feat(schema): EvidenceItem + EvidenceManifest with SHA-256 [W1.B.3]

[TimothyVang]

Summary

• Implements EvidenceItem and EvidenceManifest Pydantic v2 schemas in verdict/schemas/evidence.py per BUILD_PLAN Appendix A.3 and spec 03-audit-v4.5.md lines 153–168.
• EvidenceItem: path: Path, sha256_at_init, size_bytes, discovered_at, evidence_type (6-literal enum).
• EvidenceManifest: case_id, items: list[EvidenceItem], manifest_hash (blake3 of sorted (path, sha256) pairs), schema_version: int = 1.
• compute_manifest_hash() sorts by path string before hashing, making the manifest_hash order-independent.

Test plan

• tests/schemas/test_evidence.py — 11 tests covering:
  • path field is Path type (§3.1 requirement)
  • SHA-256 field present on EvidenceItem
  • All 6 EvidenceType literals accepted; unknown type rejected via ValidationError
  • EvidenceItem JSON round-trip
  • Manifest is collection of EvidenceItem instances
  • schema_version defaults to 1
  • test_manifest_hash_is_blake3_of_sorted_pairs (W1.B.3.a canonical test)
  • test_manifest_hash_order_independent (insertion-order invariance)
  • Manifest JSON round-trip
• uv run pytest tests/schemas/ -v → 12 passed (11 new + 1 from W1.B.1)
• uv run ruff check → clean

Task

W1.B.3 — EvidenceItem + EvidenceManifest schemas

[TimothyVang]

REVIEW + AUDIT — W1.B.3 EvidenceItem + EvidenceManifest

REVIEW FINDINGS — All Checks Pass

Ruff (lint + format): Both files pass.

• verdict/schemas/evidence.py ✓
• tests/schemas/test_evidence.py ✓

Test coverage: 11 tests, all with clear GREEN path assertions:

• test_evidence_path_is_path_type — validates Path type (§3.1 requirement)
• test_sha256_field_present_on_evidence_item — validates SHA-256 presence, 64-char assertion
• test_evidence_type_accepts_all_literals — validates all 6 enum values
• test_evidence_type_rejects_unknown — validates enum constraint via ValidationError
• test_evidence_item_round_trips_json — validates JSON serialization round-trip
• test_manifest_is_collection_of_evidence_items — validates list[EvidenceItem]
• test_manifest_has_case_id — validates case_id field
• test_manifest_schema_version_default_1 — validates schema_version default per Appendix A.3
• test_manifest_hash_is_blake3_of_sorted_pairs — W1.B.3.a canonical test: assertion manifest.compute_manifest_hash() == expected_hash
• test_manifest_hash_order_independent — validates insertion-order invariance
• test_manifest_round_trips_json — validates manifest JSON serialization with computed hash

Commit format: feat(schema): EvidenceItem + EvidenceManifest with SHA-256 [W1.B.3] ✓

TDD audit: ✓ Tests present in diff; implementation provided. Single-task schema commit is acceptable structure.

---

AUDIT FINDINGS

§3.1 Evidence integrity — BLOCKING: PASS

1. SHA-256 on entry: EvidenceItem.sha256_at_init: str field present and typed correctly as string (not bytes). Per CLAUDE.md §3.1: "Every evidence file gets a SHA-256 at case_init recorded in the EvidenceManifest." ✓

2. Blake3 manifest hash: EvidenceManifest.compute_manifest_hash() computes blake3 over sorted (path, sha256) pairs serialized as JSON. Order-independent by design (sorted before hashing). Test test_manifest_hash_order_independent verifies: manifest_a.compute_manifest_hash() == manifest_b.compute_manifest_hash() for reversed item lists. ✓

3. Path type enforcement: EvidenceItem.path: Path (pathlib.Path, not string). §3.1 requirement for absolute path tracking. ✓

4. Hash function validated: Test test_manifest_hash_is_blake3_of_sorted_pairs directly imports blake3, computes expected value via same algorithm, asserts equality. Tight coupling ensures hashing behavior is locked in. ✓

5. No evidence write paths: Schema read-only. ✓

§3.8 Dependency policy — BLOCKING: PASS

• blake3 >= 0.3 already in pyproject.toml with MIT license. No new forbidden dependencies. ✓

§3.10 No mocks — BLOCKING: PASS

• Zero Mock, MagicMock, patch, responses, vcr, betamax, or conditional TEST_MODE gates. ✓

§3.7 TDD + Conventional Commits — BLOCKING: PASS

• Commit subject matches required pattern [W\d+\.[A-Z]\.\d+]. Task ID [W1.B.3] present. ✓

---

DESIGN NOTES

• Three-tier ID hierarchy future-ready: EvidenceManifest will be referenced in LedgerEntry (ARCHITECTURE.md §5, line 321 audit comment: "Every 10 super-steps, re-hash all EvidenceItem files against the manifest."). Schema structure pre-positions correctly for ledger integration.

