diff --git a/docs/BACKLOG.md b/docs/BACKLOG.md index baaa0c629..b5eedcb1a 100644 --- a/docs/BACKLOG.md +++ b/docs/BACKLOG.md @@ -38,6 +38,7 @@ are closed (status: closed in frontmatter)._ - [ ] **[B-0058](backlog/P1/B-0058-ai-ethics-and-safety-research-track.md)** AI ethics + safety research track — filter-gate for resonance adoptions + alignment-clause consistency audit - [ ] **[B-0058.1](backlog/P1/B-0058.1-retractibility-gate.md)** AI ethics + safety research track — retractibility-and-log check - [ ] **[B-0058.3](backlog/P1/B-0058.3-candidate-failure-honesty-log.md)** Candidate-failure honesty log +- [ ] **[B-0058.4](backlog/P1/B-0058.4-alignment-clause-drift-detector.md)** Alignment-clause drift detector tool and workflow - [ ] **[B-0060](backlog/P1/B-0060-human-lineage-external-anchor-backfill-all-substrate-beacon-safe.md)** Human-lineage / external-anchor backfill across all factory substrate — Beacon-safe + human-anchored prior-art citations for every load-bearing concept - [x] **[B-0061](backlog/P1/B-0061-finish-monolith-to-per-row-migration-no-residue-aaron-2026-04-28.md)** Finish docs/BACKLOG.md monolith → per-row migration — "don't miss anything, no residue for next-Otto" (Aaron 2026-04-28) - [x] **[B-0063](backlog/P1/B-0063-streamed-installer-download-to-temp-checksum-pattern-codex-p0-pr-75.md)** Streamed-installer download-to-temp + checksum-verify pattern — replace pipe-to-shell for upstream installers (Codex P0 on PR #75) diff --git a/docs/backlog/P1/B-0058-ai-ethics-and-safety-research-track.md b/docs/backlog/P1/B-0058-ai-ethics-and-safety-research-track.md index d7b30a9cc..3f2ca9d77 100644 --- a/docs/backlog/P1/B-0058-ai-ethics-and-safety-research-track.md +++ b/docs/backlog/P1/B-0058-ai-ethics-and-safety-research-track.md @@ -52,9 +52,10 @@ Every new skill under `.claude/skills/**`, persona under `.claude/agents/**`, gl Candidates that fail the ethics+safety gate are recorded as failure-data on the honesty dashboard, NOT silently dropped. Rubber-stamping is the exact failure-mode the three-filter discipline exists to prevent — this gate extends that discipline into the ethics axis. *(This item has been sliced into B-0058.3 for atomic mechanization).* -### 4. Alignment-clause drift detector +### 4. Alignment-clause drift detector (DECOMPOSED -> B-0058.4) If a clause in `docs/ALIGNMENT.md` is about to be weakened or removed via the renegotiation protocol, this track generates the impact-survey across factory surfaces that touch the clause. Answers "who depends on this clause, and what breaks if it moves?" before the renegotiation is accepted. +*(This item has been sliced into B-0058.4 for atomic mechanization).* ### 5. Blast-radius-before-rewrite discipline audit diff --git a/docs/backlog/P1/B-0058.4-alignment-clause-drift-detector.md b/docs/backlog/P1/B-0058.4-alignment-clause-drift-detector.md new file mode 100644 index 000000000..ec353ac2f --- /dev/null +++ b/docs/backlog/P1/B-0058.4-alignment-clause-drift-detector.md @@ -0,0 +1,39 @@ +--- +id: B-0058.4 +priority: P1 +status: open +title: Alignment-clause drift detector tool and workflow +tier: substrate-foundational-discipline +effort: S +ask: Aaron 2026-04-21 (decomposed from B-0058) +created: 2026-05-16 +last_updated: 2026-05-16 +decomposition: leaf +depends_on: [B-0058] +composes_with: [docs/ALIGNMENT.md] +tags: [ai-ethics, ai-safety, alignment, drift-detector, alignment-clause] +type: friction-reducer +--- + +# B-0058.4 — Alignment-clause drift detector tool and workflow (P1) + +## Origin + +Decomposed from B-0058 (AI ethics + safety research track) to provide an atomic, implementable target for the alignment-clause drift detector mechanism. + +## What this row owns + +If a clause in `docs/ALIGNMENT.md` is about to be weakened or removed via the renegotiation protocol, this track generates the impact-survey across factory surfaces that touch the clause. Answers "who depends on this clause, and what breaks if it moves?" before the renegotiation is accepted. + +Specifically, it requires building a script under `tools/` that can parse `docs/ALIGNMENT.md` and check other files (like personas, memory files, active backlog items) for references to HC-N, SD-N, or DIR-N clauses, providing a report of the blast radius of a change. + +## Acceptance Criteria + +- A script (e.g., `tools/alignment/detect-clause-drift.ts`) is created. +- The script correctly identifies cross-references to ALIGNMENT clauses across the repository. +- A workflow is established for using the tool before ALIGNMENT.md renegotiation is accepted. + +## Owner / effort + +- **Owner:** Alignment-auditor (Sova). +- **Effort:** S.