Skip to content

Compile changing-dual junctions: allowance 288 to 224, with safe composition - #6288

Merged
Mister-Meeseeks merged 2 commits into
CrocSwap:mainfrom
CubeJerry:research/dual-exchange-safe-fusion-20261030-s79
Oct 31, 2026
Merged

Mister-Meeseeks merged 2 commits into
CrocSwap:mainfrom
CubeJerry:research/dual-exchange-safe-fusion-20261030-s79

Conversation

@CubeJerry

Copy link
Copy Markdown
Contributor

Replace a common-row, changing-dual exchange junction by three block shears. The named complete two-consumer scalar wrapper costs 288→224 (26→20 emitted ADD/SCALE instructions), after cancelling the baseline's adjacent signs. Raw emissions cost 292. Both outer exchanges and both inner calls remain. Also prove nonlinear composition, exact matching boundaries and the universal safe-observer condition O S=O (linear D=CB). New beyond #6280: the priced junction, nonlinear/observer contracts and independent verification.

Base: 2485805.
#6280 source: 4d6e437. Unchanged compiler byte/blob pins: ORIGIN.json; actual caller tariff: SOURCES.json.

Reproduce (Python 3.12, stdlib):

python3 -B scripts/research_package.py --package research/dual-exchange-independent-audit

Expected: PASS_REGISTERED_RESEARCH_PACKAGE.

Executed: eight normal/-O commands; source-emitter and free-block identities; 54 junction factorization comparisons; independent 2000 nonlinear transition cases and exhaustive 59049 F3 composition cases; exact prefix bound and all input columns. Controls catch missing restoration, wrong pairs, invalid observers, 65-bit differences, coefficient-cap overflow and actual source corruption. A valid unfavorable junction costs 212 versus 132, so improvement is cost-tested, not uniform. Source/history/modes, export and unchanged read-only Git-free replay pass. Fixed finite domains and stopping rules are documented; no random seed. Universal proofs are separate.

No new κ. No actual P104 interval or favorable complete supplier invoice is supplied. Inner calls, observers, transport, native work, restoration, precision and storage remain payable. Exact endpoint equality does not certify a rounded implementation; all five framework assumptions remain.

By @CubeJerry with substantial OpenAI Codex assistance: worker D construction, worker A independent proof/replay, and lead review. Credit Nick Petalas (#6280), Ali Romman (#5611, #5602), Chris R/Clear1nsight and retained lineage. No human/proof-kernel acceptance.

Proof, exact pins, controls and limits: https://github.com/CubeJerry/integer-mult-bounds/tree/a2d6b3920abb26f22b61b776088209e09b03590e/research/dual-exchange-independent-audit
Complete remote Git tree matches the tested contribution.

Prepared for @CubeJerry with coordinated OpenAI Codex assistance.
Credit Nick Petalas PR6280 and Ali Romman PR5611/5602 for the retained
exchange compiler and tariff, and the P104 lineage. No new kappa.
Compare after cancelling the baseline's adjacent sign pair; retain both outer
exchanges and both inner consumers. Exact coefficient admission, prefix bounds,
unfavorable-tariff control and independent nonlinear replay are included.
Prepared for @CubeJerry with OpenAI Codex assistance; credit Nick Petalas,
Ali Romman and retained P104 lineage. No new kappa.
@Mister-Meeseeks
Mister-Meeseeks merged commit 94ad794 into CrocSwap:main Oct 31, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants