# CatProof documentation

Reconcile AI agent records across configured sources. Business decisions remain separate from provider observations.

- [Clear records. Confident decisions.](https://catproof.ai/docs/): Understand CatProof, connect a supported source, and reconcile agent records while preserving business ownership.
- [Your first registry](https://catproof.ai/docs/getting-started/): Create a hosted CatProof account, add a business agent, connect a source, and verify the first reconciliation workflow.
- [Reconcile agent records](https://catproof.ai/docs/reconciliation/): Review possible identity matches, link one source identity deliberately, keep records separate, defer a decision, and reverse an incorrect link.
- [Identity evidence and matching rules](https://catproof.ai/docs/matching/): Understand deterministic source-qualified matching, exact-name clues, scope boundaries, provenance separation, and the limits of candidate generation.
- [Decisions, evidence changes and undo](https://catproof.ai/docs/decisions/): How CatProof remembers review choices, fingerprints material evidence, detects stale confirmations, and conditionally reverses binding changes.
- [The reconciliation architecture](https://catproof.ai/docs/architecture/): CatProof reconciliation separates collection, source evidence, canonical identity, human review and durable business decisions through explicit transactional boundaries.
- [Records and database invariants](https://catproof.ai/docs/data-model/): The tenant-scoped connection, observation, binding, canonical agent, asset and review model behind CatProof reconciliation.
- [Transactions and concurrency](https://catproof.ai/docs/transactions/): How authorization, registry write locks, optimistic review revisions, monotonic binding revisions and atomic audit writes prevent stale reconciliation changes.
- [Refreshes, failures and missing records](https://catproof.ai/docs/source-failures/): Understand how CatProof handles incomplete source reads, renamed identities, changed IDs, paused connections and missing observations without discarding decisions.
- [Reconciliation command contracts](https://catproof.ai/docs/api/): Developer reference for the session-authenticated reconciliation decision and undo endpoints, expected revisions, evidence fingerprints and conflict behavior.
- [Security and permissions](https://catproof.ai/docs/security/): Workspace isolation, read-only source access, credential separation, authorization rechecks and metadata limits in CatProof reconciliation.
- [Operate and recover reconciliation](https://catproof.ai/docs/operations/): Operational checks for schema revisions, safe source refresh, stale-review conflicts, backup restoration and incident triage.
- [Scale design and release gates](https://catproof.ai/docs/scale/): A clearly marked target design for durable collection jobs, versioned source snapshots, outbox events, incremental candidate indexes and paginated review queues.
- [Terms and worked examples](https://catproof.ai/docs/glossary/): Definitions of canonical agent, observation, binding, source owner, freshness and provenance with examples that distinguish business identity from technical similarity.
- [Connect and refresh sources](https://catproof.ai/docs/sources/): Configure a read-only CatProof connection, preview agent records, import inventory, rotate credentials, and diagnose failed or stale source reads.
- [Local development](https://catproof.ai/docs/local-development/): Build a CatProof Docker image from source, run it with PostgreSQL and local HTTPS, or pull the published Community Edition image.

Current source baseline: a49231f. Lifecycle/rule-version hardening shipped in deployed release 708b5af. Scale architecture is planned, not shipped.