New from the Lab·The Compass — an open moral reasoning standard for AI, tested across frontier modelsExplore →
Production AI Institute · PSF v1.1 open standard
Production AI DeskTodayAI Data Use IndexCheck My AI ToolsPolicy Change WatchContactGlobal standard · Worldwide
About · Method

How a record is made

The Production AI Desk replaces personal bylines with a repeatable pipeline: observe public evidence, extract structured records, verify sources, pass a deterministic publication gate, then publish to the graph. Failed candidates are quarantined — not silently dropped.

Pipeline

  1. 1. Observe

    Fleet loops watch public disclosures, incidents, policy changes, and Lab outputs.

  2. 2. Extract

    Candidates become structured entities, events, and sources with schema fields and fingerprints.

  3. 3. Verify

    Source tier, corroboration, and confidence are evaluated against public evidence only.

  4. 4. Gate

    All ten checks must pass; otherwise the record is quarantined — never silently dropped.

  5. 5. Publish

    Approved records promote into the public graph seed and ship with canonical URLs.

Ten-check publication gate

Every record must pass all checks before publication. This is the automated control that replaces a human approval queue.

1. Source exists and is public

Every claim traces to a URL any reader can open without credentials.

2. Source tier recorded

Primary, secondary, or tertiary tier is assigned and stored with the record.

3. Severity corroboration

High- and critical-severity incidents require at least two independent public sources.

4. Required schema fields

ID, title, summary, canonical path, source refs, and confidence are all present.

5. Confidence threshold

Confidence score is at or above 0.6 before publication.

6. Duplicate check

Fingerprint matches an existing published record are blocked.

7. PSF mapping

Incidents, assessments, and policy changes include PSF domain mapping when relevant.

8. No private data

Customer, credential, or other non-public data is rejected at the gate.

9. Canonical URL valid

Title, metadata, and site-relative canonical path pass validation.

10. Retraction path

Every published record has a documented correction and retraction route.

Standards and examinations

The Production Safety Framework is grounded in deployed production systems — exam feedback, portfolio submissions, ecosystem assessments, and incident analysis all feed revisions. PAI does not operate a formal external advisory board; framework development is driven by practitioner evidence and published criteria.

Foundation and specialist examinations are developed against real deployment scenarios, reviewed for item quality, and maintained as the PSF evolves. Credential verification remains public at /certify/verify.

Cite this page

Production AI Institute. "How a record is made." https://www.productionai.institute/about/method (accessed 2026-07-03).

Full citation guidance →