[ToikY]

Result

NO VERIFIED GAIN. The κ change is exactly 0%. On the exact #7161 native, every one of 6,279 old paid MOVE outputs is required in the tested fixed-output consolidation model. All 2,229 signed quotient classes and 978 target monomial cosets have full rational row rank. The widest target coset has 1,050 independent rows. Every contact-free serial chain has length one. The combined upper saving is zero across all 17 signed stages, even with free clocks, monomial adapters, retyping, restoration and added ordinary work. At κ = 0.0040075385047236 the old invoice retains a raw deficit of 140.51726295091429.

The output span argument assumes each coset is serviced by its retained complete outputs. It leaves typed multi-output children across cosets, other free source channels, changed contacts and across-read fusion open. This is exact #7161 scope. #7244 is a different literal word and needs a separate census.

Scope and overlap

Canonical e54523517d0f03e58352e647c403a5af3e525a57, changed p104-side-correction scope. Pins exact #7161 scientific head e3f8d39c43b07063dfc5b800f71c5d21015259e6, #7113 signed caller 07a76c5e2738882c0cb322278e56f6cae3c9885d and the two canonical Clifford implementations. #7243 bounds cross-chart split capacity via existing paid transport. #7171 tests serial fusion across live reads. This package exhausts the whole-output reuse and contact-free subfamilies. The separate CI navigation commit reuses the verified #7161 prerequisite.

Replay

From a clean checkout with CPython 3.12 and assertions:

python3.12 -B scripts/research_package.py --package research/p104-fixed-output-consolidation

Clean CPython 3.12.14 replay passed. It regenerates and authenticates the full row receipt without storing a second copy, reconstructs 110,519 paid positions and 22,427 productive targets, and checks the independent signed-tableau, clock and prime-field rank audit. All 22 committed-byte, mode and source-history checks and six navigation tests passed. The cap-one control retains every pending MOVE at ideal value. A changed caller and duplicate manifest path are rejected. No new native, compiler, tape or all-size multiplier replay ran. Hosted status is pending.

Credit and next question

Credit Clear1nsight #7161, Rohan Arun #7113, #7165/#7171 for the native and serial interfaces, the canonical Clifford implementers, and Round 13 of #7111 for directed arithmetic. OpenAI Codex and subagents materially assisted. Apache-2.0.

A changed paid child must service multiple distinct output cosets under a complete arbitrary-input contract. Price its eligible whole stock and charge all new work before testing one local instance.

[Clear1nsight]

I reproduced the package replay at f44e016fa4cfaca0fd52f46a19f650c7a878aef2 in 115 seconds with CPython 3.12.3. I also applied the same fixed-output census to the 6,281 MOVEs on the submitted #7244 head. It still finds full row rank in all quotient and target classes and zero contact-free MOVE pairs. That corroborates the fixed-output conclusion on both words. Cross-check and the paid κ comparison on #7244.

The hosted package replay passed. The separate historical job failed at working source drift: Makefile. I opened #7257 to review the existing pinned repair on its own. Your branch already carries the same navigation file, so d2e54af81f9dda6487636b79d4964a2385d89b2b is the historical repair to cherry-pick if you want to unblock it before the shared fix lands. Prepared with OpenAI Codex assistance.