Selected packing: all 21,886 cross-call 0,1,7 to 0,0,8 cycles fail timing

[navinor]

Result

On the original selected PR 1094 owner map, no simultaneous cross-call
three-role cycle among distinct occupied bins can coalesce idle ranks
{0,1,7}->{0,0,8} with the existing event windows. Complete finite domain:
21,886 capacity-feasible cycles, all window-blocked. No new supplier or κ.

This takes one bounded slice of PR 1873's open cross-call continuation. It
does not repeat the known sequential neutral-then-strict relocation census
or claim to close all three-role exchanges.

Exact obstruction

The complete source type prefilter supports only:

A,B,C (slack,width)	Cycles	A constant failing leg
(7,8),(0,8),(1,7)	17,567	C→A: event 181 outside (255,430)
(7,8),(1,7),(0,7)	4,319	A→B: event 417 outside (-1,208)

The sole (7,8) role is (stage,replica,helper)=(7,14,296), bin (13,15,14);
the sole (1,7) role is (1,15,226), bin (11,8,103). Their fixed timings
reject each whole row independently of the third role. The exclusion holds
even with active-pair and family restrictions relaxed. The favorable idle
delta -H1-H7+H8 would lower the complete fixed-word moment for all 0<a<1 by
strict concavity if legal; none of these candidates is legal.

Assignment, provenance and coordination

• Agent: Aaron Wang / navinor, with help from chatgpt; substantial research,
  math, implementation, documentation and separate read-only AI review.
• Track/card: incumbent-optimization / selected-packing, scope refined.
• OS nonce 3474c3e21150f5db2b291dfc2989aac6; catalogue key
  2eb6f6918b10e182033b; priority revision 4.
• Base/current canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2;
  selected supplier head 0fc329d0bec93ca7cd288eaea1295172e8789810.
• Canonical claim:
  Selected PR1094: exact single whole-helper relocation census #1332 (comment).
• Exact live input Git blobs/SHA256 pins are in SOURCES.json; catalogue and
  selected record are retained as historical provenance. No source copies,
  predecessor edits, selected-record changes or dependency installation.
• Fresh upstream fetch at 2026-10-14T20:39:50Z: canonical main unchanged;
  instruction/catalogue/priority/selected/source compatibility diff empty.
• Final text-based overlap refresh: 396 open PRs, 807 discussion records,
  complete pagination at 2026-10-14T20:41:25Z, plus local notes/ledger. No
  competing exact slice found; no novelty guarantee.

Nearest owning sources retain credit:
#1873
(08f1eaf5946f57de0e126e82b5e0432564cca4e0, same-call three-cycles),
#1790
(48ee04ee708081c8713e8868a370bbbf5aa099e1, two-role swaps),
#1617
(536bc90ea2de310e0e08d2cb65f7add1d9de92b5, complete sequential two-relocation
census), #1407
(d8640f5b7ae99b43f5d56146cb94ca62b8b3835d, longer sequential paths), and
#1154
(390de2866a03758e7f4ddbb2846e5b2ee4e72154, positive cycle on a different word/map).

Evidence and verification

Adds only research/selected-crosscall-rank8-cycle/: README, TASK, SOURCES,
VERIFY, RESULT, verify.py, test_verify.py.

Primary type-count/indexed join and separate role-first direct enumeration
agree on all 21,886 candidates and zero all-window-pass landings. The latter
does not call the primary transfer or landing predicate. Both methods share
source decoding and PR 1091 geometry, explicitly disclosed. Constant-block
objects and a digest of the complete direct candidate-ID list are retained.
Acceptance regenerates the census, not a hash-only assertion.

Controls: six synthetic toy checks, eleven unchanged PR 1091 owner mutations,
omitted-count and int/float receipt rejections, a rank-key serialization
round-trip/int-float/int-Boolean regression, and rejected over-budget runs
on blocked first landings or empty input. No real attaining cycle or
modified-map admission is claimed. All registered commands run optimized.

Both complete pinned supplier profiles are repriced with exact rational
intervals: selected endpoints pass, next 10^-24 ticks fail; bit's inherited
10^-16 envelope of 32*m^2 rank-one fallback calls per paid child is included.
Complex remains binding. ceiling(local,73,896,9,64) reproduces
a<0.001690054109089, balanced planning cap κ<0.001687202646;
0% of packing headroom realized. PR 1847's tighter fixed-route ceiling is
also credited in the README, not newly replayed or claimed sharp.

python3.12 -B scripts/research_package.py --package research/selected-crosscall-rank8-cycle --check-only
python3.12 -B scripts/research_package.py --package research/selected-crosscall-rank8-cycle

Local Windows/Python3.12 output:

PASS_PACKAGE_METADATA_AND_PINS (not a mathematical acceptance check)
PASS_SELECTED_CROSSCALL_RANK8_CYCLE_OBSTRUCTION candidates=21886 all_window_blocked direct_crosscheck=PASS owner_controls=11 new_kappa=NO
Ran 4 tests in 7.064s
OK
PASS_REGISTERED_RESEARCH_PACKAGE research/selected-crosscall-rank8-cycle outcome=completed

Separate read-only AI review independently rebuilt raw source geometry,
all supported types and all 21,886 candidate IDs, confirmed both constant
window blocks and exact prices, and passed a 200-case family/cache differential
check. Its deadline-bypass finding was reproduced by a failing test and fixed
with guards covering setup, skipped branches and completion; the same reviewer
reran the original bypass and confirmed rejection. No unresolved issue remained
in that targeted follow-up.

No hosted CI success, independent human review or formal-kernel execution is
claimed. No native/controller/invoice/outer/full cold replay was run. Scalar
word, all widths/calls/order, actual windows and prefix model are held fixed.
All inherited all-size/compiler, realization, hosting, precision/tape,
analytic and bit H1–H5 obligations remain open/inherited.

Handoff and credit

The larger cross-call three-cycle domain remains open. A bounded next step is to
classify another supported transition touching a slack-7 bin, including old
slacks involving rank 3, before joining exact windows/families. Three-component
transitions need not be majorization-ordered; the simple pairwise sign rule
is not sufficient. Structured next question is in TASK/README.

Rohan Arun/PR 1094; retained Jeremy Kothe mutable-leaver/Hao Yan controller
lineage; Josie Elliston/PR 1091 owner audit; PR1790/PR 1873 and cited relocation
work; Douglas Colkitt/PR 1240 ceiling; PR 1012 coverage; PR 1847 fixed-route
context; PR 589/PR 555 and retained bit lineage. Apache-2.0.

Workflow feedback: the live overlap check prevented repeating the saturated
neutral-relocation domain. The useful certificate is two constant timing
inequalities, not a large search archive. Receipt replay exposed numeric-rank
JSON key ordering; a failing regression led to normalization of keys before
sorting, while int/float and int/Boolean values remain distinct. The review
also caught a rejected-landing path that bypassed the deadline check; the
stop rule is now covered by a deterministic regression. Temporary
worktree cleanup follows publication at the donor's request.