Living Implementation Checklist

Living Implementation Checklist

Version: 1.2.0
Last Updated: 2026-01-02
Status: Complete (All 8 Phases)

---

Phase 0: Project Control Layer ✅ COMPLETE

• [x] 0.1 Create Project Charter
• Owner: Agent
• Input: Improvement document, existing architecture
• Output: `docs/00-PROJECT_CHARTER.md`
• Validation: Charter defines purpose, non-goals, success criteria
• Status: ✅ Complete
• Confidence: High

• [x] 0.2 Create System Glossary
• Owner: Agent
• Input: Existing code, architecture docs
• Output: `docs/01-GLOSSARY.md`
• Validation: All security terms defined, overloaded terms flagged
• Status: ✅ Complete
• Confidence: High

• [x] 0.3 Create Invariants Ledger
• Owner: Agent
• Input: Improvement document, existing tests
• Output: `docs/02-INVARIANTS_LEDGER.md`
• Validation: All 8 invariants documented with canaries
• Status: ✅ Complete
• Confidence: High

• [x] 0.4 Create Implementation Guide
• Owner: Agent
• Input: CLAUDE.md, improvement document
• Output: `docs/03-IMPLEMENTATION_GUIDE.md`
• Validation: Decision rules, decomposition rules, validation rules defined
• Status: ✅ Complete
• Confidence: High

• [x] 0.5 Create Living Checklist (this document)
• Owner: Agent
• Input: All Phase 0 artifacts
• Output: `docs/04-IMPLEMENTATION_CHECKLIST.md`
• Validation: All tasks listed with owners, inputs, outputs, validation
• Status: ✅ Complete
• Confidence: High

---

Phase 1: Type-Level Safety (Priority 1) ✅ COMPLETE

Goal: Make it physically impossible to pass unverified evidence to promotion APIs.

1.1 Core Type Definition

• [x] 1.1.1 Define `AdmissibleEvidenceBundle` struct
• Owner: Agent
• Input: `types/slice.rs`, glossary
• Output: `src/types/admissible.rs`
• Validation: Struct contains `SliceExport` + verified marker
• Status: ✅ Complete
• Confidence: High

• [x] 1.1.2 Implement private constructor `from_verified`
• Owner: Agent
• Input: `AdmissibleEvidenceBundle` struct
• Output: `impl AdmissibleEvidenceBundle` in same file
• Validation: Constructor requires HMAC secret, calls `verify_token()`
• Status: ✅ Complete
• Confidence: High

• [x] 1.1.3 Add public accessor methods
• Owner: Agent
• Input: `AdmissibleEvidenceBundle` struct
• Output: `slice()`, `turn_ids()`, `provenance()` methods
• Validation: Methods return borrowed references (no cloning)
• Status: ✅ Complete
• Confidence: High

1.2 Testing

• [x] 1.2.1 Unit test: successful verification
• Owner: Agent
• Input: `AdmissibleEvidenceBundle` implementation
• Output: Test in `src/types/admissible.rs`
• Validation: Test passes with correct secret
• Status: ✅ Complete
• Confidence: High

• [x] 1.2.2 Unit test: verification failure
• Owner: Agent
• Input: `AdmissibleEvidenceBundle` implementation
• Output: Test in same file
• Validation: Test returns error with wrong secret
• Status: ✅ Complete
• Confidence: High

• [x] 1.2.3 Unit test: cannot construct with invalid token
• Owner: Agent
• Input: `AdmissibleEvidenceBundle` implementation
• Output: Test in same file
• Validation: Test verifies private constructor enforces verification
• Status: ✅ Complete
• Confidence: High

1.3 Integration

• [x] 1.3.1 Add re-export to `lib.rs`
• Owner: Agent
• Input: `AdmissibleEvidenceBundle` module
• Output: Modified `lib.rs`
• Validation: `pub use types::AdmissibleEvidenceBundle`
• Status: ✅ Complete
• Confidence: High

• [x] 1.3.2 Update `ContextSlicer::slice` return type
• Owner: Agent
• Input: `AdmissibleEvidenceBundle`, `slicer.rs`
• Output: Modified `slicer.rs`
• Validation: Returns `AdmissibleEvidenceBundle` instead of `SliceExport`
• Status: ✅ Complete (Phase 8)
• Confidence: High

• [x] 1.3.3 Update service routes to use new type
• Owner: Agent
• Input: `service/routes.rs`, `AdmissibleEvidenceBundle`
• Output: Modified route handlers
• Validation: Routes serialize bundle correctly
• Status: ✅ Complete (Phase 8)
• Confidence: High

