TenantAtlas/specs/120-secret-redaction-integrity/data-model.md

Data Model — Secret Redaction Hardening & Snapshot Data Integrity (Spec 120)

This spec extends existing persistence and introduces no new base tables.

Entities

1) PolicyVersion (existing: App\Models\PolicyVersion)

Tenant-owned immutable policy evidence.

New / changed fields

• workspace_id (existing required scope field for newer rows; used for workspace-scoped fingerprint derivation)
• snapshot (JSON/array): protected snapshot payload with non-secret values preserved and secret values replaced by [REDACTED]
• assignments (JSON/array|null): protected assignment payload under the same contract
• scope_tags (JSON/array|null): protected scope-tag payload under the same contract
• secret_fingerprints (new JSON/array):
  • shape:
    • snapshot: object keyed by RFC 6901 JSON Pointer
    • assignments: object keyed by RFC 6901 JSON Pointer
    • scope_tags: object keyed by RFC 6901 JSON Pointer
  • values: lowercase HMAC-SHA256 hex digests
• redaction_version (new integer contract marker for compliant writes):
  • 1 = protected under the Spec 120 classifier contract

Relationships

• Belongs to Tenant
• Belongs to Policy
• Belongs to OperationRun (nullable)
• Belongs to BaselineProfile (nullable)

Validation / invariants

• New writes must set redaction_version = 1.
• If a protected value is persisted as [REDACTED], a matching digest entry must exist in secret_fingerprints for the same source bucket + JSON Pointer.
• If redaction_version = 1, secret_fingerprints may be empty only when no protected fields were classified.
• Version identity for dedupe must consider both the protected payload and secret_fingerprints so secret-only changes create a new version.

2) ProtectedSnapshotResult (new transient service DTO)

The canonical output of the new protection pipeline before persistence.

Fields

• snapshot (array)
• assignments (array|null)
• scope_tags (array|null)
• secret_fingerprints (array{snapshot: array<string, string>, assignments: array<string, string>, scope_tags: array<string, string>})
• redaction_version (int)
• protected_paths_count (int)

Validation / invariants

• Must be deterministic for the same input payload, workspace, and classifier version.
• Must preserve original object/list shape.
• Must never include raw secret values in any field.

3) SecretClassificationRule (new application-level value object)

Non-persisted classifier rule consumed by snapshot, audit, verification, and ops sanitizers.

Fields

• source_bucket (snapshot|assignments|scope_tags|audit|verification|ops_failure)
• json_pointer (string|null)
• field_name (string)
• decision (protected|visible)
• reason (exact_key, exact_path, message_pattern, default_visible)

Validation / invariants

• Exact-path rules take precedence over exact-key rules.
• Unknown fields default to visible unless a protected rule matches.
• Message-level sanitizers may protect by exact token pattern, but must not broad-match harmless phrases.

Derived / Computed Values

• protection_digest (implementation detail): composite hash of protected payload + secret_fingerprints, used for version dedupe.
• protected_change_detected (derived): true when compare/drift sees a fingerprint difference for the same protected path even though the visible payload remains [REDACTED].