Skip to content

Certify the minimum product and snapshot grids of the banked native DAG - #7680

Open
aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-024
Open

aliromman wants to merge 1 commit into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-024

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Result and scope

The final forward scalar graph of PR #7674 needs odd factor 225 and 12 binary coefficient bits for every literal product and addition. Its recursive denominator bound gives the sufficient values 2025 and 18. This audit computes the exact minimum and proves sharpness with basis-input witnesses.

Scalar states Minimum integer grid Odd factor Binary coefficient bits
Every literal product and addition 921600 225 12
Completed snapshots only 23040 45 9

The product grid is 576 times coarser than the source's sufficient grid. This is an arithmetic precision result, not a speedup or physical memory measurement. The snapshot grid is insufficient for product work: literal executions on two named basis inputs stop at nonexact divisions. The proof gives the separate prime-factor lower bounds and retains the exact operation order.

Read the proof, receipt and failure locations, new arithmetic checker, source inventory and scope record. The source graph constructor is reused and authenticated. This is a new coefficient audit, not an independent graph compiler.

No new kappa, selected-assumption discharge or supplier admission is claimed. The result concerns arbitrary independent normalized scalar inputs and the pinned rational DAG. Physical Clifford factors, the signed caller, recursive depth, magnitude, final unscalings and a full common-root precision substitution remain separate. All five assumptions and selected PR #7000 remain unchanged. This completes the advance claim on the arithmetic portion of the source's review request.

Evidence and reproduction

Normal and optimized CPython 3.12.16 checks passed on macOS arm64. The checker evaluates all 8533666 coefficient products at each grid, compares all 7643 snapshots under exact rescaling, and checks all 1501 public endpoint rows. A separate Fraction interpreter evaluates four basis inputs and confirms all six sharpness witnesses. Eight negative controls reject missing prime factors and use of the snapshot grid for products. At the minimum product grid, every literal division is exact.

Metadata, committed bytes, executable modes, source ancestry, the official 25-file export and its integrity check passed. An optimized read-only export replay passed without Git or a usable PATH, with all 26 payload/manifest hashes and modes unchanged. The 15 retained predecessor files match their published Git blobs and SHA-256 values, including the original mixed line endings in a source notice. No external run-time input is used. The new proof and README were rendered with GitHub Markdown and checked.

From the repository root, with CPython 3.12:

python3.12 -I -W error -B research/banked-native-minimum-grid/verify.py
python3.12 -I -W error -B -O research/banked-native-minimum-grid/verify.py
python3.12 -B scripts/research_package.py --package research/banked-native-minimum-grid
python3.12 -B scripts/check_research_commit.py --package research/banked-native-minimum-grid --check-source-history --check-modes

Expected status markers:

PASS_BANKED_NATIVE_MINIMUM_GRID
PASS_REGISTERED_RESEARCH_PACKAGE
PASS_RESEARCH_COMMIT_BYTES

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. Predecessor scientific suites, physical native operators, the selected cold replay, Lean and Docker were not run. Independent human and formal-kernel review are unclaimed.

Credit and remaining obligations

Prepared by @aliromman with substantial OpenAI Codex assistance. The banked graph and sufficient common-root proof are Ali Romman's PR #7674, exact proof and source arithmetic evaluator. Its scientific commit is 26b6c469aa6c4aa04322ce25a22150f90e5fb14e; current head 28b2af322a1fab01b3734365988930920f9dda48 adds the shared CI repair. @SYLforge supplied PR #7669; the original credit and source notices are retained unchanged. Gareth Davidson (@bitplane) supplied the prefix-coefficient grid criterion and basis-witness method in PR #866, exact precision proof. The method and source construction are prior work; this contribution binds the sharp product and snapshot grids to the new final graph. New work is Apache-2.0.

A complete physical substitution must keep the caller's separate scale and check the Clifford, unfinished-depth and final-unscaling contributions. No smaller selected root format is asserted here. A different arithmetic lowering could change the temporary grid. Base: canonical main e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8; card supplier-boundary-audit narrowed to this finite obligation. Open/closed PR and Discussion overlap was checked; sparse local notes and unpublished work remain coverage limits. No source PR or selected artifact is edited.

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