1.4 Documentation

• [x] 1.4.1 Document type-level safety in architecture doc
• Owner: Agent
• Input: `docs/architecture/15-GRAPH_KERNEL.md`, implementation
• Output: Updated architecture doc
• Validation: Section explains why promotion APIs are type-safe
• Status: ✅ Complete (inline docs in admissible.rs)
• Confidence: High

---

Phase 2: Content Canonicalization (Priority 2) ✅ COMPLETE

Goal: Ensure 100

2.1 Canonical Specification

• [x] 2.1.1 Define canonical content transformation
• Owner: Agent
• Input: Improvement doc, existing `content_hash` usage
• Output: `src/canonical_content.rs`
• Validation: Function signature `fn canonical_content(text: &str) -> Vec<u8>`
• Status: ✅ Complete
• Confidence: High

• [x] 2.1.2 Implement normalization steps
• Owner: Agent
• Input: Canonical spec
• Output: `canonical_content()` implementation
• Validation: Handles UTF-8, newlines, whitespace consistently
• Status: ✅ Complete (CRLF→LF, trim whitespace, UTF-8 encode)
• Confidence: High

• [x] 2.1.3 Implement SHA-256 hashing
• Owner: Agent
• Input: `canonical_content()` output
• Output: `fn compute_content_hash(text: &str) -> String` (hex)
• Validation: Returns hex SHA-256 of canonical bytes
• Status: ✅ Complete
• Confidence: High

2.2 Testing

• [x] 2.2.1 Test: determinism
• Owner: Agent
• Input: `canonical_content()` implementation
• Output: `test_compute_content_hash_determinism`, `test_canonical_content_determinism`
• Validation: Test passes (21 tests total)
• Status: ✅ Complete
• Confidence: High

• [x] 2.2.2 Test: newline normalization
• Owner: Agent
• Input: `canonical_content()` implementation
• Output: `test_normalize_text_crlf`, `test_normalize_text_cr`, `test_content_hash_newline_normalization`
• Validation: Test passes - CRLF, CR, LF all produce same hash
• Status: ✅ Complete
• Confidence: High

• [x] 2.2.3 Test: whitespace handling
• Owner: Agent
• Input: `canonical_content()` implementation
• Output: `test_normalize_text_trim`, `test_content_hash_whitespace_normalization`
• Validation: Test passes - leading/trailing whitespace trimmed
• Status: ✅ Complete
• Confidence: High

2.3 Enforcement

• [ ] 2.3.1 Add content hash computation to `TurnSnapshot::new` (DEFERRED)
• Owner: Agent
• Input: `compute_content_hash()` function, `types/turn.rs`
• Output: Modified constructor
• Validation: All new turns get `content_hash` automatically
• Status: ⏳ Deferred (DB trigger handles new data automatically)
• Confidence: Medium
• Note: Less critical since `content_hash_trigger` on INSERT/UPDATE handles this

• [x] 2.3.2 Backfill content_hash for legacy turns
• Owner: Agent
• Input: Existing `compute_content_hash()` DB function
• Output: 97,110 turns with content_hash (90.55
• Validation: All turns with non-empty content have hash
• Status: ✅ Complete (2026-01-03)
• Confidence: High
• Note: 10,135 turns have empty content_text → NULL hash (correct behavior)

2.4 Documentation

• [x] 2.4.1 Document canonical content spec
• Owner: Agent
• Input: Implementation
• Output: Inline docs in `src/canonical_content.rs` (lines 1-39)
• Validation: Spec includes normalization rules + examples
• Status: ✅ Complete
• Confidence: High

---

Phase 3: Token Verification Production Mode (Priority 3) ✅ COMPLETE

Goal: Make verification < 5ms p99 with caching, default to single mode.

