Lucent-Financial-Group · AceHack · Apr 23, 2026 · Apr 23, 2026 · Apr 23, 2026 · Apr 23, 2026
diff --git a/docs/FACTORY-HYGIENE.md b/docs/FACTORY-HYGIENE.md
@@ -94,6 +94,7 @@ is never destructive; retiring one requires an ADR in
 | 49 | Post-setup script stack audit (bun+TS default; bash only under exempt paths or with exception label) | Author-time (every new `tools/**/*.{sh,ps1}` decision-flow walk per `docs/POST-SETUP-SCRIPT-STACK.md`) + cadenced detection every 5-10 rounds (same cadence as skill-tune-up / row #38 / harness-surface audit) + opportunistic on-touch (every time an agent adds or edits a script under `tools/`). | Author of the script (self-check at author-time against the decision-flow doc); Dejan (devops-engineer) on the cadenced detection sweep; Kenji (Architect) on migration-order decisions when multiple violations stack up. | both | **Author-time prevention:** walk the three-question flow in `docs/POST-SETUP-SCRIPT-STACK.md` before writing any new `tools/**/*.{sh,ps1}` — (Q1) pre-setup? → `tools/setup/` bash+PowerShell exempt; (Q2) skill-bundled? → skill-compatibility rules govern, not this row; (Q3) default bun+TypeScript unless an explicit exception (trivial pipeline / thin CLI wrapper / bash scaffolding / sibling-migration guardrail) applies, in which case the script MUST carry a header comment naming the exception. **Cadenced detection:** `tools/hygiene/audit-post-setup-script-stack.sh` lists every shell/PowerShell script under `tools/` and classifies each as `exempt` / `labelled-exception` / `violation`. Exit-2 on any new violation; CI / pre-commit-eligible. **Why both layers:** prevention catches new violations at author-time (cheap); detection catches drift — labels getting stripped on edits, exceptions becoming stale, scripts moving out of exempt paths. Ships to project-under-construction: adopters inherit the canonical-stack rule + the audit script + the decision-flow doc. Aaron 2026-04-22 triggering-directive-chain: *"if post setup backlog bun/ts"* → *"now add someting that will try to prevent that and and hygene it if it happens again"*. | Author-time: commit-message rationale for any new `.sh` under `tools/` outside `tools/setup/`, OR exception-label header in the script, OR BACKLOG row queuing bun+TS migration. Cadenced: audit script output (markdown), appended to `docs/hygiene-history/post-setup-script-stack-history.md` (per-fire schema per row #44); BACKLOG row per unlabeled violation. | `docs/POST-SETUP-SCRIPT-STACK.md` (prevention surface) + `tools/hygiene/audit-post-setup-script-stack.sh` (detection surface) + `memory/project_ui_canonical_reference_bun_ts_backend_cutting_edge_asymmetry` + `memory/project_bun_ts_post_setup_low_confidence_watchlist` |
 | 50 | Missing-prevention-layer meta-audit (every hygiene row carries a prevention classification: prevention-bearing / detection-only-justified / detection-only-gap) | Round cadence (same as rows #22 / #23 / #35 / #36) + opportunistic on-touch (every time a new row is added to `docs/FACTORY-HYGIENE.md` the author classifies it at landing). Not exhaustive; the round-close sweep catches un-classified rows and gap rows. | Architect (Kenji) on round-cadence classification review + gap-closure ROI assessment. All agents (self-administered) on on-touch: every new hygiene row MUST declare its prevention classification at landing; an unclassified row is itself a violation of this row. | factory | Sweep every row in `docs/FACTORY-HYGIENE.md` and classify each as one of: (a) **prevention-bearing** — an author-time / commit-time / trigger-time mechanism (hook, CI check, decision-flow doc, pre-commit lint, skill-gate) blocks or warns the violation BEFORE it materialises; (b) **detection-only-justified** — the class is fundamentally post-hoc (e.g., cadence-history row #44 — a fire-log can only exist AFTER the fire happens; wake-friction row #29 — friction is only observable at wake-time); (c) **detection-only-gap** — no principled reason the row is detection-only; a prevention layer COULD and SHOULD be built. Classification lives in `docs/hygiene-history/prevention-layer-classification.md` (one table row per hygiene row). **Why this row exists:** Aaron 2026-04-22 *"add a hygene for missing prevention layers"* — the factory had been quietly accumulating detection-only rows without asking the