Reduce failed-attempts.md hook-debt backlog and class-count clustering drift

[DrumRobot]

Problem

cleanup's failed-attempts.md corpus has two escalating backlog signals that a recent fa-classify.py --relaxed run confirmed are not resolvable via the normal archive/dedup pass:

• Class count drift: total recorded failure-pattern classes have grown to 161-165, well past the escalation threshold of 50. COLD candidates (safe to archive) are 0, so a normal archive pass cannot shrink this — the classes are still active/relevant, just too numerous and likely contain near-duplicates or overly narrow splits that should be merged.
• Hook-debt backlog: 9 classes are marked status=hook-pending (a hook was proposed as the fix but never implemented/verified), above the threshold of 8. This has crossed the threshold twice recently (dropped to 7 after one promotion, then rose back to 9 as new patterns were registered faster than hooks were implemented).

Both signals point to the same underlying issue: pattern registration is outpacing implementation/consolidation work on this corpus.

Scope

1. Class-count clustering review: review the ~161-165 recorded classes for merge/split candidates — classes that describe the same underlying failure mode with different wording, or classes that were split too finely and should be recombined. Reduce the total count meaningfully (target: closer to the 50-class threshold, though full compliance may take multiple passes).
2. Hook-debt sprint: for the 9 hook-pending classes, either (a) implement and verify the proposed hook (promoting the class to hook-active), or (b) determine the hook is no longer warranted and reclassify/close the entry. Prioritize classes with the clearest, most mechanically verifiable fixes first.
3. Registration discipline: while working through the backlog, avoid registering further new failed-attempts classes in this repo unless truly novel — the backlog itself indicates recent registration has outpaced follow-through.

Out of scope

• Wholesale rewrite of the failed-attempts.md format or classification scheme.
• Classes with status=rule-covered or status=watch — these are intentionally left as-is pending real-world recurrence, not part of this backlog.

Implementation Plan

• Run fa-classify.py --relaxed (or the current equivalent) to get a fresh class inventory with status breakdown
• Identify merge/split candidates among the ~161-165 classes and consolidate duplicates
• For each of the 9 hook-pending classes: implement + verify the hook, or reclassify if no longer warranted
• Re-run the classifier to confirm both counts (hook-pending and total class count) have moved meaningfully toward their thresholds

Verification

Feature	Procedure	Expected Result
Hook-debt reduction	Re-run the classifier after implementing/reclassifying pending-hook classes	hook-pending count drops below the threshold of 8
Class-count reduction	Re-run the classifier after the merge/split pass	Total class count measurably lower than the 161-165 starting point
No regression	Spot-check a sample of merged/reclassified entries	No loss of information — merged entries retain all distinct failure scenarios they represented