3.1 Verification Modes

• [x] 3.1.1 Define `VerificationMode` enum
• Owner: Agent
• Input: Improvement doc
• Output: `src/types/verification.rs`
• Validation: Enum has `LocalSecret`, `Cached` with config
• Status: ✅ Complete
• Confidence: High

• [x] 3.1.2 Implement local verification
• Owner: Agent
• Input: Existing `verify_hmac()` method
• Output: `TokenVerifier::verify_token()` in local mode
• Validation: Uses shared secret, constant-time comparison
• Status: ✅ Complete
• Confidence: High

• [x] 3.1.3 Implement cached verification
• Owner: Agent
• Input: `verify_token()`, LRU crate
• Output: `TokenVerifier` with `VerificationMode::Cached`
• Validation: LRU cache keyed by xxh64 of all canonical fields
• Status: ✅ Complete (uses `lru` + `parking_lot` crates)
• Confidence: High

3.2 Performance

• [x] 3.2.1 Add benchmark for token verification
• Owner: Agent
• Input: `TokenVerifier` implementation
• Output: `benches/verification.rs`
• Validation: Criterion benchmark runs, reports latency for cold/cached verification
• Status: ✅ Complete (Phase 8)
• Confidence: High

• [x] 3.2.2 Cache key design for optimal hit rate
• Owner: Agent
• Input: Verification parameters
• Output: `VerificationCacheKey` using xxh64 hash
• Validation: Unique key per parameter combination
• Status: ✅ Complete (tests verify uniqueness)
• Confidence: High

3.3 Documentation

• [x] 3.3.1 Document verification modes in module
• Owner: Agent
• Input: Implementation
• Output: Inline docs in `src/types/verification.rs`
• Validation: Module doc explains when to use each mode
• Status: ✅ Complete
• Confidence: High

---

Phase 4: Sufficiency Framework (Priority 4) ✅ COMPLETE

Goal: Prevent gaming with low-quality homogeneous evidence.

4.1 Diversity Metrics

• [x] 4.1.1 Define `DiversityMetrics` struct
• Owner: Agent
• Input: Improvement doc, glossary
• Output: `src/types/sufficiency.rs`
• Validation: Struct captures role, phase, salience, session diversity
• Status: ✅ Complete
• Confidence: High

• [x] 4.1.2 Implement `SalienceStats` aggregation
• Owner: Agent
• Input: Turn salience scores
• Output: Min, max, mean, std_dev, high_salience_count
• Validation: Tests verify correct statistical computation
• Status: ✅ Complete
• Confidence: High

• [x] 4.1.3 Implement `DiversityMetrics::from_bundle()`
• Owner: Agent
• Input: `AdmissibleEvidenceBundle`
• Output: Computed diversity metrics
• Validation: Correctly counts roles, phases, sessions
• Status: ✅ Complete
• Confidence: High

4.2 Sufficiency Policy

• [x] 4.2.1 Define `SufficiencyPolicy` struct
• Owner: Agent
• Input: Improvement doc
• Output: Configurable thresholds in `sufficiency.rs`
• Validation: min_turns, min_roles, min_phases, min_high_salience, require_exchange, min_mean_salience
• Status: ✅ Complete
• Confidence: High

• [x] 4.2.2 Implement policy presets
• Owner: Agent
• Input: `SufficiencyPolicy` struct
• Output: `default()`, `lenient()`, `strict()` constructors
• Validation: Each preset has appropriate thresholds
• Status: ✅ Complete
• Confidence: High

• [x] 4.2.3 Implement `SufficiencyPolicy::check()`
• Owner: Agent
• Input: `DiversityMetrics`
• Output: `SufficiencyCheck` with violations list
• Validation: Returns detailed violation report
• Status: ✅ Complete
• Confidence: High

4.3 Evidence Bundle

• [x] 4.3.1 Define `EvidenceBundle` struct
• Owner: Agent
• Input: `AdmissibleEvidenceBundle`, `DiversityMetrics`
• Output: Combined admissibility + sufficiency type
• Validation: Wraps admissible bundle with metrics
• Status: ✅ Complete
• Confidence: High

• [x] 4.3.2 Implement `EvidenceBundle::from_admissible()`
• Owner: Agent
• Input: Admissible bundle, policy
• Output: Result<EvidenceBundle, EvidenceBundleError>
• Validation: Fails if policy not satisfied
• Status: ✅ Complete
• Confidence: High

• [x] 4.3.3 Add accessor methods
• Owner: Agent
• Input: `EvidenceBundle` struct
• Output: `admissible_bundle()`, `metrics()`, `policy_id()`, etc.
• Validation: All fields accessible without cloning
• Status: ✅ Complete
• Confidence: High

4.4 Testing

• [x] 4.4.1 Unit tests for diversity metrics (5 tests)
• Owner: Agent
• Input: `DiversityMetrics` implementation
• Output: Tests in `sufficiency.rs`
• Validation: Tests pass for single turn, conversation, multi-session
• Status: ✅ Complete (13 tests total)
• Confidence: High

