hygiene: tick-history row — roms/ hierarchy + #282/#398/#401 drain close

docs/hygiene-history/loop-tick-history.md

@@ -189,3 +189,4 @@ fire.
 | 2026-04-22T16:45:00Z (round-44 tick, auto-loop-44 — reproducible-stability thesis landing + bilateral-verbatim-anchor correction arc + t3.gg sponsor eval + 42-task-cleanup) | opus-4-7 / session round-44 (post-compaction, auto-loop #44) | aece202e | Tick span covered: (a) **thesis landing** — maintainer directive *"is obvious to all personas who come across our project the whole point is reproducable stability"* + *"change break to do no perminant harm and they are equel"*; landed as minimal-signal edits to AGENTS.md (new `## The purpose: reproducible stability` section with verbatim blockquote; value #3 verb substitution `Ship, break, learn` → `Ship, do no permanent harm, learn`) + README.md (new `## The thesis: reproducible stability` section with blockquote + pointer) + memory file `project_reproducible_stability_as_obvious_purpose_2026_04_22.md`. (b) **bilateral-verbatim-anchor correction arc** — maintainer flagged hallucinations mid-tick (*"you just make up resasons for me i never told you"*); I stripped AGENTS.md + README.md editorial content to verbatim-only floor; maintainer then retracted (*"i'm wrong i went back and looked and it's fine what you said"* + *"i hallicunatied not you"* + *"that was operator error lol"*); stripped state stays committed as honest floor since reconstructing editorial from summary would itself be re-synthesis — maintainer directs future expansion on own terms. Meta-lesson: both sides can mis-remember a correction; the verbatim trail (committed memory quotes) settles disputes bilaterally, not just agent→maintainer. (c) **t3.gg/sponsors evaluation** — maintainer asked if Theo's sponsor list (Blacksmith/Depot/PostHog/Sentry/Axiom/Upstash/PlanetScale/Modal/Kernel/etc.) was useful; honest answer: **no direct fit** — the roster is SaaS-heavy and antithetical to Aaron's absorb-and-contribute + Escro-maintain-every-dep + no-cloud directions; only marginal candidates were Blacksmith (GHA runner substitution, not a dep) and Axiom (log aggregation if factory ever centralises logs), neither urgent. (d) **task-list cleanup** — maintainer asked *"any to cleanup?"*; deleted 42 completed task entries, kept 5 active (#182 speculative drain, #240 email-provider mapping, #244 ServiceTitan demo, #198 batch 6, #256 this tick). (e) **SignalQuality + /btw** already landed pre-tick (commit `acb9858`): 6-dimension quality measure (Compression/Entropy/Consistency/Grounding/Falsifiability/Drift) + severity bands + ZSet-integrated claim store + composite scoring + 22 passing tests + `.claude/commands/btw.md` non-interrupting aside. Cron `aece202e` armed; minutely fire intact; tick closes clean. Carry-forward: specific "phenomenon" artifact still unresolved — maintainer described it as something that *"showed up a while back that it looked like you tried to absorbe and failed"*; grep searches under `docs/research/` + `memory/observed-phenomena/` did not produce a confident match; open question for next contact. |  |  |
 | 2026-04-22T17:00:00Z (round-44 tick, auto-loop-45 — unabsorbed-phenomenon gap closure: companion markdown for the 2026-04-19 transcript-duplication PNG) | opus-4-7 / session round-44 (post-compaction, auto-loop #45) | aece202e | Speculative-work tick per never-be-idle priority ladder — known-gap fix rather than waiting. Gap: the singular file in `memory/observed-phenomena/` (`2026-04-19-transcript-duplication-splitbrain-hypothesis.png`) had no companion markdown; Aaron's auto-loop-44 clarification that *"phenomenon was something that showed up a while back that it looked like you tried to absorbe and failed"* mapped cleanly to this artifact — a PNG filed without a written absorption. Landed: `memory/observed-phenomena/2026-04-19-transcript-duplication-splitbrain-hypothesis.md` (companion note, ~130 lines) that does three things and explicitly not a fourth: (a) names what EXISTS (the PNG, the filename-encoded hypothesis, the existing memory-file citation from Glass Halo), (b) names what does NOT exist (no written analysis, no commit msg, no ADR, no reproduction steps, no falsification plan, no explicit link to the anomaly-detection paired feature despite Aaron's verbatim framing that the phenomenon triggered that feature), (c) captures Aaron's verbatim three-claim framing from auto-loop-44, and (d) explicitly DOES NOT reconstruct what a prior Claude's absorption attempt contained — that would be exactly the re-synthesis Aaron flagged as hallucination. Open question for next contact: what axis did the prior absorption fail on (causal model / reproduction / falsifiable test / corpus landing)? The shape of the failure tells us what success looks like. Also this tick: cron-cleanup — deleted the redundant one-shot `42945668` ScheduleWakeup entry left over from the prior tick (the minutely `aece202e` heartbeat was already the canonical fire; the 25-min ScheduleWakeup was wrong-posture since the tick ALREADY fires every minute per CLAUDE.md "Tick must never stop"). Build: 0 Warning(s), 0 Error(s). |  |  |
