Evidence Before Automation

Turn messy information into inspectable, defensible decisions before automating anything consequential.

This public repository contains a methodology, generic schema, and clearly synthetic interface demonstration. It explains the public idea without publishing private operating logic, customer data, production connectors, or credentials.

No consequential automation may exceed the evidence, authority, and trust envelope supporting it.

In plain language

AI can produce a confident-sounding answer even when evidence is missing, stale, or contradictory. This project starts one step earlier: show what information supports each claim, what disagrees, what remains unknown, and which human is allowed to decide.

The included demo is a made-up equipment case. It concludes only that a packet may be prepared for human review. It does not make a legal, financial, insurance, appraisal, medical, regulatory, or payment decision.

Run the demo

Open index.html directly in a browser. No installation, account, network call, API key, or model is required.

Verify the repository

Node.js 20 or newer is required for the checks:

npm ci
npm test

The checks verify the public/private boundary, scan for recognizable secrets and sensitive file types, validate JSON, prevent unclassified files, confirm synthetic labeling, preserve the human-authority gate, and reject unsafe dynamic HTML rendering.

What is here

• METHODOLOGY.md — the ten-part evidence model;
• DECISION-TREE.md — a generic readiness flow;
• schemas/claim.schema.json — a public contract for atomic claims;
• examples/synthetic-case.json — non-real example data;
• index.html, styles.css, app.js — the educational interface;
• docs/ARCHITECTURE.md — implemented scope and anti-drift design;
• docs/PUBLIC-PRIVATE-BOUNDARY.md — what may and may not be published;
• docs/BUILD-REPORT.md — layman and technical record of the initial build;
• PUBLIC-MANIFEST.json and scripts/ — deterministic publication guardrails.

Public versus private

This repository publishes the method, not the private machinery. The private core belongs in a separate private repository with a separate Git history, access policy, and release process. See docs/PUBLIC-PRIVATE-BOUNDARY.md.

The private build prompts are governed separately with stable identifiers, versions, explicit input/output/safety/stop contracts, content locks, and deterministic checks. Their bodies, registries, lock hashes, evaluator rules, and execution tooling are intentionally absent from this public repository.

Current maturity

This is a public methodology and static synthetic demo, not a production evidence system. It does not retrieve sources, run AI reviewers, monitor drift, calculate real outcomes, or perform external actions. The interface illustrates the intended reasoning discipline; it must not be read as proof that those production capabilities exist here.

Vocabulary

• Evidence Kernel — common provenance and decision primitives.
• Domain Pack — isolated knowledge and configuration for one bounded domain.
• Decision Contract — required evidence, permitted transformations, outputs, and authority.
• Trust Envelope — conditions under which a conclusion is allowed to remain valid.
• Review Quorum — independent and adversarial checks before consequential output.
• Knowledge Quarantine — unverified knowledge cannot enter trusted operation.
• KBOM — an inventory of claims, sources, rules, formulas, assumptions, and conflicts.

License and security

The public materials are available under the MIT License. For vulnerability reporting and public-release controls, see SECURITY.md.