AI Evidence Integrity Checker

Created by Filip Majchrzak.

AI Evidence Integrity Checker is a small companion Actor for teams that want a fast sanity check before evidence is used by an AI workflow, release gate, review process, or handoff.

It checks:

• required evidence files are present;
• evidence paths are unique and safe;
• timestamps are explicit UTC values and not in the future;
• evidence is not older than the configured age limit;
• .json evidence contains valid JSON;
• optional expected SHA-256 values match the exact UTF-8 content bytes;
• evidence content is not empty.

The result is deliberately simple: PASS, FAIL, or INVALID_INPUT, with reason codes and computed SHA-256 digests.

Visual walkthrough

How it works

PASS example

FAIL example

Why this Actor exists

This is the lightweight companion to AI Claim Readiness Checkpoint. The Integrity Checker asks: “Are these evidence files structurally usable and internally bound?” The Claim Readiness Checkpoint asks the broader question: “Does the evidence satisfy the release policy well enough for a decision?”

Public project/demo context: AI Claim Readiness demo.

Example input

{
  "evaluationTime": "2026-08-23T00:00:00Z",
  "maxAgeDays": 30,
  "requiredPaths": ["tests.json", "review.json"],
  "evidenceFiles": [
    {
      "path": "tests.json",
      "observedAt": "2026-08-22T20:00:00Z",
      "content": "{\"status\":\"PASS\",\"passed\":24,\"failed\":0}"
    },
    {
      "path": "review.json",
      "observedAt": "2026-08-22T20:30:00Z",
      "content": "{\"status\":\"PASS\",\"reviewer\":\"synthetic-demo\"}"
    }
  ]
}

Output

The default Dataset and OUTPUT key-value record contain the same bounded summary. Each evidence item gets its exact UTF-8 SHA-256, byte count, timestamp, and JSON-validity flag where applicable. Failures include stable issue codes such as MISSING_REQUIRED_EVIDENCE, FUTURE_EVIDENCE, STALE_EVIDENCE, INVALID_JSON, DUPLICATE_EVIDENCE_PATH, and SHA256_MISMATCH.

Safety and claim ceiling

Use synthetic or non-sensitive evidence unless your own data-handling rules and Apify settings permit otherwise. This Actor checks structural integrity and bindings only. It does not prove truth, safety, security, compliance, authorship, or production readiness, and it does not execute an external action.

Product relationship

• Lightweight checker: AI Evidence Integrity Checker
• Full decision gate: AI Claim Readiness Checkpoint
• Creator: Filip Majchrzak