complementary question "could we have prevented this at author-time?". Without this meta-audit, the factory's reactive-cost grows silently. Parallels the existing meta-hygiene triangle (row #23 unknown-classes / #43 authored-but-unactivated / #44 cadence-history) by adding a fourth: row #47 is *"of the rows that ARE active and firing, which could have been prevented upstream"*. **Classification:** this is an **intentionality-enforcement** hygiene rule (Aaron 2026-04-22 tick-close: *"we are enforcing intentional decsions"*) — the audit cannot compute whether a row's classification is correct, but it forces every new hygiene row to carry an explicit prevention-vs-detection decision at landing. Declining to classify is itself the violation. See `memory/feedback_enforcing_intentional_decisions_not_correctness.md`. Ships to project-under-construction: adopters inherit the classification discipline + the meta-audit script + the obligation to classify any new hygiene row at landing. | `docs/hygiene-history/prevention-layer-classification.md` (classification matrix, one row per hygiene row) + cadenced audit run landed as `docs/hygiene-history/missing-prevention-layer-audit-YYYY-MM-DD.md` noting gap rows; ROUND-HISTORY row when a gap row gains a prevention layer (detection-only-gap → prevention-bearing transition); BACKLOG row per gap with prevention-design ROI estimate. | `tools/hygiene/audit-missing-prevention-layers.sh` + this row's self-reference (its own prevention layer is the at-landing-classify obligation declared in this Checks/enforces column) |
 | 51 | Cross-platform parity audit (bash / PowerShell / bun+TS twin check across macOS / Windows / Linux / WSL) | Detect-only now (landed 2026-04-22); cadenced detection every 5-10 rounds (same cadence as row #46); opportunistic on-touch every time an agent adds or edits a script under `tools/`. Enforcement deferred until baseline is green AND CI matrix runs `--enforce` on `macos-latest` / `windows-latest` / `ubuntu-latest` (WSL inherits ubuntu-latest for CI). | Dejan (devops-engineer) on cadenced detection; author of the script (self-check at author-time against the rule classes in the audit's decision-record header block). Kenji (Architect) on CI-matrix-enforcement sign-off when baseline is green. | both | `tools/hygiene/audit-cross-platform-parity.sh` classifies every script under `tools/` by rule class: (a) **pre-setup** (`tools/setup/**`) — both `.sh` AND `.ps1` required per Q1 dual-authoring rule (`memory/feedback_preinstall_scripts_forced_shell_meet_developer_where_they_live`); (b) **post-setup permanent-bash** (`thin wrapper over existing CLI` / `trivial find-xargs pipeline` / `stay bash forever`) — `.ps1` twin required per the Windows-twin obligation (`memory/feedback_stay_bash_forever_implies_powershell_twin_obligation.md`); (c) **post-setup transitional** (`bun+TS migration candidate` / `bash scaffolding`) — no twin obligation (long-term plan is one cross-platform bun+TS script); (d) **post-setup bun+TS** (`*.ts` under `tools/`) — no twin needed (cross-platform native via bun). `--summary` prints counts; `--enforce` flips exit 2 on gaps. **Why detect-only first:** baseline at first fire (2026-04-22) was 13 gaps — 12 pre-setup bash without `.ps1` twin (Q1 violation silently accumulating since `tools/setup/` existed) + 1 post-setup permanent-bash (`tools/profile.sh`) without `.ps1` twin. Turning enforcement on before triage would block every CI run. **Why this row exists:** Aaron 2026-04-22 *"missing mac/windows/linux/wsl parity (ubuntu latest) we can deffer but should have the hygene in place for when we want to enforce and it will be more obvious to you in the future that we are cross platform."