HubPublic

Tamper-evident records for submitted AI answers

Re-verify a disputed AI answer with the evidence recorded at the time.

rawctx groups answer records, model-run information, source references, and correction history into one evidence record, then reconciles it with an independent system of record to detect missing or late records.

rawctx verifies the integrity of submitted records and reconciliation results. It does not by itself guarantee answer accuracy, actual delivery, or that a source was used in reasoning.

pip install rawctx

Reconciliation guide
3 signals
missing · unexpected · late
hash-only
anonymous public receipt
1 reference
independent system of record

SHA-256 hashes are sent instead of the original text. Do not enter personal data, secrets, or predictable sensitive phrases into an anonymous public receipt. Use the tenant HMAC path in production.

This anonymous receipt is public. Anyone with the link can view it. This form does not promise a retention period.

No signup · Hash-only submission · Public link

Independent Reconciliation

Find missing, unexpected, and late answer records first.

Compare rawctx records with an independent reference source before relying on cryptographic proof. A separate business system makes coverage gaps and timing differences visible.

01 Independent reference

Reconcile against a separate system of record.

Compare answer records with an independently sourced manifest to detect missing, unexpected, and late entries.

  • missing
  • unexpected
  • late

02 Evidence record

Keep the submitted record and its changes together.

Group the answer record, model-run information, source references, and correction history for export and review.

  • pip install rawctx
  • rawctx login
  • receipt URL

03 Integrity and limits

Verify submitted records without overstating them.

Integrity checks and explicit proof states cover the submitted record. They do not prove answer accuracy, actual delivery, or source use in reasoning.

  • Merkle proof
  • explicit proof states
  • documented limits

Run it from code

The same receipt shape in three lines.

After the browser receipt, create a workspace, get a token, and submit answer records with the same hash-first shape.

Logging quickstart

Build a Python logging call.

Build the minimal rawctx logging call, then add optional parameters as needed.

Answer input
application_keyquestion_textanswer_text
Optional parameters
Proof verifier inputPaste into the verifier above
{}

Review the trust boundary

Use public contracts and explicit states as the trust basis.

Review the public verifier contract, proof-state semantics, first-party platform monitoring, and documented limits before relying on a record.

  • Public verifier specification
  • Explicit proof states
  • First-party platform status
  • Documented trust boundaries
Verifier specificationOpen the public contract

Proof bundle verifier

Paste a rawctx proof bundle to verify hashes, Merkle inclusion, signatures, and public witness bindings.

Public Merkle verifier spec