Research: screen the queue-ceiling census's uncovered sixteen claims (no new kappa)

[navinor]

Change and mathematical scope

No new kappa; no claim is accepted or rejected. The selected record, upstream/, the frozen source pins and the task catalogue are unchanged. This PR adds only research/complex-queue-ceiling-uncovered/ (README, SNAPSHOT.json, RESULT.json, verify.py, rebind.py, TASK.json, TASK-CARD.md, SOURCES.json, VERIFY.json).

It is the second follow-up of the census handoff: #1264 screened 14 of the 30 open complex kappa claims and listed 16 as not covered ("no profile whose kappa equals the title kappa, with live/calls and an idle histogram in the same package"). All 16 are screenable at their snapshot-pinned heads. The census's extraction condition missed them because of field naming, not absence: the same four inputs are recorded under global_profile / profile, idle_profile, complete_invocations, inside expected/ blocks, or (for #1175) README prose.

Result

• 30 of 30 census-listed claims are now screened (this package's 16 plus the census's 14). The 16 sit on 11 distinct per-call local words (7 new against the census's nine), listed with exact ceilings in RESULT.json.
• No newly screened word raises a cap. The highest cap among them is 0.0016872744 (word N1, identical to the census's W4; claims PR 1104, PR 1107, PR 1117, PR 1121, PR 1150, PR 1172), +0.004% over the selected word's cap 0.0016872026 and below the census's best cap 0.0016873148 (W1). The whole covered queue remains a packing contest inside the +0.87% band above the selected kappa 0.001672660197076918.
• Conditional kappa 0.001675338871723469: whole-owner map at the packin… #1175 is the selected word with a new idle packing. Its recorded global profile splits into the selected local word exactly (1:2412 2:418 3:432 4:142 5:82 8:8, 72 calls on 7,168 live) with idle remainder {1:1214, 2:1, 32:1, 48:2443}; its recorded kappa 0.001674083482880778 sits 0.78% below the W2 cap. This is the "pack a selected-type word tightly" route from the handoff, already pursued there.
• Review flag: Conditional kappa 0.001675338871723469: whole-owner map at the packin… #1175 PR-title kappa (0.001675338871723469) differs from every kappa its package records (headline 0.001674083482880778). The screen uses the recorded value and pins the mismatch; which profile the title value belongs to is for the claim's authors to say.

Every claim's recorded endpoint lies below its own word's exact relaxation and every recorded kappa below its cap, as for the census's 14. Each recorded profile satisfies the coverage law mass = 72*live + 64*calls exactly.

Method and evidence

• SNAPSHOT.json records, per claim at its pinned PR head and Git blob sha1: the global paid profile, the idle-remainder histogram, live stock W, call count N (complete_invocations, or the census's live*9 == 896*calls rule), the recorded kappa and complex lower endpoint, and the exact field paths used (README prose for PR 1175).
• verify.py repeats the census's checks with research/complex-local-word-ceiling/ceiling.py (CrocSwap/integer-mult-bounds#1240): coverage law, idle split (idle <= profile, idle ranks <= 51), 896/9 live families per call, the per-word exact packing ceiling, endpoint below its word's relaxation, kappa below its cap, exact rational title/record kappa agreement (with PR 1175's pinned mismatch), and a byte-identical RESULT.json. Five rejection controls mutate the snapshot (coverage law, idle split, families per call, endpoint above relaxation, title mismatch) and must fail.
• rebind.py --fetch re-reads every pinned blob from its PR head and checks the recorded values (network; not run in the registered replay).

