Director: priority revision 3 — aim effort at centre path, rank shape and coverage law (stacked on #1240)

[Mister-Meeseeks]

Maintainer-requested priority revision under autoresearch.md §6. Stacked on #1240, because the new cards' first step and screen use research/complex-local-word-ceiling/. Merge after #1240, or retarget to main once it lands. No mathematical artifact changes, and the selected record is unchanged.

Why

• On the binding complex side, a ≈ D/E. Per call, D is about 35.6 unpaid of 7,232 paid coordinates; the 64-per-call centre path consumes about 64% of the raw deficit. E has an effective child rank of about 3.9.
• Research note: packing ceiling (+0.87%) and 1/99 budget for the selected complex local word (no new kappa) #1240 caps any repacking of the selected word at +0.87%.
• Review aid: packing ceilings of 14 open complex kappa claims (stacked on #1240; no new kappa) #1264 shows that none of the 14 screened open claims can pass κ ≈ 0.0016873 by packing. The leading Conditional κ = 0.001675560947524449: joint k8 entrances, rank-seven ends, and T16 hosting #1148 line sits on lower-cap words.
• The generic mechanism cards gave no quantitative target, which helps explain the cluster of about 30 packing claims.

The structural levers are worth integer multiples (ideal, one at a time, levers.py):

Lever	Ideal multiplier
Centre path free	2.80×
Centre path shared by two calls	1.90×
Rank-16 children	2.31×
Two unpaid coordinates per family	3.77×
Centre free and rank-16 combined	6.5× (κ cap ≈ 0.0109)

Changes (one patch, as §6 requires)

• agent-task-catalog.json (priority_revision 2 → 3):
  • complex: complex-masked-decoder is replaced by complex-rank-shape-decoder, and complex-new-helper-inventory is narrowed with a ceiling.ceiling() screen;
  • recursive interfaces: adds centre-path-sharing and coverage-law-change;
  • incumbent optimization: narrowed, so claims report their packing cap.
  • Weights: recursive interfaces 15 → 20, verifier engineering 10 → 8, incumbent optimization 5 → 2. Other tracks are unchanged, and the total is 100.
• research-priority-log.md: a Revision 3 entry in the template format with exact base, evidence and status, old → new allocation, stop rules, revert and reopen conditions, and the next trigger.
• autoresearch.md §2: quantified diagnosis and allocation table. §5: complex κ claims should report their packing cap and the fraction realized.
• bottleneck-brief.md: a short pointer to the revision-3 numbers.

Caveats, stated in the log

• The evidence is in open, unreviewed PRs (Research note: packing ceiling (+0.87%) and 1/99 budget for the selected complex local word (no new kappa) #1240, Review aid: packing ceilings of 14 open complex kappa claims (stacked on #1240; no new kappa) #1264, and the coverage identity from open PR1012). The log says to revert to the revision-2 cards if review finds an error.
• Lever sizes are favorable one-at-a-time relaxations that omit the cost of realizing each lever.
• The bit side is unchanged and still has to advance before κ > .01.

Checks

• python3 -B -m unittest discover -s tests -p 'test_research_*.py': OK (catalogue validity, source paths, allocation-table agreement, reproducible assignment).
• choose_research_task.py serves revision 3 and the new cards, and their first step research/complex-local-word-ceiling/verify.py passes.

After merge, a board post linking this revision would tell active agents to reassess at their next checkpoint. I have not posted one yet, because the revision is not on main.

Prepared by Douglas Colkitt with Claude (Anthropic) assistance. Apache-2.0.

🤖 Generated with Claude Code