[petalas]

Summary

An independent, separately written recomputation of the merged
research/p104-cone-cut receipt, from the same pinned witness and certificate,
plus one added structural comparison. No new κ; the selected frontier is
unchanged.

 one-step cones: 1,725 forward edges, backward 12 -> 22 -> 37, intersection 0
+ every published merged field reproduces under a second implementation
+ added: the emptiness is a property of the one-step-per-dimension schedule

What reproduces

Published value	Merged receipt	Independent run
forward cone at dimension 5	1,725	1,725
backward seeds / growth	12 / 12, 22, 37	identical
backward departure-slot histogram, writer arrivals	12 slots, 142:2, 143:10	identical
intersection, 37-edge sha256	0, 9f44aa43…	identical

Two inherited quantities also reproduce: the marker counts at dimensions 2 to 5
(25, 52, 365, 1725) and the dimension-2 incoming departure-slot histogram
(165:9, 204:7, 205:9).

What is added

The merged receipt stops at the first disagreement between two implements of
the same rule; this package re-derives the geometry, the relation and both
propagations in separate code and compares field by field. Beyond that, it
iterates the same relation to its fixed point at the same dimension. The
backward closure then grows to 99 edges, 26 of which lie in the one-step
forward cone, and the settled forward propagation reaches all 2,309 edges by
dimension 3. The emptiness of the cut is therefore a property of the frozen
one-step-per-dimension transfer, not of the adjacency relation alone, so no
relation-level strengthening of this cut is available. The merged claim is not
contradicted; the scope of what it can be asked to support is narrowed.

Reproduce

CPython 3.12, from a repository root:

python3.12 -B scripts/research_package.py --package research/p104-cone-cut-independent-check
python3.12 -B scripts/check_research_commit.py --package research/p104-cone-cut-independent-check --check-source-history
python3.12 -B research/p104-cone-cut-independent-check/verify.py
python3.12 -B -O research/p104-cone-cut-independent-check/test_verify.py

The registered replay of all three commands takes about 77 seconds locally
(verify about 20 s, controls about 35 s) under a 300-second per-command budget,
so the 10-minute CI bound has margin. Local results are from macOS with the
repository's CPython 3.12 virtualenv; the hosted replay is the authoritative
certification and is recorded below once it finishes.

Limits

• The 4,992 required-pair accumulation, the interval/literal missing-pair
  readings, the backward-cone licence, dirty restoration and the fixed-cover
  consumer contract are not re-derived here; no construction, physical
  invoice, supplier or consumer is supplied.
• The saturation comparison is about iterated closures of the same relation,
  not about any executable schedule in the frozen word.
• This is an independent finite check of one published receipt, not a proof of
  the transfer model and not a new mechanism.
• Independent human and formal proof-kernel review are unclaimed.

Credit

The checked receipt is research/p104-cone-cut by jwr, with the assistance that
package records; it credits Nick Petalas/PR3124's pooled-cut checker for the
archived relation rule. The frozen witness and certificate are used unchanged
and retain their own predecessor credit. The separate recomputation, the
saturation comparison and the structural guards are this package's new content.
Produced by Nick Petalas with AI assistance.

Hosted replay

The registered replay job certified both commands on the hosted runner at head
4bf5553ba77c6951c7854c26e4d856bee7ac4351: the plan job passed in 2m04s and
replay (research/p104-cone-cut-independent-check) passed in 1m58s.