• [x] 4.4.2 Unit tests for sufficiency policy (5 tests)
• Owner: Agent
• Input: `SufficiencyPolicy` implementation
• Output: Tests for each violation type
• Validation: Tests cover insufficient turns, roles, salience, exchange
• Status: ✅ Complete
• Confidence: High

• [x] 4.4.3 Unit tests for evidence bundle (3 tests)
• Owner: Agent
• Input: `EvidenceBundle` implementation
• Output: Tests for success and failure cases
• Validation: Tests verify construction fails appropriately
• Status: ✅ Complete
• Confidence: High

4.5 Integration

• [x] 4.5.1 Add module to `types/mod.rs`
• Owner: Agent
• Input: `sufficiency.rs`
• Output: `pub mod sufficiency;` + re-exports
• Validation: Module accessible from `types::`
• Status: ✅ Complete
• Confidence: High

• [x] 4.5.2 Add re-exports to `lib.rs`
• Owner: Agent
• Input: Sufficiency types
• Output: `pub use types::sufficiency::*`
• Validation: Types accessible from crate root
• Status: ✅ Complete
• Confidence: High

---

Phase 5: Database-Level Slice Boundaries (Priority 5) ✅ COMPLETE

Goal: Enforce INV-GK-008 at the database query level—prevent out-of-slice access.

5.1 Boundary Guard Type

• [x] 5.1.1 Define `SliceBoundaryGuard` struct
• Owner: Agent
• Input: Invariants ledger (INV-GK-008), glossary
• Output: `src/types/boundary.rs`
• Validation: Struct contains turn_ids, slice_fingerprint, boundary_hash
• Status: ✅ Complete
• Confidence: High

• [x] 5.1.2 Implement `from_slice()` constructor
• Owner: Agent
• Input: `SliceExport` type
• Output: `SliceBoundaryGuard::from_slice(&SliceExport)`
• Validation: Extracts turn IDs, computes boundary hash
• Status: ✅ Complete
• Confidence: High

• [x] 5.1.3 Implement `as_uuid_array()` for SQL binding
• Owner: Agent
• Input: `SliceBoundaryGuard` struct
• Output: `fn as_uuid_array(&self) -> Vec<uuid::Uuid>`
• Validation: Returns UUIDs safe for parameterized queries
• Status: ✅ Complete
• Confidence: High

5.2 Query Builder

• [x] 5.2.1 Define `BoundedQueryBuilder` struct
• Owner: Agent
• Input: Security requirements
• Output: Builder in `boundary.rs`
• Validation: Forces `WHERE id = ANY($1)` pattern
• Status: ✅ Complete
• Confidence: High

• [x] 5.2.2 Implement `select()`, `filter()`, `order_by()` methods
• Owner: Agent
• Input: SQL query patterns
• Output: Builder methods
• Validation: Only parameterized patterns allowed
• Status: ✅ Complete
• Confidence: High

• [x] 5.2.3 Implement `build()` method
• Owner: Agent
• Input: Builder state
• Output: SQL string with $1 placeholder for turn IDs
• Validation: Generated SQL is injection-safe
• Status: ✅ Complete
• Confidence: High

5.3 Boundary Violation Detection

• [x] 5.3.1 Define `BoundaryViolation` struct
• Owner: Agent
• Input: Security requirements
• Output: Violation report type
• Validation: Contains slice_fingerprint, unauthorized_ids, timestamp, context
• Status: ✅ Complete
• Confidence: High

• [x] 5.3.2 Implement `check_access()` method
• Owner: Agent
• Input: Requested turn IDs
• Output: `BoundaryCheck` enum (Authorized | Violation)
• Validation: Detects out-of-slice access attempts
• Status: ✅ Complete
• Confidence: High

• [x] 5.3.3 Implement violation logging
• Owner: Agent
• Input: `BoundaryViolation`
• Output: `log()` method using tracing
• Validation: Logs with SLICE_BOUNDARY_VIOLATION marker
• Status: ✅ Complete
• Confidence: High

5.4 Testing

• [x] 5.4.1 Unit tests for boundary guard (4 tests)
• Owner: Agent
• Input: `SliceBoundaryGuard` implementation
• Output: Tests in `boundary.rs`
• Validation: Tests creation, contains, as_uuid_array, as_set
• Status: ✅ Complete
• Confidence: High

• [x] 5.4.2 Unit tests for boundary check (2 tests)
• Owner: Agent
• Input: `check_access()` method
• Output: Tests for authorized and violation cases
• Validation: Correctly identifies out-of-slice access
• Status: ✅ Complete
• Confidence: High