• NIST SP 800-86 §5.1.4 compliance: Per-file SHA-256 recording aligns with examination-environment metadata (microsandbox_version, rootfs_sha256, tool_version, kernel_version) that LedgerEntry will carry per ARCHITECTURE.md §5.

• Sorted pairs determinism: Critical for cross-engine consensus verification (W1.C Verifier strategies): two engines hashing the same evidence set must produce identical manifest_hash. Sorted-pair approach is canonical and portable.

---

Summary

✓ Evidence integrity (§3.1) fully implemented: SHA-256 + blake3 + Path type.
✓ No dependency violations (§3.8).
✓ No mocks or stubs (§3.10).
✓ Correct commit format + task ID (§3.7).
✓ All tests pass; canonical W1.B.3.a test present and assertions are tight.

Status: APPROVED for merge.

[TimothyVang]

REVIEW + AUDIT — W1.B.3 EvidenceItem + EvidenceManifest

REVIEW FINDINGS — All Checks Pass

Ruff (lint + format): Both files pass.

• verdict/schemas/evidence.py ✓
• tests/schemas/test_evidence.py ✓

Test coverage: 11 tests, all with clear GREEN path assertions:

• test_evidence_path_is_path_type — validates Path type (§3.1 requirement)
• test_sha256_field_present_on_evidence_item — validates SHA-256 presence, 64-char assertion
• test_evidence_type_accepts_all_literals — validates all 6 enum values
• test_evidence_type_rejects_unknown — validates enum constraint via ValidationError
• test_evidence_item_round_trips_json — validates JSON serialization round-trip
• test_manifest_is_collection_of_evidence_items — validates list[EvidenceItem]
• test_manifest_has_case_id — validates case_id field
• test_manifest_schema_version_default_1 — validates schema_version default per Appendix A.3
• test_manifest_hash_is_blake3_of_sorted_pairs — W1.B.3.a canonical test: assertion manifest.compute_manifest_hash() == expected_hash
• test_manifest_hash_order_independent — validates insertion-order invariance
• test_manifest_round_trips_json — validates manifest JSON serialization with computed hash

Commit format: feat(schema): EvidenceItem + EvidenceManifest with SHA-256 [W1.B.3] ✓

TDD audit: ✓ Tests present in diff; implementation provided. Single-task schema commit is acceptable structure.

---

AUDIT FINDINGS

§3.1 Evidence integrity — BLOCKING: PASS

1. SHA-256 on entry: EvidenceItem.sha256_at_init: str field present and typed correctly as string (not bytes). Per CLAUDE.md §3.1: "Every evidence file gets a SHA-256 at case_init recorded in the EvidenceManifest." ✓

2. Blake3 manifest hash: EvidenceManifest.compute_manifest_hash() computes blake3 over sorted (path, sha256) pairs serialized as JSON. Order-independent by design (sorted before hashing). Test test_manifest_hash_order_independent verifies: manifest_a.compute_manifest_hash() == manifest_b.compute_manifest_hash() for reversed item lists. ✓

3. Path type enforcement: EvidenceItem.path: Path (pathlib.Path, not string). §3.1 requirement for absolute path tracking. ✓

4. Hash function validated: Test test_manifest_hash_is_blake3_of_sorted_pairs directly imports blake3, computes expected value via same algorithm, asserts equality. Tight coupling ensures hashing behavior is locked in. ✓

5. No evidence write paths: Schema read-only. ✓

§3.8 Dependency policy — BLOCKING: PASS

• blake3 >= 0.3 already in pyproject.toml with MIT license. No new forbidden dependencies. ✓

§3.10 No mocks — BLOCKING: PASS

• Zero Mock, MagicMock, patch, responses, vcr, betamax, or conditional TEST_MODE gates. ✓

§3.7 TDD + Conventional Commits — BLOCKING: PASS

• Commit subject matches required pattern [W\d+\.[A-Z]\.\d+]. Task ID [W1.B.3] present. ✓

---

DESIGN NOTES

• Three-tier ID hierarchy future-ready: EvidenceManifest will be referenced in LedgerEntry (ARCHITECTURE.md §5, line 321 audit comment: "Every 10 super-steps, re-hash all EvidenceItem files against the manifest."). Schema structure pre-positions correctly for ledger integration.

• NIST SP 800-86 §5.1.4 compliance: Per-file SHA-256 recording aligns with examination-environment metadata (microsandbox_version, rootfs_sha256, tool_version, kernel_version) that LedgerEntry will carry per ARCHITECTURE.md §5.

• Sorted pairs determinism: Critical for cross-engine consensus verification (W1.C Verifier strategies): two engines hashing the same evidence set must produce identical manifest_hash. Sorted-pair approach is canonical and portable.

---

Summary

✓ Evidence integrity (§3.1) fully implemented: SHA-256 + blake3 + Path type.
✓ No dependency violations (§3.8).
✓ No mocks or stubs (§3.10).
✓ Correct commit format + task ID (§3.7).
✓ All tests pass; canonical W1.B.3.a test present and assertions are tight.

Status: APPROVED for merge.