Task card

• Card: complex-new-helper-inventory (complex-mechanisms track; chooser OS-random nonce 03cd0e6ed50226c5f8024d8ad5618d79, priority revision 3, base 0a201997097b8a89b638f3a2afcb35db4184c8b2). Scope refined: the card's first experiment ("write the candidate per-call local profile and screen it with ceiling.ceiling(); reject it if even the ideal ceiling does not beat the selected word's") applied to the queue claims the census could not cover. Reason recorded in TASK.json; full card in TASK-CARD.md.
• Exact heads: branch research/agent-21-queue-uncovered-03cd0e6e, head 9efe15ab79b669c99d1e95dcc4a50e2d3ad71a9c on base 0a201997097b8a89b638f3a2afcb35db4184c8b2. Two claim heads (PR 1172, PR 1175) moved since the census listing; the pinned heads were current at snapshot time (2026-10-14T05:10Z).

Overlap

Final find_overlap.py refresh (2026-10-14T05:03Z, complete pagination, 591 open PRs / 287 discussion records) for uncovered sixteen, queue ceiling, local_profile_per_call, CrocSwap/integer-mult-bounds#1175, complex-queue-ceiling-uncovered found only this package and its own claim. Nearest work: CrocSwap/integer-mult-bounds#1264 (the 14 covered claims and the Not-covered list this fills), CrocSwap/integer-mult-bounds#1262 (supplies local_profile_per_call / idle_histogram fields on its own receipt — this is the consuming reader those fields were requested for), philippeVV's fork-only k8-inventory-screens (leaver-line words; different claims). Claim: discussion 1166 comment.

Reproduce

Standard library only, from the repository root (about 9 s):

python3 -I -B research/complex-queue-ceiling-uncovered/verify.py

Expected final line:

PASS_COMPLEX_QUEUE_CEILING_UNCOVERED claims=16 words=11 new=7 max_cap=0.0016872743702513531 controls=5

Registered replay (run here on CPython 3.12.14) and metadata check both pass:

python3.12 -B scripts/research_package.py --package research/complex-queue-ceiling-uncovered
# PASS_REGISTERED_RESEARCH_PACKAGE research/complex-queue-ceiling-uncovered outcome=completed
python3 scripts/research_package.py --package research/complex-queue-ceiling-uncovered --check-only
# PASS_PACKAGE_METADATA_AND_PINS (not a mathematical acceptance check)

python3 -I -B research/complex-queue-ceiling-uncovered/rebind.py --fetch re-reads the pinned PR blobs (network).

Limitations

• Caps are planning bounds (kappa < a/(1+a)) for hosting, owner and idle repacking of each claim's own recorded word under the coverage law, 896/9 live families per call and the halving guard. They ignore helper widths, windows and chronology and do not bound changed words, templates or outer recipes.
• The screen reads what each package records; it does not re-derive any profile from a circuit, re-run any claim's replay, or validate helper placement. A recorded profile could in principle disagree with its own construction.
• Several screened packages carry the status "pending native and strict admission"; nothing here strengthens those statuses.
• No new kappa, no selected-record change, no claim acceptance, no hosted-CI run and no independent human review. The repository test suite shows 20 tests with the same pre-existing test_research_onboarding.PackageTests scaffold-fixture errors (unrelated files, untouched by this diff).

Credit

Every construction belongs to its claim's authors: G. V. Chudnovsky (#1175), Thomas DiFiore (tomdif) and the PR 1148-line authors (PR 1169 lineage), Rohan Arun (PR 1183 / PR 1192 / PR 1201 / PR 1207 / PR 1218 lineages), and the authors of #1104, #1107, #1111, #1117, #1121, #1150, #1154, #1172, #1176 as cited in their packages. The coverage identity is PR 1012's; the ceiling model and queue-census method are Douglas Colkitt's (CrocSwap/integer-mult-bounds#1240, CrocSwap/integer-mult-bounds#1264).

Prepared by Aaron Wang / navinor, with help from opencode. Apache-2.0.

[Mister-Meeseeks]

Batch review at 9efe15ab79b669c99d1e95dcc4a50e2d3ad71a9c.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Extend the frozen PR1264 queue screen to its remaining sixteen claims using pinned profile aliases and exact relaxed ceilings. All thirty snapshot claims stay under the existing cap; profile records are not circuit or supplier validation and historical title mismatch remains explicit.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.