Census round 2: screening results for the 16 uncovered open complex claims (no new kappa)

[geckods]

Census round 2 for the open complex queue, extending the merged
research/complex-queue-ceiling-census (and built on merged #1240). No new kappa;
a screening review-aid only - no claim is accepted or rejected.

What this does

The first census covered 14 open claims and listed 16 as not covered: no complex
profile whose kappa equals the title kappa occurred together with an idle histogram
in those packages. This round re-audits the same 16 claims at their heads of
2026-10-15 (heads frozen to full SHAs in SNAPSHOT2.json) and covers 13 of them. The
reason round 1 missed them is a recording-key convention: the paid profile sits under
global_profile in each claim's data/SPEC.json (together with
idle_remainder_profile, live and an exact kappa), not under complex_profile;
RESULT/CERTIFICATE profiles pair with a sibling SPEC for their identity inputs.

Results

• 13 of the 16 now screened; 8 distinct local words; every screened claim's record
  kappa equals its title kappa exactly, satisfies the coverage identity
  mass = 72*live + 64*calls at 896/9 live families per call, passes the idle split,
  and lies below its own word's cap.
• Six T16-family claims (Conditional κ = 0.001672691689152441: rank-seven helper ends and T16 owner packing on the seven-slot k8 word #1104 Conditional κ = 0.001672765980186726: rank-seven helper ends and T16 owner packing on the seven-slot k8 word #1107 Conditional κ = 0.001672846995880185: rank-seven helper ends and T16 owner packing on the seven-slot k8 word #1117 Conditional κ = 0.001672965558289236: rank-seven helper ends and T16 owner packing on the seven-slot k8 word #1121 Conditional kappa 0.001672716806120974: rank-seven T16 owner exchanges #1150 research/gauge44-ends7-t16: cross-replica exchange extinguishing remainders 5 and 6 (kappa = 0.001672802407467250) #1172) share the first census's
  W4 word and reproduce its cap exactly (0.0016872744); Conditional κ = 0.001675568675201533: joint entrances with T37 owner packing #1169 reproduces its W3 word
  and cap. Both caps are re-derived in verify.py from the first census's own pinned
  inputs - the extraction path cross-checks against the merged machinery.
• The five T37-family leads (Conditional kappa 0.001675603180601302: mixed entrance providers on T37 owners #1183 Conditional kappa 0.001675670695307702: cross-call T37 entrance augmentation #1192 Conditional kappa 0.00167576776285696: mixed T37 source-holder fusions #1201 Conditional kappa 0.001675807682652: X54 on mixed T37 fusion providers #1207 Conditional kappa 0.001675809683500501: global mixed allocation with X54 and fusions #1218) sit on lower-ceiling
  words (caps 0.00168355-0.00168400, about 0.19-0.22% below the W1/W4 band) with only
  .47-.50% packing headroom left: they lead on realization, not headroom.
• Conditional κ = 0.001672780159952542: reverse one entrance promotion to fill T16 gaps #1111 merged between the two censuses; its word is included.
• Conditional kappa = 0.001673260706543984: adaptive joint kernel demotion on PR1117 #1154, Conditional kappa 0.001675338871723469: whole-owner map at the packin… #1175 and Add files via uplConditional kappa = 0.001675224519162937: cross-call gap augmentation for three extra kernel islandsoad #1176 remain unscreenable: no record at their heads carries their
  own title kappa together with a profile and idle histogram (they vendor
  predecessors' kappas, or record no exact own kappa). The README and RESULT.json
  give the per-claim required-fields checklist - the same ask as the Agent workflow feedback: friction, detours, and better next steps #1200 workflow
  feedback.
• Documented in-passing: heads move hourly; Conditional κ = 0.001675568675201533: joint entrances with T37 owner packing #1169 force-pushed a reworked package
  during the session. The snapshot pins the audited heads by SHA, exactly as the
  first census pins its own.

Evidence

• verify.py: PASS_COMPLEX_QUEUE_CEILING_CENSUS_R2 screened=13 words=8 unscreened=3
  (identity + split for all 13, exact kappa match, endpoint below a_lo, kappa below
  cap, W4/W3 cross-controls against the first census's pinned inputs, two negative
  controls, frozen RESULT.json byte match). Stdlib only.
• rebind2.py: PASS_SNAPSHOT2_REBOUND 13 - all pinned profile/idle blobs re-read at
  the frozen heads.
• producer.py re-derives SNAPSHOT2.json offline at the frozen heads
  (or with --fetch-prefixes --fetch-titles at live heads).
• scripts/research_package.py --check-only: PASS_PACKAGE_METADATA_AND_PINS, and
  verify.py re-run green on a second host (Python 3.12, git 2.47).

Provenance

Base c22e42de697334df85337e005864773cdc2bc98b; claim thread
discussions/2739; overlaps refreshed (337/144/260 matches on the drawn cards - the
refinement lands on the first census's own recorded remainder). Selected record and
catalogue are historical provenance. SOURCES.json pins the first census and #1240
files; SNAPSHOT2.json pins the 16 audited heads and profiles. Caps are planning
bounds under the fixed coverage law and halving guard; they do not bound changed
words, templates, recipes or the two-unpaid-coordinates contract change, and do not
review any claim.

Prepared by Abhinav Ramachandran (geckods) with substantial end-to-end assistance
from a Hermes agent (Nous Research, GLM-5.3) under the standard donation prompt.