Repository navigation
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Exact row arithmetic reduces the required stored scale from 45158400 to 161280 on the fixed chart-595 graph. Accumulate each row's integer common numerator and divide once after the row is complete. Every completed normalized value stays the same. For W-bit donors, W+20-bit temporaries suffice for every product, partial sum, quotient and remainder; an exact-dividing source-row control fails at W+19.
Read the proof, receipt, runtime kernel, checker and scope record. This changes the arithmetic representation: the accumulator carries a pending row divisor. It does not contradict the literal-product grid minimum. The stored-grid minimum is inherited and attained by the new program.
No new kappa or complete supplier substitution is claimed. This is normalized scalar arithmetic, with immutable donors, exact final division and a separate output-amplitude requirement. The graph and paid/target invoice are unchanged. The physical Clifford/root, guarded-stream, movement, dirty-state and cleanup obligations remain open, as do all five selected assumptions. The W+20 arithmetic cap lies below the source's W+26 allowance, but this alone does not prove a lower physical allocation or bill. This completes the advance claim; the active helper-successor search remains separate.
The exact coefficient census finds maximum denominator 6720, maximum integer coefficient 84840 and maximum absolute coefficient sum 683350, all at row 151738. Per real component, the full program has 903276 coefficient-bit iterations, 245747 conditional product additions, 170652 numerator additions and 3607 nontrivial row divisions. Each division has W+19 bit iterations. These are explicit word-operation counts, not a runtime measurement or bit-tape tariff.
Normal and optimized CPython 3.12.16 checks pass on macOS arm64. The complete basis replay covers 8738317 products and 1558861 row-final coefficient divisions; all 7644 row vectors match the pinned literal-product trace. Two full shift/add and binary-division executions agree at every row with a separate Fraction oracle and preserve donors. All 30576 signed row-boundary fixtures pass. Large rows reach W=4096. Five controls reject insufficient width, a nonzero remainder, an out-of-range input, a boolean input and excess coefficient mass. Metadata, source history, committed bytes, modes, export, archive integrity and a read-only optimized Git-free replay pass.
With CPython 3.12 and its standard library, from the repository root:
Expected markers:
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 requiresmake 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. Full candidate and selected replays, physical root/tape checks, Lean and Docker were not run. The all-width statement is a written proof, not finite extrapolation or formal-kernel acceptance. Output installation in W bits still needs the caller's amplitude bound. The finite full runs check that condition at W=64.Prepared by @aliromman with substantial OpenAI Codex assistance. @Clear1nsight supplied PR #7697, the fixed graph and grid figures in its pinned physical proof. The graph and notice are retained unchanged with source pins. Ali Romman's PR #7701, pinned grid proof, supplies the complete reference trace and prior grid audit. Its method descends from PR #7680. No source verifier is imported.
Common-denominator accumulation and binary arithmetic are classical. Nick Petalas (@petalas) supplied the related once-rounded gate in PR #6429, pinned proof. That proof was read; its code, rounding contract and width formula are not transferred. Wentao Li (@VictorLiwentao), crediting Wronskode, supplied related integer-row verification in PR #2409. Credit details and the retained notice preserve the banked/helper/caller lineage. New work is Apache-2.0.
Base: canonical main
e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8. Carduniform-transfer-audit, scope changed to this local arithmetic contract. Live handoffs, refreshed open PR/Discussion overlap, source-reference intake, complete closed snapshot and updated-order closed delta were checked. Sparse local and unpublished work remain coverage limits. The next useful step is a full guarded-stream/root substitution, with every ordinary cost included.