• [x] 5.4.3 Unit tests for query builder (3 tests)
• Owner: Agent
• Input: `BoundedQueryBuilder` implementation
• Output: Tests for basic, columns, filters
• Validation: Generated SQL matches expected patterns
• Status: ✅ Complete
• Confidence: High

• [x] 5.4.4 Unit tests for boundary hash (2 tests)
• Owner: Agent
• Input: `boundary_hash` implementation
• Output: Tests for determinism and difference detection
• Validation: Same turns → same hash, different turns → different hash
• Status: ✅ Complete
• Confidence: High

5.5 Integration

• [x] 5.5.1 Add module to `types/mod.rs`
• Owner: Agent
• Input: `boundary.rs`
• Output: `pub mod boundary;` + re-exports
• Validation: Module accessible from `types::`
• Status: ✅ Complete
• Confidence: High

• [x] 5.5.2 Add re-exports to `lib.rs`
• Owner: Agent
• Input: Boundary types
• Output: `pub use types::boundary::*`
• Validation: Types accessible from crate root
• Status: ✅ Complete
• Confidence: High

---

Phase 6: Replay Provenance Expansion (Priority 6) ✅ COMPLETE

Goal: Capture all parameters needed to exactly reproduce a slice retrieval result.

6.1 Embedding Model Reference

• [x] 6.1.1 Define `EmbeddingModelRef` struct
• Owner: Agent
• Input: Improvement doc, glossary
• Output: `src/types/provenance.rs`
• Validation: Contains model_id, version, dimensions, quantization, deterministic
• Status: ✅ Complete
• Confidence: High

• [x] 6.1.2 Implement builder pattern methods
• Owner: Agent
• Input: `EmbeddingModelRef` struct
• Output: `new()`, `with_quantization()`, `non_deterministic()` methods
• Validation: Fluent API for construction
• Status: ✅ Complete
• Confidence: High

• [x] 6.1.3 Implement `to_ref_string()` for unique identification
• Owner: Agent
• Input: Model reference
• Output: Format: `model_id@version:dDIM:qQUANT`
• Validation: Unique string per model configuration
• Status: ✅ Complete
• Confidence: High

6.2 Normalization Version

• [x] 6.2.1 Define `NormalizationVersion` struct
• Owner: Agent
• Input: Canonical content spec
• Output: Struct in `provenance.rs`
• Validation: Contains version, config_hash, features list
• Status: ✅ Complete
• Confidence: High

• [x] 6.2.2 Implement `current()` factory
• Owner: Agent
• Input: `CANONICAL_CONTENT_VERSION` constant
• Output: Returns current normalization version
• Validation: References the actual version constant
• Status: ✅ Complete
• Confidence: High

6.3 Retrieval Parameters

• [x] 6.3.1 Define `RetrievalParams` struct
• Owner: Agent
• Input: RAG++ retrieval API
• Output: Struct in `provenance.rs`
• Validation: Contains k, threshold, reranking, max_tokens, policy_version
• Status: ✅ Complete
• Confidence: High

• [x] 6.3.2 Implement builder methods
• Owner: Agent
• Input: `RetrievalParams` struct
• Output: `new()`, `with_reranking()`, `with_max_tokens()`, `with_policy_params_hash()`
• Validation: Fluent API for construction
• Status: ✅ Complete
• Confidence: High

6.4 Replay Provenance

• [x] 6.4.1 Define `ReplayProvenance` struct
• Owner: Agent
• Input: All provenance components
• Output: Complete provenance type
• Validation: Contains timestamp, embedding_model, normalization, retrieval_params, graph_snapshot, slice_fingerprint
• Status: ✅ Complete
• Confidence: High

• [x] 6.4.2 Implement `is_complete()` validation
• Owner: Agent
• Input: `ReplayProvenance` struct
• Output: `fn is_complete(&self) -> bool`
• Validation: Checks all required fields are present
• Status: ✅ Complete
• Confidence: High

• [x] 6.4.3 Implement `fingerprint()` for comparison
• Owner: Agent
• Input: Provenance fields
• Output: xxh64 hash of canonical provenance string
• Validation: Deterministic fingerprint
• Status: ✅ Complete
• Confidence: High

• [x] 6.4.4 Implement `is_replay_compatible()` comparison
• Owner: Agent
• Input: Two provenances
• Output: `fn is_replay_compatible(&self, other: &Self) -> bool`
• Validation: Returns true if replay would produce same slice
• Status: ✅ Complete
• Confidence: High

