Skip to content

Execute the half-sum CLEAR on fixed tapes with a complete scan bill - #7728

Open
aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-056
Open

aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-056

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

This gives a fixed-tape execution and complete local scan bill for the exact half-sum CLEAR circuit in PR #7721. At width 1076, the emitted program uses 589,841 unit data instructions and 11,206,982 interpreter steps. It preserves all donors, implements the target XOR and invalid flag, restores every work cell and temporary bit, and returns all 26 heads. No new kappa is claimed.

Read the proof, receipt, emitted binary program, compiler and tape executor, checker, source pins and credit. The scoped claim answers the source's released tape-layout question.

The machine has 25 data tapes, one read-only program tape, a five-symbol alphabet, two temporary control bits and a width-independent finite decoder. Nine binary bits encode each unit movement, control read, target toggle or cell initialization/retirement. The program is read and returned in full. No random-access gate lookup is assumed during execution.

Resource at width 1076 Exact count
Data bit cells 24,788
Work bit cells, included above 15,103
Peak data markers 50
Marked program cells 5,308,571
Unit data instructions 589,841
Interpreter steps 11,206,982

For every positive width, the proof gives L<=1222w+2952 instructions and 19L+3<=23218w+56091 interpreter steps. It charges ripple access, flags, output copy, reverse execution, initialization, retirement and all head returns. This bound follows from the literal column loops; it is not fitted to measured values. Work cells start blank and return blank. The original external markers and program remain unchanged.

The entrance supplies the marked width-dependent program, input blocks, origins and reserved blank regions. Uniform physical program generation/loading and enclosing stock/reset services remain open. This is a classical multi-tape execution model; transition-level reversible control is not claimed. The complete external XOR map is reversible and its local workspace is restored. No selected owned-erasure or precision-tape assumption is discharged.

Base: e54523517d0f03e58352e647c403a5af3e525a57. The six scientific files from PR #7721, published head 8f6279527c4362e2638737c90fa18c3cfec724fa, are retained unchanged. Its source proof and credit establish the all-width arithmetic map. Its source verifier is rerun. The native event88937 identification is inherited from PR #7687 and its exact interface; that whole native check was not rerun. PR #3233 prices a different pooled field ADD; no code or constant is reused from it.

Run with CPython 3.12 and its standard library:

python3.12 -B scripts/research_package.py --package research/half-sum-clear-tape

Each registered command prints the retained source result, then:

PASS_HALF_SUM_CLEAR_FIXED_TAPE {'1': 1024, '2': 65536, '1076': 32} 11206982

Normal and optimized replay pass. The tape checker executes all 1,024 width-one combinations, all 65,536 width-two donor tuples with varied output bits, and the 32 published width-1076 fixtures. It checks every tape cell and head against direct integer arithmetic. Five damaged streams are rejected. An endpoint-preserving extra right/left detour costs exactly 38 additional steps. Source bytes, committed history/modes, package export, sealed read-only Git-free replay, public links and rendered publication readback pass. Hosted CI did not certify this package. The plan job failed in the shared test_current_ci_section_matches_selected_record: the base guide names frontier6030 while the selected record requires make frontier7000-verify. The log reports 760 tests, one failure and 24 skipped; package replay was skipped. This package changes no shared CI file. Local replay remains the scientific evidence.

No full caller or outer calculation changes: the selected saving remains .00363005817 with all five framework assumptions. Program preparation, global stock/reset, rounded producer reachability and native/caller admission remain explicit obligations. Full predecessor suites, selected cold replay, Lean and Docker were not run. This is source-aware author checking, not blind, independent human or formal acceptance. No selected-frontier or other workstream change is included.

Credit Nick Petalas for the specialized circuit, proof and verifier; SYLforge for the source word, coordinate identification, validity interface and comparator; and the retained native/compiler contributors. The source README verifies these attributions. Original files and notices remain. Prepared by Ali Romman with substantial OpenAI Codex assistance, Apache-2.0.

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.

1 participant