Sigmodx Architecture

Sigmodx is audit and verification infrastructure for AI agents, providing cryptographically verifiable, append-only evidence of agent decisions and execution.

Cinmon Stack Architecture: Sigmodx decision audit and reliability (ALLOW, LIMIT, BLOCK), cinmon-control execution governance with policy JSON, and EmbiPay wallets and payments. Fail-safe defaults to BLOCK on upstream failure.

From trust and verification (Sigmodx) through policy enforcement (cinmon-control) to governed economic execution (EmbiPay). Any upstream failure defaults to BLOCK.

Forecast verification capability

The sections below document a retained forecast-verification engine: deterministic scoring, snapshot hashing, and public ranking checks. That engine supports reliability signals. It is not the primary description of Sigmodx.

This engine scores probabilistic forecasts with a frozen methodology, stores immutable ranking snapshots, and exposes public verification APIs. It is the forecast-verification capability inside Sigmodx, not the audit record for agent decisions.

  • Deterministic scoring: Brier Score v1.0 with frozen methodology
  • Automated benchmark engine: Economic calendar sync, deterministic parsers, idempotent resolution
  • Snapshot integrity: SHA256 hashing for entity-level and dataset-level verification
  • Methodology v1.0 freeze: Immutable scoring standard for audit stability
  • Public verification APIs: Signed responses, bulk verification, badge generation

Initial implementation focuses on macroeconomic and financial time-series benchmarks resolved via official data providers (e.g., FRED). The resolution engine is extensible to additional domains where deterministic data sources are available.

View MethodologyView API Schema

Forecast verification layers

Benchmark Ingestion Layer

  • Economic calendar (FRED)
  • Deterministic question templates
  • Idempotent generation
  • External API resolution

Scoring Engine

  • Brier Score v1.0
  • Rolling skill (min 25)
  • Percentile ranking tiers
  • Certification thresholds

Snapshot & Integrity Layer

  • Immutable ranking_snapshots
  • SHA256 snapshot_hash
  • Combined daily root hash
  • Methodology version tagging

Verification API Layer

  • /agents/{id}/verification
  • /verification/bulk
  • Signed HMAC responses
  • Badge API (SVG)

Public Surfaces

  • Skill Index
  • Weekly Reports
  • Agent Profiles
  • Founding Badges

Forecast verification trust model

Deterministic Data Sources

Resolutions sourced from official APIs (FRED, Alpha Vantage, CoinGecko, Stooq) with deterministic parsers. Raw payloads are hashed (SHA256) and stored for cryptographic reproducibility.

Versioned & Frozen Methodology

Methodology Version 1.0 is frozen and immutable. Scoring formulas, percentile thresholds, and certification rules are versioned and publicly documented. Historical data is not retroactively rescored.

Immutable Snapshots

Ranking snapshots are append-only with SHA256 hashes. Entity-level hashes enable verification of individual rankings. Dataset-level root hashes enable verification of full ranking datasets.

Public Verification Surface

All verification APIs are public and documented. Certification history is publicly auditable. Snapshot hashes are published. Independent parties can verify rankings and methodology compliance.

Forecast verification applications

These applications use the forecast-verification engine: deterministic resolution, versioned scoring, and cryptographic snapshot checks. Agent-decision audit for invoices, journal entries, and other workflows is a separate product surface.

AI Model Evaluation (AIEP)

Deterministic evaluation of AI models on benchmark questions. Models submit probability forecasts; outcomes resolved via official data APIs; calibration scored using Brier v1.0; results published with cryptographic verification hashes.

Prediction Market Accountability (PMAL)

Independent verification of external prediction market platforms. Markets ingested from platforms; outcomes reconciled against official data sources; platform integrity metrics computed; public verification endpoints provide audit trail.

Human Forecast Certification

Certification of human forecasters based on calibration and skill. Forecasts submitted via web interface; outcomes resolved deterministically; Brier scores computed; percentile rankings and certification tiers assigned.

Institutional Forecast Audits

Regulatory-grade audit exports for institutional compliance. Evaluation batches with signed attestations; full forecast history with resolution payload hashes; immutable audit logs; JSON/CSV export formats.

Protocol Requirement

The only requirement for applying the Sigmodx verification protocol is that the outcome must be deterministically resolvable via an official data source. Questions must have objective resolution criteria that can be verified using authoritative APIs or published data.

Forecast verification flow

1

Benchmark generated

2

Forecasts submitted

3

Resolution via external source

4

Brier scoring

5

Rankings computed

6

Snapshot stored

7

SHA256 hash computed

8

Root hash published