6.5 Provenance Builder

• [x] 6.5.1 Define `ProvenanceBuilder` struct
• Owner: Agent
• Input: `ReplayProvenance` requirements
• Output: Builder in `provenance.rs`
• Validation: Optional fields with required field enforcement
• Status: ✅ Complete
• Confidence: High

• [x] 6.5.2 Implement builder methods
• Owner: Agent
• Input: Builder struct
• Output: All setter methods + `build()`
• Validation: `build()` returns `Result<ReplayProvenance, ProvenanceError>`
• Status: ✅ Complete
• Confidence: High

6.6 Testing

• [x] 6.6.1 Unit tests for embedding model ref (1 test)
• Owner: Agent
• Input: `EmbeddingModelRef` implementation
• Output: Tests in `provenance.rs`
• Validation: Tests construction and ref_string
• Status: ✅ Complete
• Confidence: High

• [x] 6.6.2 Unit tests for normalization version (1 test)
• Owner: Agent
• Input: `NormalizationVersion` implementation
• Output: Tests `current()` returns expected values
• Validation: Version and features match constants
• Status: ✅ Complete
• Confidence: High

• [x] 6.6.3 Unit tests for retrieval params (1 test)
• Owner: Agent
• Input: `RetrievalParams` implementation
• Output: Tests builder pattern
• Validation: All fields set correctly
• Status: ✅ Complete
• Confidence: High

• [x] 6.6.4 Unit tests for provenance builder (2 tests)
• Owner: Agent
• Input: `ProvenanceBuilder` implementation
• Output: Tests for success and missing field error
• Validation: Builder enforces required fields
• Status: ✅ Complete
• Confidence: High

• [x] 6.6.5 Unit tests for fingerprint determinism (1 test)
• Owner: Agent
• Input: `fingerprint()` method
• Output: Test that same inputs → same fingerprint
• Validation: Determinism verified
• Status: ✅ Complete
• Confidence: High

• [x] 6.6.6 Unit tests for replay compatibility (1 test)
• Owner: Agent
• Input: `is_replay_compatible()` method
• Output: Tests same vs different provenances
• Validation: Correctly identifies compatible provenances
• Status: ✅ Complete
• Confidence: High

6.7 Integration

• [x] 6.7.1 Add module to `types/mod.rs`
• Owner: Agent
• Input: `provenance.rs`
• Output: `pub mod provenance;` + re-exports
• Validation: Module accessible from `types::`
• Status: ✅ Complete
• Confidence: High

• [x] 6.7.2 Add re-exports to `lib.rs`
• Owner: Agent
• Input: Provenance types
• Output: `pub use types::provenance::*`
• Validation: Types accessible from crate root
• Status: ✅ Complete
• Confidence: High

---

Phase 7: Incident Response Infrastructure (Priority 7) ✅ COMPLETE

Goal: Production-grade metrics, alerting, and quarantine for security violations.

7.1 Severity and Incident Types

• [x] 7.1.1 Define `Severity` enum
• Owner: Agent
• Input: Security requirements
• Output: `src/types/incident.rs`
• Validation: Critical (15min), High (1hr), Medium (4hr), Low (1day) with response times
• Status: ✅ Complete
• Confidence: High

• [x] 7.1.2 Define `IncidentType` enum
• Owner: Agent
• Input: Invariants ledger (INV-GK-001 through INV-GK-008)
• Output: Enum variants for each invariant violation
• Validation: 8 variants mapping to all invariants
• Status: ✅ Complete
• Confidence: High

• [x] 7.1.3 Implement `severity()` method on `IncidentType`
• Owner: Agent
• Input: `IncidentType` enum
• Output: `fn severity(&self) -> Severity`
• Validation: Maps each incident type to appropriate severity
• Status: ✅ Complete
• Confidence: High

• [x] 7.1.4 Implement `invariant()` method on `IncidentType`
• Owner: Agent
• Input: `IncidentType` enum
• Output: `fn invariant(&self) -> &'static str`
• Validation: Returns invariant ID (e.g., "INV-GK-001")
• Status: ✅ Complete
• Confidence: High

7.2 Incident Struct

• [x] 7.2.1 Define `Incident` struct
• Owner: Agent
• Input: Security requirements
• Output: Struct in `incident.rs`
• Validation: Contains id, timestamp, incident_type, severity, context, acknowledged
• Status: ✅ Complete
• Confidence: High