 | 2026-04-24T02:00:00Z (autonomous-loop tick, auto-loop-48 — Craft production-tier ladder bootstrapped + first module landed) | opus-4-7 / session continuation | 20c92390 | Tick executed foreground-axis directly on Aaron's Otto-47 directive by landing the Craft production-tier ladder v0 + first module. Tick actions: (a) **Step 0 state check**: PR #207 (Otto-47 BACKLOG rows) MERGEABLE but BLOCKED on build-and-test IN_PROGRESS; 5 Phase 1 PRs (#199/#200/#202/#203/#204) updated from BEHIND via `gh pr update-branch`; #206 BLOCKED same as #207. Background axis clean; foreground picks new substrate. (b) **Production-tier ladder bootstrapped**: created `docs/craft/subjects/production-dotnet/README.md` naming the ladder distinctly from onboarding (different audience, different prerequisites, different lessons). Structural concept added: `docs/craft/subjects/production-{lang}/{topic}/` directory convention. Four neighbour module stubs named (zero-alloc-hot-loops, simd-vectorisation, struct-vs-ref-semantics, jit-inlining-rules) for future landing. (c) **First module landed**: `docs/craft/subjects/production-dotnet/checked-vs-unchecked/module.md` (~260 lines). Six-class site decision matrix (bounded-by-construction / bounded-by-workload / bounded-by-pre-check / unbounded-stream-sum / user-controlled-product / SIMD-candidate). Decision tree read top-to-bottom. Measurement gate: ≥5% BenchmarkDotNet delta required per demotion; F#-specific `Checked.` vs. `(+)` benchmark harness shown. Three bound-proving techniques (type-system / algebraic / FsCheck property). Canonical `src/Core/ZSet.fs:227-230` site cited as **keep Checked** exemplar. Concrete demotion candidates named: ZSet.fs:289-295 (SIMD-candidate), NovelMath.fs:87 (bounded-by-workload counter), CountMin.fs:77 (bounded-by-workload), Aggregate.fs:30 (unbounded — keep Checked). Self-check section with 4 observable outcomes. Composes-with pointers + explicit NOT-list (not mandate-to-demote-every-site / not project-flag-flip / not replacement for property tests / not onboarding / not micro-opt-for-its-own-sake). (d) **Split-attention model held**: background = 5 PR update-branches applied via `gh pr update-branch` loop; foreground = production-tier module. No interrupt-break-on-blocker (audit BACKLOG row doesn't block module because module teaches decision framework, not specific audit results). (e) **CronList verified live**: `20c92390` minutely fire. | PR `<pending>` `craft/production-dotnet-checked-vs-unchecked-v0` | Observation 1 — tier-split was genuinely structural. A "harder onboarding module" would gatekeep beginners at the `subjects/zeta/` surface; a separate `subjects/production-dotnet/` welcomes a different audience at their correct entry point. Same applied-default-theoretical-opt-in discipline inside the module, but prerequisites are level-appropriate (BenchmarkDotNet literacy, span fluency) instead of onboarding metaphors. Observation 2 — landing the module v0 *before* the per-site audit executes is the right sequencing. The module teaches the *decision framework*; the audit produces *specific decisions*. Decision framework doesn't depend on audit outcome — audit outcome will be informed by the framework. Sibling-not-sequential. Observation 3 — the six-class matrix is already load-bearing for the audit: Naledi (perf) will use it as the classification spine; each of ~30 sites slots into one class; the "keep Checked" column catches half. Landing the taxonomy now prevents ad-hoc classification later. Observation 4 — module self-check (4 observable outcomes) gives future readers a concrete way to flag if the module failed pedagogically. Bidirectional alignment built in from v0. |
+| 2026-04-25T01:45:00Z (autonomous-loop tick — #282 lint fix finish + #401 upstreams sentinel landed + #402 roms/ canonical hierarchy v0 with BIOS-availability filter) | opus-4-7 / session continuation (post-compaction, Otto-NNN cluster) | f38fa487 | Tick executed a three-thread day: (a) **#282 drained to green-lint floor** — landed 4 remaining markdownlint failures (MD018 ×2 for `#280`/`#266` line-leading heading-parse, MD056 table-pipe-inside-code-span broken gate-name fix, MD032 missing-blank-line-before-list) after earlier 9-thread Copilot reply+resolve pass. PR now at `f30be23`; lint pending CI re-run; auto-merge armed. (b) **#401 upstreams-sentinel-parity PR opened + armed** per earlier Aaron *"we should if not"* — `references/upstreams/.gitignore` + `references/upstreams/README.md` sentinels land with root `.gitignore` switched