned supervisory haircut table (policy input, not hardcoded) to each collateral item's asset class/maturity bucket, scales for a non-standard holding period via the CRE22.68 square-root-of-time rule, adds an FX-mismatch haircut where collateral currency differs from exposure currency, and computes net exposure E* = max(0, E*(1+He) - sum(C*(1-Hc-Hfx))). An item haircut override without a reason_code is flagged, the item-level basis for a separate signed §27 human_accountability_record, not minted by this kernel. An unmatched asset_class/maturity_bucket defaults to a conservative 100% haircut, flagged, never silently valued. Deterministic per-item haircut application and summation only, no collateral-to-exposure allocation/optimization solver. Not calculate_repo_haircut (508, Canton 24/7 timing-gap-specific SFT calculator) or compute_stock_token_collateral_haircut (art-320, RHC liquidation-risk layering), this is the generic Basel comprehensive-approach net-exposure engine across asset classes and currencies. Not a capital-return filing tool, evidence artifact only, never regulator-submittable.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:1d708d8172b5...
Credit-concentration screen over a flat exposure list (name, sector, amount): returns the top-N single-name exposures by amount, a per-sector rollup, single-name and sector Herfindahl-Hirschman Index (0-10000 scale), and a breach list against caller-declared single-name and sector limit percentages. Concentration limits are your own institution's risk-appetite policy, not a fixed regulatory threshold, so nothing is baked in. Distinct from the shipped IRRBB/NII shock kernels (art-183/art-185/art-369), which measure rate-risk exposure rather than name/sector diversification. NaN-safe. Zero network, zero PII.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:2c48c8919701...
Counterparty internal credit-limit check: compares each counterparty's caller-supplied current exposure against its board-approved internal limit line (settlement, pre-settlement/PFE, or aggregate limit type), computes utilization percent and headroom, and flags each counterparty WITHIN_LIMIT, WARNING (above the caller-set soft-warning threshold, e.g. 90% of the limit), or BREACH (exposure exceeds the approved limit). This is a deterministic point-in-time check against internally governed limit lines, distinct from the Basel/Reg-YY regulatory single-counterparty threshold check (art-425), and is NOT a real-time exposure monitor: no live feed, no intraday polling, no scheduled job. Not X: use art-425 for the Basel III / Regulation YY 25%/15%-of-Tier-1 regulatory large-exposures limit; use this node for internal (board- or risk-committee-approved) counterparty limit-line governance.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:235144f13ce1...
Comptroller's Handbook IRR basis-risk delta-NII calculator: sweeps a single reference-rate shock across multiple priced indices (Prime, SOFR, Fed Funds, CD portfolio, etc.), each with a caller-declared historical beta vs the reference rate, and isolates the incremental delta-NII from indices not moving in lockstep. Distinct from art-369 (Rate Shock Ladder Replay), whose parallel-curve convention assumes one shock moves the entire gap schedule uniformly and cannot see basis risk. Complements art-369 and art-442 as the third leg of a full NII/EVE-shock toolkit. NaN-safe. Zero network, zero PII.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:6a53b2b9e42a...
