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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 takes the suffix-sensitivity task left open by #4505. It adds an exact reverse free-word response for an additive logical defect after every scalar, warp, climb and center event in the complete original and rewritten THREE-base words, plus per-output/per-class rational error weights and universal sufficient perturbation/feedback theorems.
Under the explicit model that each entire-field ADD/SCALE defect has norm at most δ, the other modeled classes are exact, inputs are exact and logical free warps act isometrically, every persistent output error is bounded by 1,328 δ forward or 4,879 δ backward. Those worst-output constants are unchanged by #4505's 100-unit scalar saving. Unequal defect budgets use the full row ledger; the API requires an explicit nonnegative value for every present class. Initial persistent input error E0 contributes at most 2 E0. A caller-provided recursive defect contract can be closed using the sufficient feedback theorem in PROOF.md.
The 11/13 extra-bit corollary is about absolute entire-field logical defect tolerance, not machine mantissa width or a componentwise rounding quantum. Physical norm conversions, annotation errors and actual recursive defect contracts remain to be proved. No new κ or full rounded induction is claimed.
Coordinatewise nonregression does not follow: 33 forward and 90 backward row bounds change; individual increases reach 2 and 12. These are recorded explicitly. For ADD-only noise with all other classes exact, the backward maximum decreases from 4,308 to 4,300; this does not assert that physical SCALE is exact.
Checks pass: authenticated main source inputs and guarded #4505 rewrite; reverse endpoint columns for all 633 independent initial variables in both original/new words and both orientations; eight complete-suffix forward-injection comparisons; 720 all-event oracle checks on 30 small programs in both orientations, including noncommuting warps, clear/copy and signs; explicit class-budget and malformed-budget checks. Registered replay, committed-byte checks and isolated optimized Git-free export replay pass. Python standard library only; Windows tested, Linux execution unavailable.
Canonical main advanced during final refresh to f9f85c1. This branch is rebased onto it; guidance, selected record and the four frozen input bytes are unchanged. SOURCES.json preserves the initial frozen provenance base 2a13eb1, and its pins replay on the refreshed branch. No selected-frontier change.
Credit #4505 for guarded whole-word integration, #4502 for borrowed adapters, #4499 for gauges, #4494/#4489 for source selectors and rank sums, and the retained zxv / Jacob Alan Sussman / Cinna Davis / Jeremy Kothe native/compiler lineage. MathProof's explicit feedback lemmas are source-level precedents, not freshly checked Lean dependencies. ATTRIBUTION.md records overlap with #3181/#4501/#371 and incomplete Discussion/local-note coverage. Reverse sensitivity and impulse-response bounds are classical; the source-bound application and exact ledgers are the contribution. Substantial OpenAI assistance; Apache-2.0.