from `references/upstreams/` (blanket) to `references/upstreams/*` + `!references/upstreams/.gitignore` + `!references/upstreams/README.md` exemptions. Same shape as `drop/` + `roms/`. (c) **#402 roms/ canonical hierarchy landed as a living design through five maintainer iterations**: (1) initial 63-directory tree grouped by manufacturer with EmulationStation slugs; (2) Aaron Otto-*"under atari you would have like 2600 and that level of category too"* + *"mame is separate we don't need per emulator folder"* — atari/ children stripped of `atari` prefix (2600, 5200→lynx, 800, st, etc.), arcade/ removed, mame/ + fbneo/ promoted top-level; (3) Aaron Otto-*"we don't need an extra roms folder under fbneo / same for mame"* — dropped the `mame/roms/` + `fbneo/roms/` sub-path misreading; (4) Aaron Otto-*"if there are any you need bios files you can't create yourself lets remove those"* + *"just keep the ones you don't need anything but your code"* — stripped Sony/Saturn/Dreamcast/Neo Geo/3DO/Xbox/GC-Wii-DS/PCE-CD/Intellivision/ColecoVision/Apple II/Amstrad/BBC/C64/VIC-20/Atari-5200-7800-Lynx; (5) Aaron Otto-*"open source bios is fine too"* + *"keeping only those that work standalone or have viable open BIOS replacements or ones we can write ourself from scratch without cheating"* — restored Atari 800 (Altirra OS, BSD), Atari ST (EmuTOS, GPL), Commodore Amiga (AROS, APL), MSX (C-BIOS, BSD), ZX Spectrum (Open Source Speccy ROM). Final state: 37 directories, 38 READMEs (branch + leaf + top-level). Per-folder sentinel distinguishes branch ("not empty — enumerates children") from leaf ("drop ROMs per top-level protocol"). MAME/FBN stay removed because per-board BIOS requirement has no viable open-source alternative. Commits: `548320d` (initial hierarchy) + `bb5b900` (trim + BIOS-availability filter). (d) **Otto-279 policy clarification captured + BACKLOG-extended**: *"research counts as history, give first-name attribution, agents get attributions too. we can add it to the list. backlog that that will be a lot of churn after the drain"* — Otto-52 BACKLOG row (name-attribution policy clarification) extended with Otto-279 reinforcement; post-drain sweep to RESTORE stripped names on research docs (PR #351 et al). Memory file `feedback_research_counts_as_history_first_name_attribution_for_humans_and_agents_otto_279_2026_04_24.md` + MEMORY.md index pointer. Reverted my own mid-tick name-stripping edits on #282 when policy was re-clarified. (e) **#398 drained** — 3 threads (Codex P2 + Copilot ×2) about `dotnet` example commands vs `mise exec --` rule-rationale mismatch; fixed in `f7ca762` (both Context + Verified-2026-04-24 examples now route through `mise exec --` with inline discipline-echoing note); replied + resolved all three. (f) **CronList verified**: `f38fa487` minutely autonomous-loop fire armed; cron stays armed. | `548320d` + `bb5b900` (#402 chore/roms-hierarchy-sentinels branch) + `f30be23` (#282 lint finish) + `f7ca762` (#398 mise exec fix) + `0f4d9ee` (#401 upstreams sentinel) | **Observation 1 — living-design-through-iteration cadence held cleanly.** Aaron iterated the ROM hierarchy rule through **five** clarification bursts in the same loop; tree mutated through four intermediate states before stabilising at 37 dirs. Each mutation landed as a discrete regeneration (Python script re-wrote all READMEs from metadata on each iteration) rather than cumulative patches — cheaper to rebuild from source than to chase patches across 30+ markdown files per iteration. **Observation 2 — BIOS-availability filter is a cleaner cut than manufacturer-completeness.** Initial impulse was "list every common emulator"; Aaron's rule "just your code + safe ROM" cut 30+ platforms in one move and the remaining 28 are *coherently usable* under the factory's safe-ROM protocol. Completeness was the wrong axis; self-containment was. **Observation 3 — Otto-279 name-attribution surface-class refinement composes with Otto-237 mention-vs-adoption.** Both are "rule-applies-differently-per-surface-class" clarifications. Otto-220 was the literal rule; Otto-237 carved out mention-in-research (don't strip public-info references); Otto-279 carves out names-in-history-surfaces (research docs ARE history, names stay). Same shape applied to two different content axes. Backlog + not-in-drain scheduling is correct (churn after drain). **Observation 4 — reply+resolve discipline held across 12 threads today** (9 on #282 + 3 on #398); zero breadcrumb-unresolved state left behind. Otto-236 discipline intact. **Observation 5 — speculative work in-flight** while waiting on #282/#398/#401/#402 CI completion: the ROM hierarchy landed as the speculative move with highest-value per the never-be-idle ladder (generative factory improvement: makes substrate ready for a future emulator-absorption milestone). |