EU Securitisation Regulation (EU) 2017/2402 Art.6 and U.S. Credit Risk Retention Rule (Dodd-Frank Sec.941, Reg RR, 12 CFR Part 244) 5% risk-retention checker: verifies the originator/sponsor/original-lender retains at least 5% net economic interest via a recognized method (vertical slice, horizontal first-loss, L-shaped, representative sample, seller's interest), applies the U.S. Qualified Residential Mortgage (QRM) exemption where flagged, and flags EU-specific structural breaches, retainer sole-purpose-entity prohibition and retained-interest hedging/sale. Deterministic point-in-time structural check from caller-supplied retention method, exposure/retained amounts, and jurisdiction-specific flags for a single reporting date. Not a compute_rbc_action_level (NAIC insurer capital ladder) or calculate_solvency2_scr_ratio (EU insurer capital) check, this is bank/ABS securitization risk-retention structure, not insurer solvency. QRM is the only exemption modelled. The other U.S. exemptions have no inputs and sit outside scope: the qualifying commercial loan, commercial real estate loan, and automobile loan exemptions of 12 CFR 244.14 through 244.18; the general exemptions of 244.19, covering United States government-backed, agricultural, state and municipal, qualified scholarship funding bond, pass-through resecuritization, FDIC-sponsored, student loan, community-focused lending, and three-to-four unit mortgage transactions; the foreign-related safe harbor of 244.20; and the additional exemptions of 244.21. Separately, 244.3(a) lets a sponsor satisfy the base retention requirement through any one of 244.4 through 244.10, so the routes given to eligible ABCP conduits by 244.6, to open market CLOs by 244.9(b) and (c)(1), where the lead arranger of each CLO-eligible loan tranche retains 5 percent of its face amount, and to qualified tender option bonds by 244.10 are alternative ways to satisfy the requirement rather than exemptions from it. None of them is modelled here. The EU non-performing-exposure derogation of Art 6(3a), which measures retention against the net value of the exposures after the purchase-price discount, is not modelled either. A structure covered by one of these, a CLO for instance, is read as an ordinary securitization: if its retained interest falls below 5 percent the verdict returned is NONCOMPLIANT rather than not-applicable.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:7c47eb666218...
Sample across other categories
A fixed cross-category sample proving the generator is not banking-only. Every one of the 660 live nodes has its own entry under chaingraph/register/<tool_id>.register.json, listed in index.json.
Validates AP2 v0.2 Intent→Cart→Payment mandate trio: signature-chain integrity, scope/limit consistency, TTL/expiry, over-spend detection, Human-Not-Present autonomous-agent flows. Publishes conformance test-vector fixtures.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:623495f9378c...
Six-domain PSD3/PSR readiness rubric: Open Finance access rights (Art.35/36), TPP categorisation (PISP/AISP/PIISP), SCA exemption alignment (Art.85–90), consent framework maturity, fraud liability model (Art.59–65), and embedded finance/BaaS scope. Radar chart + prioritised gap table. Root node (no upstream dependency). Feeds ART-04 (DORA Mapper) and PTG-01. EU transposition ~2027; UK PSR enacted 2024.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:3826069610c6...
12-question weighted diagnostic across six readiness domains for Canton Network pilots: settlement ops, custody, cash-leg, privacy, AML/KYA, and capital. Routes to the correct workflow chain based on gap scores.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:da4585f3af87...
12-question A–F readiness diagnostic for digital trade / electronic trade documents (MLETR). Grades corridor legality, document digitisation, platform connectivity, trade-rule basis (eUCP/URDTT), financing, and AML/TBML controls; routes to the right chain and emits a remediation checklist.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:77e46eb235da...
12-question A-F diagnostic that classifies which carbon/climate obligations bind a firm (CBAM authorised-declarant duty, EU Taxonomy alignment, EU Green Bond conformance, climate stress) and routes to the right carbon-compliance chain. Separates in-force CBAM definitive liability (since 1 Jan 2026) from prepare-ahead items (first declaration 30 Sep 2027, downstream scope 1 Jan 2028).
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:a8f0b6c702c3...
12-question A-F diagnostic scoping a crypto-asset service provider's MiCA Title-V lifecycle readiness (authorization, Art 67 own-funds, whitepaper, MAR-crypto, travel rule) and routing to the right MiCA chain. Config-only; ART/EMT-issuer cases route to existing stablecoin chains.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:f3a4a041f414...
Reconstruct the RFC 9421 signature base and verify the Ed25519 Web Bot Auth signature against a caller-supplied public key, zero network. Checks alg=ed25519, tag=web-bot-auth, and freshness. Feeds the signatures-directory validator (art-130).
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:540c9d7518ba...
