Summit CognitiveDocs

REFERENCE

Glossary

Definitions of the terms used across these docs. Each entry is grounded in the live Decision Receipt API contract or the public concept pages, with links where a fuller treatment exists.

TermDefinition
Admissibility / admissible decision Whether an agent action carries enough verifiable evidence, under a stated policy, to be allowed to proceed. An admissible decision is one that passed its policy checks and replay determinism. See Decision admissibility.
Decision Receipt The cryptographically signed record emitted by POST /v1/evaluate. It captures the claim, the evidence, the policy result, the replay result, and an Ed25519 attestation, and is hash-chained to the previous receipt. See The Decision Receipt.
Claim (ClaimInput) The request body submitted for evaluation. A claim names the entity, a human-readable claim description, an optional action, and one or more evidence sources. See the API reference.
Entity The subject being evaluated — typically a repository or service, for example acme-corp/widget-api. Supplied as the required entity field on a claim.
Evidence source A single piece of supporting material attached to a claim, carrying an id, type, uri, confidence, content, and optional provenance. A claim requires at least one. See Evidence and provenance.
Source type The kind of evidence a source represents, drawn from a fixed set including ci, code_review, test_result, human_approval, agent_trace, and runtime_log. Policies can require a minimum number of distinct types.
Confidence A score from 0.0 to 1.0 on each evidence source indicating how strongly it supports the claim. A policy's minConfidence sets the minimum average confidence across sources.
Provenance / chain of custody Metadata recording where a source came from and how it was handled — captured_at, snapshot_id, collector, and a chain_of_custody list. A policy may require provenance on every source. See Evidence and provenance.
Policy (policyConfig) The rules a claim is evaluated against — minimum sources, distinct source types, average confidence, provenance and replay requirements, and risk-scope limits. Supplied inline as policyConfig or selected server-side. See Configuring policies.
Verdict The policy outcome for a claim: ALLOWED when all checks pass, BLOCKED when a check fails, or ESCALATED when human review is called for. See Policies and verdicts.
Admissibility status The receipt-level disposition: ACCEPTED, NON_DETERMINISTIC (replay did not reproduce the result), or REJECTED. See Decision admissibility.
Replay / replay determinism Re-running the evaluation over the same inputs to confirm it reproduces the same result. A policy's requireReplay makes determinism a condition of admissibility. See Replay and determinism.
Replay hash The replay_hash (a sha256: digest) recorded for the replay result. Matching replay hashes across runs are what make a decision deterministic.
Attestation A signature block on a receipt naming the signer, the algorithm (Ed25519), the signature, and a timestamp. It lets anyone confirm the receipt was issued by the server and not altered.
Ed25519 The public-key signature scheme used for attestations. The server's public key is published at GET /v1/keys/server in PEM form, so receipts can be verified offline. See Verifying receipts.
Hash chain The linkage that ties receipts together in order. Each receipt's chain block carries a sequence number and the previous_receipt_hash, making any retroactive edit detectable.
Inputs hash The inputs_hash on a receipt's evidence block — a sha256: digest of the evaluated inputs, binding the receipt to the exact claim and evidence it was issued for.
Ledger The append-only store of all issued receipts. Verification recomputes a receipt's canonical hash and matches it against the ledger; aggregate counts are available at GET /v1/ledger/stats.
Tier The plan attached to an API key, setting its hourly rate limit: free (100 requests/hour), pilot (10,000), or enterprise (unlimited). New signups start on free.
Standing / contestability The principle that a decision should be open to challenge by an affected party. Because every receipt is signed and independently verifiable, anyone can recheck the evidence and result behind a decision rather than having to take it on trust.