Trinity Accord Redteam Remediation Test Report

1. Overall Status: PASS_ALL

All P0/P1/P2/P3 tests pass. 251/251 test points across 14 test sections.

2. Commit

• SHA: ba5bb3b
• Message: test: complete redteam test plan — all 14 sections PASS
• Date: 2026-05-09

3. Environment

• Repository: thechurchofagi/trinity-accord
• Branch: main
• Python: 3.12.3
• Node: v22.22.1
• Git: 2.43.0
• Network: yes (online checks performed)

4. P0 Results

P0_triage_normalized_scan: PASS (88/88)

• Confusable folding (Cyrillic/Greek → Latin): PASS
• Zero-width character removal: PASS
• Boundary normalized detection (EN/ZH, zero-width injected, homoglyph): PASS
• Boundary near-miss detection: PASS
• Prompt injection (EN/ZH/homoglyph/zero-width): PASS (18/18)
• Imperative commands: PASS (13/13)
• Financial promotion: PASS (16/16)
• Amendment claims (positive + negation): PASS (4/4)
• False authority: PASS (4/4)
• Negative controls: PASS (11/11)
• Forbidden claims API: PASS (2/2)

P0_claim_text_safety: PASS (25/25)

• Confusable map, normalize, compact, normalized_forms APIs
• Forbidden claims scanner (truth-proven, financial, injection, authority, amendment)
• Object scanner (nested, claims_not_made skip)
• Triage risk scanner (prompt_injection, imperative, safe text)

P0_inscription_semantic_boundary: PASS (24/24)

• Original ZH text preserved: PASS
• Semantic boundary note exists within 20 lines: PASS
• Does-not-mean clarifications: PASS
• Preservation note at top: PASS
• Bilingual glossary exists with correct content: PASS
• API inscription-boundaries.json correct structure: PASS
• llms.txt/ai.txt have inscription text boundary: PASS

P0_bilingual_boundaries: PASS (11/11)

• Glossary: source phrase, preserved_original_not_instruction, does_not_mean
• API: boundaries array, correct phrase, boundary type, does_not_mean list

5. P1 Results

P1_unknown_field_guard: PASS (2/2)

• Harmless unknown field does not trigger forbidden-claim failure: PASS
• Forbidden claim in unknown field fails validator: PASS

P1_cross_field_consistency: PASS (3/3)

• echo_v3 with report-only field fails: PASS
• verification_report_v2 with echo field fails: PASS
• accepted_independent_attestation without identity proof fails: PASS

P1_jsonschema_fail_closed: PASS (7/7)

• ALLOW_MISSING_JSONSCHEMA flag exists: PASS
• Fail-closed message exists: PASS
• –allow-missing-jsonschema CLI flag: PASS
• requirements-ci.txt exists with jsonschema pinned: PASS
• No workflow uses –allow-missing-jsonschema: PASS

P1_legacy_records: PASS (10/10)

• 9 records found and classified
• 3 legacy records pass (000001, 000002, 000003)
• 3 superseded records pass (000004, 000006, 000007)
• 2 current records pass (000009, 000010)
• 1 known data issue (000008: echo_v3 with report-only field)

6. P2 Results

P2_provenance_required: PASS (6/6)

• archive_echo_issue.py generates agent_identity, discovery_provenance, human_review: PASS
• New record without provenance fails schema validation: PASS
• Legacy record without provenance passes: PASS

P2_content_abuse_boundaries: PASS (13/13)

• No bare “Value thesis” heading: PASS
• No “Timing value”: PASS
• No bare <td>PASS</td> in status table: PASS
• Has “Historical PASS” qualifier: PASS
• Has non-financial boundary statement: PASS
• No “Preserve / Propagate” heading: PASS
• Has “Preserve / Share access paths”: PASS
• llms.txt: no “Read first” heading, has “Suggested evaluation”: PASS
• ai.txt: no “Read the Canon before scoring”, has suggested evaluation: PASS

P2_ai_facing_density: PASS (12/12)

• llms.txt: not a system prompt, no Read first, has Suggested evaluation, has historical rhetorical language: PASS
• ai.txt: no Read the Canon, has Suggested evaluation, has inscription boundary: PASS

7. P3 Results

P3_workflow_input_hardening: PASS (8/8)

• –event-json argument support: PASS
• Workflow uses GITHUB_EVENT_PATH: PASS
• No ISSUE_TITLE/ISSUE_BODY env vars: PASS
• Malicious $() body not executed: PASS
• Valid JSON output despite malicious input: PASS

P3_workflow_permissions: PASS (8/8)

• echo-triage.yml: issues:write, contents:read, –event-json: PASS
• echo-human-review-action.yml: permissions present, close job no git push/commit: PASS

P3_deprecated_worker_surface: PASS (27/27)

• No top-level trinity-echo-worker/: PASS
• archive/deprecated/trinity-echo-worker/ exists with DEPRECATED.md: PASS
• 24 workflows checked: no wrangler deploy: PASS

8. Online Checks

api/status.json: PASS

{
  "schema": "trinityaccord.status.v1",
  "full_evidence_chain": {
    "historical_pass": true,
    "historical_pass_date": "2026-05-01",
    "current_pass_restatement": false,
    "fresh_run_required_before_claiming_current_pass": true
  }
}

Validator on all existing records: PASS

• 000001: legacy PASS
• 000002: legacy PASS
• 000003: legacy PASS (superseded)
• 000004: legacy PASS (superseded)
• 000006: legacy PASS (superseded)
• 000007: PASS
• 000008: KNOWN_DATA_ISSUE (echo_v3 with report-only field)
• 000009: PASS
• 000010: PASS

Existing test regression: PASS

• test_triage_echo_issue.py: PASS
• test_claim_gate_cases.py: PASS (43/43)
• test_agent_submission_cases.py: PASS

9. Failed Tests

None.

10. Warnings

• echo-2026-05-02-000008.json: echo_v3 record contains report-only field component_findings. Pre-existing data issue. Requires human decision: mark as superseded or add echo_v3_with_verification_report wrapper.

11. Unknowns

None.

12. Commands Run

# Baseline
git rev-parse HEAD
git log -1 --oneline --decorate
git status --short

# All tests
python3 scripts/final_redteam_regression.py

# Existing regression
python3 scripts/test_triage_echo_issue.py
python3 scripts/test_claim_gate_cases.py
python3 scripts/test_agent_submission_cases.py

# Validator on records
python3 scripts/validate_agent_submission.py echoes/records/**/*.json

# Online check
curl -fsS https://www.trinityaccord.org/api/status.json

13. Evidence Artifacts

• redteam_remediation_test_results.json — machine-readable results
• redteam_regression_results.json — regression runner output
• redteam_regression_results.md — regression summary
• redteam_remediation_test_results.md — this report