* Cross-platform-first must be a *visible* factory property (audit exists, runs, prints the gap) before it becomes an enforced gate. Same pattern as FACTORY-HYGIENE rows #23 / #43 / #47. See `memory/feedback_cross_platform_parity_hygiene_deferred_enforcement.md`. **Classification (row #47):** **prevention-bearing** — the audit runs at author-time (opportunistic on-touch) and surfaces the gap before it lands, same as row #46. The audit itself is a detect-only mechanism but detect-only surfaces the obligation at author-time when the author runs it. Ships to project-under-construction: adopters inherit the parity audit + the decision-record-block pattern + the CI-matrix obligation once it's wired. | Audit output in repo root on each fire; cadenced runs appended to `docs/hygiene-history/cross-platform-parity-history.md` (per-fire schema per row #44); BACKLOG row per gap at triage time; ROUND-HISTORY row when a gap resolves. | `tools/hygiene/audit-cross-platform-parity.sh` (detection + decision-record header block) + `memory/feedback_cross_platform_parity_hygiene_deferred_enforcement.md` + `memory/feedback_stay_bash_forever_implies_powershell_twin_obligation.md` + `memory/feedback_preinstall_scripts_forced_shell_meet_developer_where_they_live` + `docs/POST-SETUP-SCRIPT-STACK.md` |
+| 53 | AutoDream cadenced consolidation (upstream Anthropic Q1 2026 + factory overlays A/B/C/D) | Anthropic's cadence verbatim: **≥24h AND ≥5 sessions** since last cycle. Gate marker lives at the top of `MEMORY.md` as `[AutoDream last run: YYYY-MM-DD]`. Session wake checks the gate silently; manual invocation ("dream" / "consolidate memory") may override with a maintainer confirmation. When Anthropic un-gates the backend (currently `tengu_onyx_plover`), the manual approximation retires and the automatic run takes over; factory overlays continue on top. | Architect (Kenji) on cadenced fires (dispatches an audit subagent; main agent applies only what the audit surfaces). Overlay D pulls the alignment-auditor role in on clause-touching consolidations. The autodream-hygiene skill (BACKLOG P1 follow-up) will wrap the four upstream phases + four factory overlays as a single invocable procedure once authored. | both | **Upstream four phases** (Orientation → Gather Signal → Consolidation → Prune & Index) run as Anthropic prescribes — factory does not replace them. **Factory overlays layered on top:** (A) cross-substrate mirror check before Orientation — scans per-user memory for generic factory-shaped rules that should live in-repo per the in-repo-preferred discipline (`memory/feedback_in_repo_preferred_over_per_user_memory_where_possible_2026_04_23.md`), migrates what qualifies with the signal-preservation rule; (B) harness-surface drift sync during Gather Signal — cross-checks `docs/HARNESS-SURFACES.md` adoption statuses for surfaces that have moved watched → adopted or cut since last cycle; (C) governance-promotion pass during Consolidation — ≥3 memories citing the same rule surfaces a BP-NN / ADR / GOVERNANCE promotion candidate for Kenji; (D) alignment-observability sync during Prune & Index — any memory edit that touched an `ALIGNMENT.md` clause (HC-1..HC-7 / SD-1..SD-8 / DIR-1..DIR-5) queues for the alignment-auditor time-series. **Invariants preserved:** Anthropic's four (load-bearing memories unconditionally / distinct-query-axes stay distinct / cross-references bidirectional / corrections recorded not deleted) plus factory additions (verbatim quotes stay verbatim / migration supersedes but does not delete per-user source / supersede markers over silent retirement). **Why this row exists:** Aaron 2026-04-23 *"continue our AutoDream reserach to make sure we are running it on a cadence and we plug in tightly with the existing claude Q1 2026 feature for AutoDream and AutoMemory, we should by defintion be an extension of theirs as it will get upgrades we want over time."* Ships to project-under-construction: adopters running Claude Code inherit AutoMemory / AutoDream from Anthropic; this row adds the factory's overlay policy (migration discipline, governance-promotion signal, alignment sync) that any adopter can opt into by following the research doc. **Classification (row #50):** **detection-only-justified** — consolidation is fundamentally post-hoc by Anthropic's own design (daytime logger is additive; the hygiene pass is the subtractive counterpart that cannot run before the log exists). The factory overlays inherit the same post-hoc shape. | `[AutoDream last run: YYYY-MM-DD]` marker update at top of `MEMORY.md` (upstream-schema); per-fire entry appended to `docs/hygiene-history/autodream-fire-history.md` (factory-schema, per row #44: date / agent / overlays fired / findings count / promotion candidates / next-fire-expected-date); ROUND-HISTORY row on the round the fire landed if overlay findings materially shifted something. | `docs/research/autodream-extension-and-cadence-2026-04-23.md` (this row's authoritative policy doc) + `memory/reference_autodream_feature.md` (upstream feature description) + `memory/reference_automemory_anthropic_feature.md` (upstream substrate) + `memory/feedback_in_repo_preferred_over_per_user_memory_where_possible_2026_04_23.md` (Overlay A migration discipline) + `memory/feedback_signal_in_signal_out_clean_or_better_dsp_discipline.md` (consolidation signal preservation) + `docs/HARNESS-SURFACES.md` (Overlay B's drift surface) + `docs/ALIGNMENT.md` (Overlay D's sync target) |
