Verified Autonomy alignment — tracking issue

[antnewman]

Tracking issue for the Verified Autonomy gap-closure work.

Source paper: Newman et al., Verified Autonomy: A Field Guide to Engineering Trust in AI Systems, May 2026.
DOI: 10.5281/zenodo.19096229

Gap analysis: see PDA-platform-gap-analysis-verified-autonomy.docx in the repo root, and the plan at ~/.claude/plans/serene-sleeping-globe.md.

Summary. The paper presents a nine-layer defence-in-depth architecture for engineering trust in AI systems. PDA Platform is cited as the production reference for Layers 1, 2, and 3, with partial gaps where the paper-grade architecture is not surfaced to consumers. Layers 4, 6, 7 are absent. Layers 5 and 8 have substantive primitives but with scope gaps (Layer 5 has no rule-engine gate between AI-authored output and consumer; Layer 8's hash chain only covers NISTA submissions).

This issue tracks 37 child issues across two phases:

• Phase 2 — Foundations (8 PRs). Each introduces a primitive that integrations consume.
• Phase 3 — Integrations (29 PRs). Each wires a foundation into a specific tool or module.

Conventions

• Branch off dev; PR to dev; never push to main.
• Commit author: antnewman <antjsnewman@outlook.com>. Never Co-Authored-By: Claude.
• One regression-test class per PR.
• Each Phase 3 PR scoped to a single tool/module.

Progress

(Child issues are added to the task list below by scripts/create_va_issues.py.)

Phase 2 — Foundations

• A1 · L8 foundation — Generic AuditChain primitive #28 · L8 foundation — Generic AuditChain primitive
• A2 · L5 foundation — Output-guardrail policy engine #29 · L5 foundation — Output-guardrail policy engine
• A3 · L6 foundation — Groundedness checker #30 · L6 foundation — Groundedness checker
• A4 · L4 foundation — Calibration and conformal-prediction primitives #31 · L4 foundation — Calibration and conformal-prediction primitives
• A5 · L2 foundation — Four-tier escalation router with OR fail-safe #32 · L2 foundation — Four-tier escalation router with OR fail-safe
• A6 · L3 foundation — Quality-score and potential-hallucinations primitives #33 · L3 foundation — Quality-score and potential-hallucinations primitives
• A7 · L7 foundation — RedTeam harness #34 · L7 foundation — RedTeam harness
• A8 · L9 foundation — Confidence→RAG monotonicity verification kernel #35 · L9 foundation — Confidence→RAG monotonicity verification kernel

Phase 3 — Integrations

• B1 · L5 integration — generate_board_exception_report guardrail #36 · L5 integration — generate_board_exception_report guardrail
• B2 · L5 integration — generate_assumption_report guardrail #37 · L5 integration — generate_assumption_report guardrail
• B3 · L5 integration — generate_gate_review_summary guardrail #38 · L5 integration — generate_gate_review_summary guardrail
• B4 · L5 integration — generate_narrative guardrail #39 · L5 integration — generate_narrative guardrail
• B5 · L5 integration — generate_premortem_questions guardrail #40 · L5 integration — generate_premortem_questions guardrail
• B6 · L5 integration — generate_benefits_narrative guardrail #41 · L5 integration — generate_benefits_narrative guardrail
• B7 · L5 integration — generate_portfolio_summary guardrail #42 · L5 integration — generate_portfolio_summary guardrail
• B8 · L5 integration — generate_lessons_section guardrail #43 · L5 integration — generate_lessons_section guardrail
• B9 · L5 integration — generate_pir_template guardrail #44 · L5 integration — generate_pir_template guardrail
• B10 · L6 integration — generate_board_exception_report groundedness #45 · L6 integration — generate_board_exception_report groundedness
• B11 · L6 integration — generate_assumption_report groundedness #46 · L6 integration — generate_assumption_report groundedness
• B12 · L6 integration — generate_gate_review_summary groundedness #47 · L6 integration — generate_gate_review_summary groundedness
• B13 · L6 integration — generate_narrative groundedness #48 · L6 integration — generate_narrative groundedness
• B14 · L6 integration — generate_premortem_questions groundedness #49 · L6 integration — generate_premortem_questions groundedness
• B15 · L6 integration — generate_benefits_narrative groundedness #50 · L6 integration — generate_benefits_narrative groundedness
• B16 · L6 integration — generate_portfolio_summary groundedness #51 · L6 integration — generate_portfolio_summary groundedness
• B17 · L6 integration — generate_lessons_section groundedness #52 · L6 integration — generate_lessons_section groundedness
• B18 · L6 integration — generate_pir_template groundedness #53 · L6 integration — generate_pir_template groundedness
• B19 · L8 integration — pm-assure audit chain #54 · L8 integration — pm-assure audit chain
• B20 · L8 integration — pm-assumptions audit chain #55 · L8 integration — pm-assumptions audit chain
• B21 · L8 integration — pm-reporting audit chain #56 · L8 integration — pm-reporting audit chain
• B22 · L8 integration — pm-knowledge audit chain #57 · L8 integration — pm-knowledge audit chain
• B23 · L8 integration — pm-simulation audit chain #58 · L8 integration — pm-simulation audit chain
• B24 · L4 integration — evaluate_calibration MCP tool #59 · L4 integration — evaluate_calibration MCP tool
• B25 · L4 integration — Monte Carlo conformal bands #60 · L4 integration — Monte Carlo conformal bands
• B26 · L4 integration — Reference-class calibration #61 · L4 integration — Reference-class calibration
• B27 · L2 surfacing — escalation router MCP tool #62 · L2 surfacing — escalation router MCP tool
• B28 · L3 surfacing — hallucinations list on AI tools #63 · L3 surfacing — hallucinations list on AI tools
• B29 · L1 surfacing — surface confidence gap to consumers #64 · L1 surfacing — surface confidence gap to consumers