Validate the required-field subset of an EMIR Refit ISO 20022 auth.030 derivative trade report: action type, both counterparty LEIs (20-char ISO 17442), UTI, UPI, notional, currency, effective date, and asset class. Catches missing or malformed fields before submission to the Trade Repository. Feeds UTI completeness checker (art-154).
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:4a915e240b0b...
Classify insurance contracts to their IFRS 17 measurement model: Premium Allocation Approach (PAA) for coverage periods of 12 months or less, Variable Fee Approach (VFA) for direct-participating contracts with investment-linked features (not reinsurance), or General Measurement Model/BBA as the default. Returns measurement_model (GMM/VFA/PAA), eligible_models, paa_eligible, vfa_eligible flags, and direct_participating indicator. Root node of the ifrs17-measurement-conformance chain. IFRS 17 live since Jan 1 2023. Zero network, zero PII.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:681001a8db77...
Computes prediction market PnL, implied probability, break-even, no-vig fair value, expected value, Kelly stake, and odds conversion for binary and scalar contracts. Covers Polymarket, Kalshi, CME Event, and Robinhood. Includes Brier and log-score forecast accuracy metrics. Not financial advice.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:3db60f420ca0...
Builds an EU AI Act Art 12(2)-conformant decision log record for high-risk AI systems in financial services. Computes chain_position (first/chained), art12_completeness_score (12 required fields), retention_months (>= 6 months, configurable), and anchor_surface instructions for composing anchor.ainumbers.co/mcp. subject_ref is a STRUCTURAL field only (opaque reference, never a real natural-person identifier; no PII enters this kernel). When the caller declares which sealed decision artifact this record evidences, this node also wraps it as a section-27.6 evidence bundle over that subject. Disambiguates from build_session_receipt (cry-01): that node logs MCP session I/O; this node builds a regulatory Art 12 decision record. Run classify_annex3_decisioning_obligations (art-238) first to confirm is_high_risk before generating Art 12 records.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:ddb7da11f556...
Classifies ISO 20022 camt.053 BkTxCd entries by Domain, Family, and SubFamily per the CGI-MP camt.053 Usage Guide v5.0 and the ISO 20022 ExternalBankTransactionCode1Code registry 2023-03. Validates the OPBD + sum(movements) = CLBD balance equation. Scores structured-remittance match rate. Emits reconciliation_status (CLEAN / PARTIAL_MATCH / LOW_MATCH_RATE / FAILED_BALANCE), match_rate_pct, and domain_buckets[]. Used in corporate TMS straight-through reconciliation. ZERO PII: no account-holder names or identifiers enter this kernel.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:483a887df68a...
Lints an extracted monthly reserve disclosure against GENIUS Act S.394 §4: composition-category eligibility, tenor, custody locations, a dual_control(2) CEO/CFO certification gate (distinct CEO and CFO identities required, per FDIC NPR 2026-04-10 and OCG SPEC.md §27.3), registered-examiner presence, month-over-month diff, and an on-chain supply cross-check against a pasted figure. Successor to the pre-issuance precheck_reserve_attestation (art-06): that tool is the pre-issuance readiness gate, this is the recurring post-issuance monthly filing check. Never claims cryptographic verification of the source PDF.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:0f28358a1106...
Checks a receipt's optional W3C BitstringStatusList credentialStatus reference and reads the revocation bit at statusListIndex from a supplied, zero-egress status list credential. A set bit means revoked, and revocation overrides the receipt's own signature validity even when that signature is cryptographically valid. Absence of a credentialStatus reference is its own no-signal state, never treated as evidence of active status. Not-X-use-Y: this kernel checks revocation status only, it does not verify the underlying §16 signature itself.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:1bba28fe1629...
Reconciles Robinhood Chain stock-token corporate actions against the ERC-8056 scaled UI amount surface. Stock tokens never rebase; splits and dividends land as a uiMultiplier() change plus a UIMultiplierUpdated event while raw balanceOf stays static until redemption. Checks declared corporate-action ratio against the multiplier transition, monotonic event sequencing, and raw-balance invariance. First tooling anywhere for ERC-8056 reconciliation. Zero network, zero PII.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:429e97740306...