 | 52 | Tick-history bounded-growth audit (`docs/hygiene-history/loop-tick-history.md` line-count vs threshold) | Detect-only (landed 2026-04-22); cadenced detection once per round-close (same cadence as row #44 cadence-history sweep, since this is the canonical row #44 worked example auditing itself); opportunistic on-touch whenever the tick-history file is read or edited. Archive action itself remains manual for now; deferring automation to the larger BACKLOG row that also covers threshold-revision and append-without-reading refactor. | Dejan (devops-engineer) on cadenced detection; the tick itself (self-administered at tick-close) on the opportunistic on-touch — each tick's end-of-tick sequence can invoke this audit after the append + commit to get a `within bounds: 96/500 lines` visibility signal. | factory | `tools/hygiene/audit-tick-history-bounded-growth.sh` checks the file's line count against a threshold (default 500, overrideable via `--threshold N`) and exits 0 within bounds / 2 over threshold. The threshold is set lower than the stated 5000-line paper bound because the file is read on every tick-close append — a per-tick context cost that scales linearly with file size — and 5000 lines represents too large a context hit on a 1-minute cadence. The audit's header block carries a mini-ADR decision record for the 500-line choice (context / decision / alternatives / supersedes / expires-when). **Why this row exists:** Aaron 2026-04-22 tick-fire interrupt: *"does loop tick history grow unbounded? that's an issue if so you just read it"*. Honest state was stated-bound-no-enforcement: file header named 5000 lines, nothing checked it. This row closes the enforcement gap for the threshold-check half of the full BACKLOG row (archive-action + append-without-reading refactor remain deferred). **Self-referential closure:** the tick-history file IS the canonical row-#44 cadence-history-tracking worked example (named explicitly in row #44's "Durable output" citation). Until this row landed, the most-cadenced surface in the factory — the tick itself — had its fire-log surface unaudited for its own growth. Meta-audit triangle remains intact (existence #23 / activation #43 / fire-history #44), and row #49 adds a fourth: fire-history files themselves need bounded-growth audits because they grow at the cadence of the surface they track. **Classification (row #47):** **prevention-bearing** — the audit surfaces approaching-threshold warnings at 80% so the archive action can be planned, rather than reactive-only at over-threshold. Ships to project-under-construction indirectly: adopters inherit the pattern (fire-log files under their own `docs/hygiene-history/` need the same bounded-growth treatment), not this exact script. | Audit output on each fire; cadenced runs appended to `docs/hygiene-history/tick-history-bounded-growth-history.md` (per-fire schema per row #44); BACKLOG row when archival is due (archive-action itself queued as part of the larger tick-history enforcement BACKLOG row); ROUND-HISTORY row when threshold changes or archive action executes. | `tools/hygiene/audit-tick-history-bounded-growth.sh` (detection + mini-ADR header block) + `docs/hygiene-history/loop-tick-history.md` (target surface, canonical row #44 worked example) + BACKLOG row *"Loop-tick-history bounded-growth enforcement"* (larger follow-up: threshold revision + append-without-reading refactor + archive action) |
 
 ## Ships to project-under-construction

diff --git a/docs/hygiene-history/autodream-fire-history.md b/docs/hygiene-history/autodream-fire-history.md
@@ -0,0 +1,21 @@
+# AutoDream fire history
+
+Per-fire ledger for FACTORY-HYGIENE row #53 (AutoDream cadenced
+consolidation — upstream Anthropic Q1 2026 + factory overlays
+A/B/C/D). Schema per row #44 (date / agent / output / link /
+next-expected).