• [x] 7.2.2 Implement `new()` and type-specific constructors
• Owner: Agent
• Input: `Incident` struct
• Output: `new()`, `slice_boundary_violation()`, `token_verification_failure()`, etc.
• Validation: Factory methods for each incident type
• Status: ✅ Complete
• Confidence: High

• [x] 7.2.3 Implement `acknowledge()` method
• Owner: Agent
• Input: `Incident` struct
• Output: `fn acknowledge(&mut self, by: &str)`
• Validation: Sets acknowledged_at and acknowledged_by
• Status: ✅ Complete
• Confidence: High

• [x] 7.2.4 Implement `alert()` method
• Owner: Agent
• Input: `Incident` struct
• Output: `fn alert(&self)` using tracing
• Validation: Logs with appropriate level based on severity
• Status: ✅ Complete
• Confidence: High

7.3 Quarantine System

• [x] 7.3.1 Define `QuarantinedToken` struct
• Owner: Agent
• Input: Security requirements
• Output: Struct in `incident.rs`
• Validation: Contains slice_fingerprint, token, reason, incident_id, timestamps
• Status: ✅ Complete
• Confidence: High

• [x] 7.3.2 Implement `new()` constructor
• Owner: Agent
• Input: `QuarantinedToken` struct
• Output: Constructor with required fields
• Validation: Sets quarantined_at timestamp
• Status: ✅ Complete
• Confidence: High

• [x] 7.3.3 Implement `review()` method
• Owner: Agent
• Input: `QuarantinedToken` struct
• Output: `fn review(&mut self, by: &str, approved: bool)`
• Validation: Sets reviewed_at, reviewed_by, approved
• Status: ✅ Complete
• Confidence: High

7.4 SQL Schemas

• [x] 7.4.1 Define `QUARANTINE_TABLE_SCHEMA` constant
• Owner: Agent
• Input: `QuarantinedToken` struct
• Output: SQL CREATE TABLE statement
• Validation: All fields mapped to appropriate SQL types
• Status: ✅ Complete
• Confidence: High

• [x] 7.4.2 Define `INCIDENT_TABLE_SCHEMA` constant
• Owner: Agent
• Input: `Incident` struct
• Output: SQL CREATE TABLE statement
• Validation: Includes indexes for common queries
• Status: ✅ Complete
• Confidence: High

7.5 Metrics Interface

• [x] 7.5.1 Define `IncidentMetrics` trait
• Owner: Agent
• Input: Prometheus metrics conventions
• Output: Trait in `incident.rs`
• Validation: Methods for `record_incident()`, `record_quarantine()`
• Status: ✅ Complete
• Confidence: High

• [x] 7.5.2 Implement `NoOpMetrics` default
• Owner: Agent
• Input: `IncidentMetrics` trait
• Output: No-op implementation
• Validation: Compiles, does nothing
• Status: ✅ Complete
• Confidence: High

• [x] 7.5.3 Implement `TestMetrics` for testing
• Owner: Agent
• Input: `IncidentMetrics` trait
• Output: Test implementation with counters
• Validation: Tracks call counts for verification
• Status: ✅ Complete
• Confidence: High

• [x] 7.5.4 Define Prometheus metric name constants
• Owner: Agent
• Input: Prometheus naming conventions
• Output: Constants in `incident.rs`
• Validation: Names follow `graph_kernel_*` pattern
• Status: ✅ Complete
• Confidence: High

7.6 Testing

• [x] 7.6.1 Unit tests for severity (1 test)
• Owner: Agent
• Input: `Severity` enum
• Output: Tests response time mapping
• Validation: Each severity has correct response time
• Status: ✅ Complete
• Confidence: High

• [x] 7.6.2 Unit tests for incident type (2 tests)
• Owner: Agent
• Input: `IncidentType` enum
• Output: Tests for severity and invariant mapping
• Validation: Correct mappings for all variants
• Status: ✅ Complete
• Confidence: High

• [x] 7.6.3 Unit tests for incident (2 tests)
• Owner: Agent
• Input: `Incident` struct
• Output: Tests for creation and acknowledgment
• Validation: Timestamps set correctly
• Status: ✅ Complete
• Confidence: High

• [x] 7.6.4 Unit tests for quarantine (2 tests)
• Owner: Agent
• Input: `QuarantinedToken` struct
• Output: Tests for creation and review
• Validation: Review sets all fields correctly
• Status: ✅ Complete
• Confidence: High

• [x] 7.6.5 Unit tests for metrics (2 tests)
• Owner: Agent
• Input: `IncidentMetrics` implementations
• Output: Tests for metric names and TestMetrics
• Validation: Names correct, TestMetrics tracks calls
• Status: ✅ Complete
• Confidence: High

