From narratives to verifiable proof bundles

How RPO works deterministic structure, public integrity anchors

The Sandbox is deliberately strict: no AI model, no interpretation, no “storytelling”. It performs a deterministic extraction of basic anchors (time, place, actions), generates a draft rpo.json, and exposes a public sha256 proof hash.

Important: the hash proves integrity, not truth. This is a format demo, not legal advice.

Open the Sandbox Back to overview

The 4-step mechanism

Write a short narrative

Keep it concrete: include at least one time anchor (date/time), one place, and 2–3 actions. You may anonymize names and dates.

Deterministic extraction

The sandbox extracts basic anchors, detects minimal signals, maps them to conservative axes and thresholds. No “interpretation”, no inference, no hidden model.

Bundle draft

You get a draft rpo.json (portable, machine-readable) designed to be paired with a human-readable report in professional deployments.

Public integrity anchor

A sha256 hash is computed. If the file changes, the hash changes. That’s the whole point: integrity, not “truth”.

What RPO is (and what it is not)

RPO is a file bundle format

A shared envelope: rpo.json + (optional) rpo.pdf + proof_hash. Everyone reads the same structure.

Goal: interoperability across courts, regulators, research, platforms.

RPO is deterministic

The sandbox is conservative by design: same input → same output. It’s meant to be auditable, not impressive.

This is why we explicitly say “No AI model”.

RPO does not decide truth

It helps package a coherent handover. It does not “judge”, it does not “prove intent”. Integrity is measurable; truth requires human decision.

Hash = tamper-evidence, not truth serum.

Minimal draft example

{
  "rpo_version": "0.1",
  "context": "workplace",
  "case": { "title": "Incident — March 2024" },
  "issuer": { "label": "Internal audit team, Company X" },
  "subject": { "label": "Employee — Sales department" },
  "anchors": {
    "dates_detected": ["2024-03-14T10:30:00"],
    "places_detected": ["Paris office"]
  },
  "signals": ["threat", "pressure"],
  "canonical_map": [
    { "axis": "threats/coercion", "level": "present", "why": ["threat"] }
  ],
  "proof": {
    "hash_algo": "sha256",
    "public_hash": "…computed…"
  },
  "disclaimer": "Hash proves integrity, not truth. Format demo, not legal advice."
}

In the Sandbox you can copy JSON or download rpo.json. For the full deployment track (with a human-readable report), you request pilot access below.

Want the full pilot (CNRS × TruthX): interpretive coherence, structured narrative layers, advanced signal fusion, and a human-readable probative report designed for professionals.

Request pilot access