+
+Authoritative policy:
+`docs/research/autodream-extension-and-cadence-2026-04-23.md`.
+
+Upstream feature references:
+
+- `~/.claude/projects/<slug>/memory/reference_autodream_feature.md`
+- `~/.claude/projects/<slug>/memory/reference_automemory_anthropic_feature.md`
+
+| Date | Agent | Gate | Overlays fired | Findings | Next expected |
+|---|---|---|---|---|---|
+| 2026-04-20 | Claude (manual approximation pre-row-#53) | No marker — first known fire | Pre-overlay era | MEMORY.md bootstrap — no overlay framework yet | Superseded by the 2026-04-23 policy doc |
+| 2026-04-23 | Claude (first row-#53 fire) | 3 days + ≥5 sessions since 2026-04-20 marker — both gates pass | Light pass: Orientation + Gather Signal (findings-only); Overlays A/B/C/D scheduled for follow-up tick | Per-user MEMORY.md is 345 lines (well over the 200-line cap); 14+ new 2026-04-23 memories not yet indexed cleanly; generic rules in per-user worth candidate migration to in-repo. Full consolidation deferred to a dedicated tick — this fire records the cadence observance and the findings. | Next fire gate-open at ≥24h AND ≥5 sessions from 2026-04-23 (likely 2026-04-26 or later). Overlay A migration candidates land on that fire or earlier opportunistic-on-touch. |
+| 2026-04-23 (later, same-day) | Claude (auto-loop-50 opportunistic-on-touch Overlay A) | Same session — not cadence-gated; opportunistic-on-touch per row #53 | Overlay A — first execution | Migrated `feedback_signal_in_signal_out_clean_or_better_dsp_discipline.md` from per-user to in-repo `memory/` (PR #157). Dangling citations from `docs/FACTORY-HYGIENE.md` + `docs/research/autodream-extension-and-cadence-2026-04-23.md` resolved. Per-user source retained with "Migrated to" marker (provenance preserved). Candidate next migrations identified: `feedback_outcomes_over_vanity_metrics_goodhart_resistance.md`, `feedback_deletions_over_insertions_complexity_reduction_cyclomatic_proxy.md`, `feedback_external_signal_confirms_internal_insight_second_occurrence_discipline_2026_04_22.md`, `project_semiring_parameterized_zeta_regime_change_one_algebra_to_map_others_2026_04_22.md` — all generic factory-shape, all cited from the just-migrated file. | Next full-cadence fire at ≥24h AND ≥5 sessions from 2026-04-23. Opportunistic-on-touch continues for the 4 identified candidates as future ticks have budget. |
+| 2026-04-23 (later ×2, same-day) | Claude (auto-loop-52 opportunistic-on-touch Overlay A) | Same session — opportunistic-on-touch | Overlay A — second execution (working the identified queue from the prior fire) | Migrated `feedback_outcomes_over_vanity_metrics_goodhart_resistance.md` from per-user to in-repo `memory/` (PR #158). Sibling to signal-in-signal-out (same 2026-04-22 tick pair — Goodhart resistance + DSP preservation as paired disciplines). Resolves one composes-with dangling citation from the just-migrated signal-in-signal-out memory. Queue now 3 candidates (deletions-over-insertions / external-signal-confirms-internal-insight / semiring-parameterized-zeta). | Opportunistic-on-touch continues; cadenced fire still gate-open ≥2026-04-26. |