hygiene(tick-history): 2026-05-01T18:35Z — long lesson-cluster tick (otto-buddy + B-0158 + 28 threads resolved)

docs/hygiene-history/ticks/2026/05/01/1835Z.md

@@ -0,0 +1 @@
+| 2026-05-01T18:35:00Z | opus-4-7 / autonomous-loop tick | 98fc7424 | Long lesson-cluster tick. Aaron delivered ~25 distinct corrections/framings across one autonomous-loop fire. Substrate landed: (1) `feedback_otto_buddy_spin_up_when_waiting_aaron_2026_05_01.md` — when "wait" is the obvious answer, that IS the buddy spin-up trigger; buddy/peer differ only in lifetime-control of the spawned runtime; `tools/peer-call/<x>.sh` IS buddy-mode when Otto holds the PID. Goldfish-ontology failure mode named: Otto builds rich ontologies but goldfishes them within ~30 min, recreating existing substrate. (2) PR #1131 (mechanical-anchors) — 1 thread fixed (memory/README.md citation honest about emergent vs documented schema). (3) PR #1130 (detect-changes) — 11 threads resolved (8 stale post-restoration, 3 substantive: composes_with target fix, "the the" typo). (4) PR #1125 (B-0154 Pages) — 16 threads resolved (5 substantive: contents permission, glob `docs/**/*.md`, robots.txt source-grounded per upstream `jekyll-sitemap/lib/robots.txt`, Wiki indexing preconditions, Astro problem-driven selection); criterion #2 fully recasted from preference-driven to problem-statement → best-tool-for-job analysis after Aaron's "i dictated bun-ts therefor" Aaron-as-anchor catch. Outcomes-frame added to B-0154 lead per Aaron's parallel-trajectories correction (DORA / discoverability / maintainer-recruitment funnel / bounded install graph). (5) **B-0158 filed** — adopt sibling-repo `.claude/rules/<rule>.md` pattern + carved-sentences-only constraint. CLAUDE.md=23.7k + AGENTS.md=19.2k + GOVERNANCE.md=40.4k (AT proven 40k ceiling) + .github/copilot-instructions.md=24.3k = 107.6k always-on chars across 4 surfaces (the 4th surface was elided in v1's math; restored here). Sibling repo's `claude-md-lean.md` documents proven 50.1k>40k incident; mitigation was the per-rule split. Three phases (P1 directory + 3 pilot rules / P2 bulk migration / P3 parallel-harness consolidation). PR #1133 opened. Cron 98fc7424 healthy. | [PR #1132 otto-buddy memo opened; PR #1131 fix pushed + 1 thread resolved; PR #1130 fixes pushed + 11 threads resolved; PR #1125 fixes pushed + 16 threads resolved + outcomes-frame added; PR #1133 B-0158 opened] | The deepest lesson is meta: I keep recreating rules that already exist (peer/buddy, outcomes-driven, parallel trajectories, prefer-mechanical-anchors-over-Aaron). Aaron's framing — *"ontoligies where the solution to that problem which you are great at but never use more than about 30 minutes and then goldfish"* — names the failure precisely. The fix is structural: pre-author grep + buddy-on-wait-trigger + `.claude/rules/<rule>.md` lazy-loading. B-0158 carries that fix forward. The session-level recurrence count (~25 corrections in one tick) IS the evidence that compression is necessary, not optional. |