7.7 Integration

• [x] 7.7.1 Add module to `types/mod.rs`
• Owner: Agent
• Input: `incident.rs`
• Output: `pub mod incident;` + re-exports
• Validation: Module accessible from `types::`
• Status: ✅ Complete
• Confidence: High

• [x] 7.7.2 Add re-exports to `lib.rs`
• Owner: Agent
• Input: Incident types
• Output: `pub use types::incident::*`
• Validation: Types accessible from crate root
• Status: ✅ Complete
• Confidence: High

---

Blocking Issues

None currently

---

Phase 8: Last Mile Integration ✅ COMPLETE

Goal: Wire security primitives into the hot path—API boundary enforcement, content hash verification, performance benchmarks.

8.1 API Boundary Enforcement (INV-GK-003)

• [x] 8.1.1 Change `ContextSlicer::slice()` to return `AdmissibleEvidenceBundle`
• Owner: Agent
• Input: `slicer.rs`, `AdmissibleEvidenceBundle`
• Output: Modified `slice()` return type
• Validation: Cannot get unverified slice from slicer
• Status: ✅ Complete
• Confidence: High

• [x] 8.1.2 Update service routes for new return type
• Owner: Agent
• Input: `service/routes.rs`
• Output: Handlers extract slice from bundle for serialization
• Validation: Routes still return valid JSON responses
• Status: ✅ Complete
• Confidence: High

• [x] 8.1.3 Update batch slicer for new return type
• Owner: Agent
• Input: `atlas/batch_slicer.rs`
• Output: `slice_all()` handles bundles correctly
• Validation: Atlas tests pass
• Status: ✅ Complete
• Confidence: High

8.2 Content Hash Verification (INV-GK-004)

• [x] 8.2.1 Add `verify_content_hash()` to `TurnSnapshot`
• Owner: Agent
• Input: `types/turn.rs`, `canonical_content.rs`
• Output: Method that verifies content matches stored hash
• Validation: Unit tests for valid, mismatch, and missing hash cases
• Status: ✅ Complete
• Confidence: High

• [x] 8.2.2 Add `ContentHashError` type
• Owner: Agent
• Input: Error requirements
• Output: Error enum with `Mismatch` variant
• Validation: Error contains turn_id, stored hash, computed hash
• Status: ✅ Complete
• Confidence: High

• [x] 8.2.3 Add `get_turn_with_verified_content()` to PostgresGraphStore
• Owner: Agent
• Input: `store/postgres.rs`
• Output: Method that fetches turn + content and verifies hash
• Validation: Returns error on mismatch, logs warning for legacy data
• Status: ✅ Complete
• Confidence: High

8.3 Performance Benchmarks (INV-GK-006)

• [x] 8.3.1 Add criterion benchmark infrastructure
• Owner: Agent
• Input: `Cargo.toml`
• Output: criterion dev-dependency, bench target
• Validation: `cargo bench` runs
• Status: ✅ Complete
• Confidence: High

• [x] 8.3.2 Create verification benchmark
• Owner: Agent
• Input: `TokenVerifier` implementation
• Output: `benches/verification.rs`
• Validation: Benchmarks cold, cached, and concurrent verification
• Status: ✅ Complete
• Confidence: High

8.4 Test Updates

• [x] 8.4.1 Update golden tests for new return type
• Owner: Agent
• Input: `tests/golden.rs`
• Output: Tests use `bundle.slice()` pattern
• Validation: All 13 golden tests pass
• Status: ✅ Complete
• Confidence: High

• [x] 8.4.2 Update atlas integration tests
• Owner: Agent
• Input: `tests/atlas_integration.rs`
• Output: Tests converted to async, use Arc wrapping
• Validation: All 14 integration tests pass
• Status: ✅ Complete
• Confidence: High

---

Completion Tracking

Overall Progress: 104/105 tasks (99

Test Count: 160 tests passing (131 unit + 14 integration + 13 golden + 2 doc)

Database Status (2026-01-03):
- `content_hash` column: ✅ Exists in `memory_turns`
- DB trigger: ✅ `content_hash_trigger` on INSERT/UPDATE
- Legacy backfill: ✅ 97,110 turns with hash (90.55
- Empty content: 10,135 turns with `content_text = ''` → NULL hash (correct)

Remaining Deferred Item:
- 2.3.1: Add content hash to `TurnSnapshot::new` (low priority—DB trigger handles this)