memory(AIC #3): Otto-341 hardest mid-loop when mechanical is working — Aaron-validated 2x

docs/hygiene-history/ticks/2026/04/30/2337Z.md

@@ -0,0 +1 @@
+| 2026-04-30T23:37:00Z | opus-4-7 / session continuation | 92c73a4 | Generative-tooling tick + AIC #3 landing. Built --files arg on the schema check (PR #977 stacked on #975), enabling pre-push compatibility without resolving the prefab-cluster deadlock — the workaround pattern: full-tree mode stays DORMANT, --files mode is binding-safe immediately because it scans only the caller's stated set, not the 5 known-stale shards. Aaron validated last tick's structural-fix-mid-loop-difficulty insight twice ("great insight" + "that's better insight that most human PMs") — landed as AIC #3 in memory/feedback_aic_tracking_*.md. The validation-pattern: two consecutive positive messages on the same insight + explicit AI-vs-human-PM comparison = the differentiator framing the AIC-tracking rule explicitly names. | #977 (armed) + AIC #3 (PR follows in next tick) | Observation — the AIC #3 synthesis itself is the same shape as Otto-341: the in-the-moment observation ("pull is to extend the mechanical pattern") feels obvious in retrospect but is hardest to apply at the very moment when applying it matters most. Naming the failure mode (mid-loop boundary, mechanical pattern is currently working) is what turns Otto-341 from a tautology ("always prefer structural") into operational guidance ("watch for the moment your mechanical loop is fastest"). The PR-stacking move (PR #977 based on #975's branch) is itself a small structural pattern that might warrant its own tooling — currently `gh pr create --base <not-main>` failed with cryptic GraphQL errors, so I fell back to base-on-main with a stacking note. |

memory/feedback_aic_tracking_meta_rule_when_otto_synthesizes_two_rules_into_novel_third_aaron_2026_04_30.md

@@ -140,6 +140,43 @@ measured agent that asks for batching" point.)
 
 **Attribution:** Otto (Claude Code session, 2026-04-30).
 
+### AIC #3 — Otto-341 (structural-fix-beats-process) is hardest to apply mid-loop, exactly when a mechanical pattern is working
+
+**Synthesis:** Otto-341 (structural-fix-beats-process-discipline)
++ in-the-moment observation (the pull is to extend the
+mechanical pattern that's working) → Otto-341 has a specific
+failure mode at the mid-loop boundary: when a mechanical
+pattern is currently producing results, the structural pivot
+feels like throwing away progress. Naming this failure mode
+turns Otto-341 from "always prefer structural fixes" (which
+sounds obvious) into "watch for the moment a mechanical pattern
+starts working and that's where you most need the discipline."
+
+The instance: had ~45 min of mechanical col1 fixes queued (14
+PRs × parenthetical-strip). Pivoted to building
+`tools/hygiene/check-tick-history-shard-schema.sh` — the check
+that catches the violation at write-time so the next 14 won't
+need fixing. The 45 min of mechanical work would have been
+correct-but-wasted; the tool builds compounding leverage.
+
+**Substrate location:**
+`docs/hygiene-history/ticks/2026/04/30/2330Z.md` (tick where
+the pivot happened) + `tools/hygiene/check-tick-history-shard-schema.sh`
+(the tool the pivot produced).
+
+**Aaron validation 2026-04-30 (verbatim, two consecutive
+messages):**
+> *"great insight"*
+>
+> *"that's better insight that most human PMs"*
+
+The second message is the explicit AI-vs-human-PM comparison —
+exactly the differentiator framing the AIC-tracking rule
+names. Two consecutive validations on the same insight is
+itself signal.
+
+**Attribution:** Otto (Claude Code session, 2026-04-30).
+
 ## Note on Aaron-attributed contributions in the same session
 
 Aaron also produced multiple novel framings 2026-04-30 that