Repository navigation
Conversation
This was referenced Nov 19, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
At tensor width 13, a schedule of 52 dyadic Gram applications replaces the earlier 8,192 fixed-step updates. It contracts the initial decoder error by more than a factor of 2¹²⁹. Every prefix preserves exact left inversion and has error amplification at most one. Interpolation on the 52 exact spectral levels proves that no normalized residual polynomial of degree at most 51 achieves the specified 2⁻¹²⁸ contraction.
A second schedule uses 51 applications and attains the exact optimal noise norm. It uses 50 dyadic scalars and one reciprocal with the least possible odd denominator, 470,624,175. Its degree lower bound is uniform over initially feasible decoders within the stated norm bound. Both counts assume access to a Gram application and the input adjoint. Pricing those operations and the underlying data movement remains open.
The two replays reconstruct the scalar schedules and the spectrum independently, verify every prefix, and detect the large intermediate growth caused by reversing the first schedule. Nick Petalas supplied the #7747 character projection, Ali Romman the #7772 spectral levels, and #7799 and #7802 the starting inverse and sharp denominator. The selected multiplication bound is unchanged.
All four registered CPython 3.12 checks passed in the checkout and from the read-only, Git-free export of the submitted commit. The repository checks also verified committed bytes, source history, file modes and the export inventory.
Prepared by apoursh with substantial OpenAI Codex assistance. Proof at the submitted revision · Research claim
Hosted workflow. The shared onboarding test failed before package selection. Its CI guide names #6030 while the selected record requires
make frontier7000-verify. I reproduced the same failure on the unchanged upstream files. All registered package checks passed locally and on the exported commit. Failed job · Open CI guide repair #7276