Repository navigation
Research: independently check QQ16 recovery frames and tensor arithmetic - #4708
Merged
Mister-Meeseeks merged 1 commit intoOct 22, 2026
Merged
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Change and mathematical scope
PR #4109 already proves the common finite format for PR #4015 and its paired successor, including the producer substitution. This adds independent checks of the PR4015 baseline at its native-frame and integer-execution boundaries. It claims no new format theorem, supplier admission or kappa improvement.
The new package,
research/qq16-streamed-decoder-format/, reads authenticated canonical data without importing predecessor executable code. It reconstructs the 73-axis owner maps, checks every ADD frame and the borrowed/private ownership ledger, and rederives the all-cut capacity bound. The actual event-1707 negative q=19 line is executed on dense 512- and 262,144-coordinate complex inputs; bounded integer butterflies agree with a separate translation-convolution oracle, and the inverse restores the input exactly.The existing sufficient formats are corroborated, with
t=lambda-1:The package also separates two reference protocols. PR4109's fused convention gives
(12352+2240 lambda)N; the explicit unfused convention here, including the same cover-transfer subtotal, gives(19434+2240 lambda)N+12. Their difference is accounted for by term copies, explicit negations, monomial sign allowances and final scratch clearing. It is not a new speedup or a contradiction between optimal-cost claims.Task and overlap
Task:
joint-query-controller, refined to deliberate independent verification. Submission base: canonical mainb58ad9ef4d4e861b0eaa5dba41742bdc072dca51, priority revision 7. The historical arithmetic/card baseline remains pinned at20ecf2891d80b5acf18d980b2164fb44ea7faa90. Exact sources and attribution are inSOURCES.jsonandTASK.json.0fff81d8997cef646538bcd4f60d7d455cf93b8e.30dcf52a7722d0919e125f77a4fb921094578242; the format and producer composition retain its priority.Repository overlap tooling refreshed 404 open PRs and 3,573 Discussion records with complete pagination. Relevant QQ16 local notes and targeted merged-PR searches were also checked; unrelated sparse notes and arbitrary diffs remain outside this coverage. No novelty guarantee is inferred from search. Current selected PR4470 and shared/frozen sources are unchanged by this PR.
Validation
Standard-library CPython 3.12.13, macOS arm64:
The registered replay passes: 3,813 events, 10,336 primitive observations, 224 owner maps, 114,688 signed native basis columns, and six behavioral groups. A separate integer/dyadic implementation matches every one-layer observation and explicitly checks depths 1, 2, 3, 8 and 32. Controls include native-interior overflow despite fitting endpoints, off-grid halves, ownership/frame mutations, complete constant-payload word execution, and the dense nonconstant tensor fixtures above.
The committed export passes all three direct scripts with isolated, optimized, warning-strict Python in a read-only Git-free directory, with unchanged files. Commit-byte and full source-history preflight passed for all 13 package/input files at
c55c8d6fed77543ae6c51955c55720cf8a9351c4. Hosted CI remains a separate check.No complete O73 payload, actual installed caller, matched physical tariff, selected cold replay, human review or formal-kernel proof was executed here. Native functional/cover semantics remain inherited; the audit does not repeat PR4109's paired-candidate or approximate-error theorem.
Credit
Nicholas Stone, with substantial OpenAI Codex assistance. Credit Chris R / Clear1nsight for PR4015, PR4043 and PR4109; Nick Petalas for PR3303/3270; the independent PR3652 check; commdata2338, ywz65, Arham Tahir, Sricharan Vagvala, Ali Romman and the retained QQ16/native/geometry lineage. Apache-2.0; upstream source bytes and notices remain unchanged.
Open next questions
Follow the revision-7 matched whole-caller protocol comparison: preserve original parent parking/restore, external-call copying and return, descendant contexts and all later root calls. The numerical certificate does not price that comparison or establish physical ownership, representation conversion or a favorable supplier budget. Another guard refinement is not the missing step.
Merge danger
Two-way; one isolated, opt-in eight-file research package. No shared verifier or selected-result changes.