Canonical clause
A verbatim excerpt from a gate-verified policy snapshot. It carries the source URL, deep link, structural citation, character offsets, surface, tier applicability, risk label, confidence, and snapshot provenance.
Canonical clauses first.
Answers second.
Clause Intelligence is AIRIN's evidence layer for policy search, Ask answers, evolution timelines, and PDF reports. It starts with source-verified clauses, then builds deterministic pattern matches and dated stance events on top.
74,115
Canonical clauses
15,101
Pattern matches
15,101
Stance events
8
Platforms with clauses
Risk pattern
A deterministic matcher over canonical clause text. Patterns identify repeatable language families such as training use, IP license grants, retention exceptions, commercial-use limits, confidentiality, audit rights, and tier conditionality.
Stance event
A dated policy position derived from a matched clause: training permitted, training excluded, opt-out available, broad license granted, retention exception present, or similar. Stance events are the substrate for policy evolution.
Evolution diff
A before-and-after comparison of stance events across snapshots. AIRIN labels a change improved, worsened, or changed only when the prior and current positions both have citations.
Clause retrieval
Ask AI and machine search retrieve canonical clauses first, then answer from those clauses. Embeddings may rank relevant clauses, but they are never allowed to create facts, ratings, or citations.
Report surface
Reports package the same canonical clauses into a PDF-first deliverable: tier matrix, policy evolution, top hazardous clauses, and source records for auditability.
The retrieval boundary
Semantic search is allowed to make relevant clauses easier to find. It is not allowed to decide what is true. AIRIN can rank clauses with embeddings, but the final answer still has to cite exact canonical clauses and stance events. If the retrieved evidence does not answer the user's question, the product must say so.
Publication guarantees
These constraints are the reason AIRIN can expose clause intelligence publicly without a human review queue. The system is only as expressive as the evidence it can verify.
- Verified: No displayed clause intelligence exists without a policy_clauses row.
- Verified: No policy_clauses row can be inserted unless its ingestion run passed Gate 2.
- Verified: No Ask answer may cite a clause id that was not retrieved from the canonical clause store.
- Verified: No policy evolution claim is emitted unless before/after citations exist.
- Verified: No human manually approves, edits, or overrides clause intelligence findings before publication.
Need the full pipeline?
The main methodology page explains harvest, extraction, verification gates, consensus, scoring, contradictions, and the source archive.
📢 POLICY UPDATES ALERT
AIRIN Brief
Built for compliance officers, legal counsel, and SaaS founders. Subscribe to the email digest — one short brief when a tracked vendor materially changes its terms, training policy, or risk rating. Prefer in-app? Watch platforms in your alerts inbox instead.