Glossary

Terms used across the AXIS documentation, in the canonical sense defined by the project.

ALLOW / BLOCK / REQUIRE_APPROVAL

The three policy decision outcomes. ALLOW executes (200), BLOCK rejects with no execution (403), REQUIRE_APPROVAL creates an approval record and returns approval_id (202).

Approval

Human consent recorded for a pending write. Consent is not execution: the client must retry with approval_id. Resolution is immutable (approve or reject).

approval_id

Identifier returned in 202 responses; required on the explicit retry so AXIS can bind the retry to the approved consent.

AXIS gate

The Rust/Axum service that implements the write-path control pipeline: classify, evaluate policy, enforce, record evidence.

Classifier

The SQL classification component bound by the AXIS Security Classification Contract v1.0; produces QueryType and risk signals.

Control Plane

The Next.js operator surface (port 3000) that proxies /api/axis/* server-side; the browser never sees the backend URL or token.

Deterministic

Same SQL + context + policy version always yields the same decision. AXIS explicitly rejects AI-based probabilistic judgment.

Dry-run

Decision preview without execution or audit writes: would_execute: false, audit_written: false.

Evidence

Durable audit records (WAL + JSONL projection) covering decisions, executions, approvals and policy lifecycle events.

Evidence Bundle V1

Range export (bundle_type: axis.evidence_bundle.v1) with first/last seq and hashes, optional Ed25519 signature, for out-of-band verification.

event_hash / previous_hash

Per-event SHA-256 binding the event to its predecessor; tampering breaks the chain.

execution_state

Outcome of an allowed execution: success, failure, or unknown (timeout after dispatch — verify before retry).

Fail-closed

Uncertain or unsupported inputs are rejected or blocked; there is no permissive default and no silent read assumption.

Fingerprint

Normalized identifier of a classified SQL statement, carried in decisions and audit events.

JSONL projection

Human-readable audit log (audit.log); non-canonical and safe to delete.

Manifest (policy)

policy_manifest.json — authoritative startup source pointing to the active policy with its expected SHA-256.

Operating mode

shadow, approval_first, enforce, emergency_bypass — how decisions are applied (classification and evaluation are identical in all modes).

Operator token

AXIS_OPERATOR_TOKEN, sent as X-AXIS-Operator-Token or Bearer; required for policy activation/rollback and approval resolution.

Policy version

Immutable policy identity (e.g. prod_main@1.0.0); carried into decisions and evidence.

QueryType

Primary classification: Read, Write, Unknown (Unknown defaults to Write/Block).

Risk signal

Defense-in-depth classification fallback (e.g. select_into, cte_write, volatile_function, copy_program) that can only tighten a read default.

Runtime profile

AXIS_RUNTIME_PROFILE: local (pilot defaults) or production (stricter posture).

Session store

In-memory, session_id-scoped metadata for AXIS-side PREPARE/EXECUTE/DEALLOCATE; lost on restart (fail-closed).

WAL

Write-ahead log; the canonical append-only evidence store (audit.wal) with fsync-backed durability.

Related: FAQ · Architecture · Audit & Evidence