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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This scoped proof gives the complete local native/scalar error operator for the current three-field repair in PR #7669. It includes all three fractional scalar ADDs and all three native MOVE residuals, with the actual mixed output charts. If E is the joint Euclidean norm of those six residuals, the new block's output error is at most
C_new E, where 5.435547392075 < C_new < 5.435547392076. This constant is sharp for arbitrary residual vectors with that joint budget. No new kappa, native error allowance or complete caller bound is claimed.Read the all-size local proof, exact spectral and native receipt, literal residual derivation, native field verifier, current-main inputs and copied source pins.
This follows the transport limit in PR #7671, exact source
f34696319eb19d05d1b09f7372c3146be3ea1635. That source supplies complete reconstruction and local window binding. The candidate remainsee251f6005d2a92b887c5b804bd7aac05989d278b2c9ccb3c3091b50d533b08d, with the PR7669 source files pinned at95d476623f1468822057554475dc8badb19bfa44. This result takes residual allowances as inputs. The source contributor's actual rounding and reachable-grid work remains separate.The exact frame quotients are unitary. Their product relation permits an explicit alignment of the six residuals and three outputs. The resulting 3-by-6 matrix includes every path through the literal nine-event block. Its first scalar residual cancels from x but remains in y and z; the native errors have their own coefficients. The old four-port matrix and the ten-port old/new difference are also proved. Their induced constants lie in
(4.582849288401,4.582849288402)and(7.107859846622,7.107859846623), respectively. All three constants have exact characteristic polynomials and rational leading-minor certificates. The tensor and complete-spectator lift preserves them at every positive column multiplicity.The comparison rotates the old x exit and both z outputs into a common chart only to measure the norm. Those isometries are not unpaid program operations. The bounds need no independence among realized errors. Sharpness refers to unrestricted residual vectors; it is not a claim that a particular floor quantizer can attain each extremal vector. Earlier input error must still pass through the exact input/output matrix.
Use Python 3.12 and the standard library:
Expected output includes:
Normal and optimized replay passed. The literal formal-residual calculation matches a separately derived coefficient matrix. The retained direct Gaussian implementation checks 1536 forward/adjoint basis columns across charts 265,751,927. Full 512-component complex-field executions check both exact outputs and one isolated injection at each of the ten old/new residual ports. Four controls detect an omitted native error, a lost factor two, a wrong global phase and missing old-x exit alignment. Committed bytes, source ancestry, the 22-file official export and read-only Git-free optimized replay passed. 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.Limits: this is additional local error on the same entry fields, with exact integral ADDs under a no-overflow contract. Actual native quantizers and their allowances, reachable scalar grids, signed widths, physical conversions, the full sixteen-call error, later analytic recovery and all five inherited framework assumptions remain open. The whole native word, caller, price CLI and multiplier were not replayed. The selected frontier is unchanged.
Base:
e54523517d0f03e58352e647c403a5af3e525a57. @SYLforge supplied the candidate, exact edit and prior rational identity. @CubeJerry supplied the retained direct Gaussian frame implementation, as credited in its notice. @aliromman supplied the pinned source reconstruction and directed this new residual proof and verification with substantial OpenAI Codex assistance. The unchanged source provenance preserves the earlier native, caller and helper lineage. New work is Apache-2.0; copied sources are unchanged.