Front-end (housing) and back-end (total) debt-to-income ratios per Fannie Mae Selling Guide B3-6-02 and Freddie Mac Single-Family Seller/Servicer Guide 5401.2. Classifies the back-end ratio into a standard-manual / extended-manual-compensating-factors / DU-LPA-only / exceeds-max tier and flags whether the loan is within the max DTI for the selected underwriting type (DU, LPA, or manual). Feeds art-222-agency-eligibility-matrix as one of its DTI inputs. Not check_agency_eligibility_matrix itself, which performs the full multi-check eligibility decision.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:2e686160add3...
US OCC/FDIC interest-rate-risk parallel shock ladder: sweeps four prescribed parallel magnitudes (+/-100/200/300/400bp) over a bucketed repricing-gap schedule, returning both a duration-weighted delta-EVE leg and a 12-month cumulative-gap delta-NII leg per shock. Optional non-parallel steepener/flattener presets are caller-declared (short/long tenor bps split), not baked-in regulatory scalars. Distinct from the shipped BCBS d368 / EBA standardised six-scenario convention (art-183/art-185): sweeps multiple magnitudes rather than one, and combines EVE+NII in a single kernel. Complements CC-A repricing-gap schedules or user-supplied gap tables. NaN-safe. Zero network, zero PII.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:225cfd55e3d8...
Assembles a shareable evidence bundle around an artifact and stamps the SPEC.md §SIDECAR.1 tiered label it qualifies for: OCG-Verify (envelope well-formed, execution_hash recomputes), OCG-Execute (additionally §21 chain-execution and §22 mandate gates hold), OCG-Prove (additionally a §18 compute-integrity proof verifies). Tiers are cumulative, any gate false at a level makes every level above it unavailable. The label adds no new gate and mints no new trust claim: it re-expresses existing gate-pass results the caller declares for the referenced artifact, and this node never re-runs those gates itself. When the caller supplies collected section-27.2 human-accountability records, they consume/emit as a section-27.6 evidence bundle keyed to the same artifact hash, so the tier label and the accountability trail ride one diffable object. Zero network, zero PII.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:005c23e5672a...
Hashes a third-party eval log (e.g. an Inspect AI transcript) and binds it into a receipt that a compiled Work Mandate (art-274) can reference. VERIFY-ONLY: never executes, re-runs, or re-scores the eval, composes shipped §4 hash, §16 signature, and §20 anchor carriers around a digest the caller already produced. claim_strength is the weakest-link status across the eval log hash and the mandate reference, never inflated by one strong leg covering a missing other.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:4dd15343a439...
Computes the OECD Pillar Two substance-based income exclusion (SBIE) for a jurisdiction from a caller-declared payroll-cost figure, tangible-asset carrying value, and a versioned transition-year rate table (payroll % + tangible-asset % looked up by target year, table supplied whole as a policy input, not hardcoded). Derives excess profit (jurisdictional GloBE income less SBIE, floored at zero), the resulting top-up tax from a caller-supplied top-up-tax percentage, and the final jurisdictional top-up after a QDMTT-paid offset, flagging any QDMTT over-collection informationally. Consumes art-454's jurisdictional GloBE income + top-up-tax-percentage output shape directly and does not recompute an ETR itself, so it also runs standalone. Election choices and GloBE-income adjustments are human judgment and stay upstream. NaN-safe. Zero network, zero PII.
independently verified: zkVM execution proof (risc0/groth16-bn254)
kernel digest sha256:f883e9af5be1...
Trust labels and last-validated dates are derived from each node's compute_proof_ready status and pinned compute-image dates at generation time: they are never hand-set. A node moving from a deferred proof to a ready one is picked up automatically the next time node scripts/gen-euc-register.mjs runs; nothing here is hardcoded against a future proving pass.