Repository navigation
Audit QQ16 child-error grid closure and certify sharp projected guards - #3189
Merged
Mister-Meeseeks merged 1 commit intoOct 17, 2026
Merged
Conversation
This was referenced Oct 17, 2026
Merged
Mister-Meeseeks
pushed a commit
that referenced
this pull request
Oct 17, 2026
…the local native core Lifts PR #3189's B1,2 child-error witness through the literal local native suffix instead of treating non-scalar maps as identities. The Gaussian-integer direction w = T_236 ONE (entries in {-1,0,1}, squared norm 512) is injected after child 13489 on L344, and the 13,529-event suffix is executed exactly. After scalar 26472, L93 is exactly (-81921/262144) ONE: 17 fractional positions cannot represent this channel, while 18 suffice at every completed event for this direction. Final data errors are L75 = -ONE/2 and L93 = ONE/2; six borrowed helpers (L337, L338, L339, L341, L343, L346) retain errors and all private fields are zero, so an endpoint-only grid check misses the intermediate requirement. A separately implemented sparse Fraction engine matches the complete changed-field trace and every endpoint vector, and four reverse adjoints check the attaining and terminal components. This is a finite certificate for one direction at completed events under a completed-child defect model. Actual child rounding, recipe interiors, dirty-input bounds, helper restoration, global embedding and the paid consumer remain open; no supplier, selected-record or kappa change is claimed. Adds research/qq16-native-noise-lift/ only. Submitted by @commdata2338 with substantial AI assistance (OpenAI Codex research and implementation; packaged with Claude Code). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01X86sCSXdoDd61SDFfC1PKG
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
2 participants
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.
Result and scope
PR3181 uses the dyadic denominators of ideal input columns to set scalar rounding error to zero while allowing approximate native children. A returned value on the working grid does not by itself make the next half-ADD exact. This package gives two adjacent, source-bound child/half-ADD countermodels and a corrected lattice criterion that includes independent child-error columns at every intermediate cut.
In the same scalar projection used by PR3181, a complete census of 7,004 child-error columns gives sharp uniform guards:
Twelve B1,2 child sites need the eighteenth position. One unit defect after child 13489 on L344 has coefficient
-81921/262144in L93 after scalar 26472. The odd numerator supplies an exact lower witness; the all-column replay supplies the upper bound in the stated projection.The global census treats non-scalar operators as identities, explicitly records that scope, and covers F0/B1,2 only. The adjacent local witnesses retain the actual instructions with no intervening operation. These establish a missing receiver-side condition under an error-only child contract. They do not establish a full native rounding failure, disprove every possible
50+betaaccuracy implementation, or supply an admitted physical bit width. A defect lattice is also distinct from a returned-value lattice. No new kappa or selected-frontier change.Verification
Expected:
PASS_QQ16_NOISE_LATTICE_AUDITwith input guards4,17and child guards4,18; six independent witness checks; eleven passing tests;PASS_REGISTERED_RESEARCH_PACKAGE.The primary exact replay checks every input and child-error column at all 42,426 completed scalar cuts. A separate implementation uses dense single-channel forward propagation and reverse duals for the four sharp attainments and two adjacent examples. Negative controls cover altered witnesses, insufficient guards, source corruption, endpoint cancellation and later clearing. Registered Python 3.12.14 replay passes; a read-only, Git-free export of the committed nine package/input files passes on optimized Python 3.14.7 with unchanged bytes. The original PR3181 read-only verifier also passes with its repository-root variable redirected to the authenticated input checkout; its files were unchanged.
Provenance and credit
Base:
997e6a9ca7b4f95d50e366f61412fd640d72a88a. Audited PR3181 head:80fc72d14546088e9fce3c06dc12bc67270da0af. The live joint-query certificate and decompressed payload are hash-bound; task metadata records audited claim hashes and the seed-61 scope choice. Fresh overlap covered 306 open PRs and 1,973 discussion records, with no matching child-error guard census identified. Earlier #2995/#3017/#3036 faults have different coverage.Credit dcharb78 for PR3181's ideal-column/adjoint work, the retained QQ16 native-word authors, Nick Petalas / #2965, Arham Tahir / #2995/#3036, and Tekalign / #3017. Prepared by Tekalign (
tekuuu) with substantial OpenAI Codex assistance. Apache-2.0. Only the new isolated research package is added; existing sources and shared tools remain unchanged.Investigation claim.