Repository navigation
D-BAG-FOLD evidence: multi-hop existence from resident presence (lab) - #1937
Merged
Merged
Conversation
Preserve the co-addressed fold falsifier and add the witnessed existential probe under .claude/probes/2026-11-03-d-bag-fold/ (own [workspace], not built by the root workspace). Originals are byte copies of the session artifacts with sha256 in the README. Findings, against a nested-loop oracle on 8 cases: - a key-prefix range is not edge membership (false positive with R or S deleted); a range count is not DISTINCT; a concept-only prefix crosses classids; - the existential evaluates in one Program in two arms: an ADMISSION B prepared projection (gather-free, copies presence) and pointer tables plus Semijoin over the table's own planes (zero-copy, uses Gather); - plane order, key uniqueness and presence version are each a silent wrong answer when unchecked. No production code changed. Board: new entry, ISS-MULTIHOP-PRESENCE-ARM-UNRULED, regenerated entries and fruits indexes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars.
 |
…OT IN FORCE) Add the decision-preparation memo for the zero-copy pointer lens (Arm L) beside the evidence it rests on, plus the first draft as a byte copy. The memo recommends keeping Arm L lab-only and requesting owner classification (EXO-140). It records that ADMISSION B's "execution remains co-addressed" and the later zero-copy $i doctrine have not been reconciled, and that a structurally derived index does not admit a MaskOp::Gather. It separates provenance, bounds, ordered-lane identity, key uniqueness and presence-version invalidation, and lists frame, ClassId, codebook and scope consistency as required witnesses. Arm P is kept as a negative comparison, not a fallback. The eight-case oracle results and the six disable-run falsifiers are preserved. The DT64 WHEN/DO route is marked independently gated. The memo ends with an open operator question. No code, probe, mask-risc or board change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF
AdaWorldAPI
marked this pull request as ready for review
November 3, 2026 14:08
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
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 PR preserves the co-addressed fold falsifier and adds a probe for the open question: can
∃d. R(a,b(d)) ∧ S(b(d),c(d)) ∧ T(c(d),d)be evaluated natively from resident presence, without storing any composed relation?Everything is under
.claude/probes/2026-11-03-d-bag-fold/. The probe carries its own[workspace]table, so the root workspace never builds it. No production code changes; mask-risc, Quack and the xlog cube are used as they are.What was found
Tested against an independent nested-loop oracle on 8 cases: all edges present; each edge deleted; several candidates with one, two or none valid, including decoys that break at different hops; a chain held by another classid; a version change.
And → Any | Count. No Gather.Filter::Semijoinover the table's own resident planes. No presence bit is copied, but it executesMaskOp::Gather.r2il::preparewithGuardNotExecutable. No admitted DO word shape expresses the conjunction.Full classification and the smallest missing pieces are in
REPORT.md; how to run the probe and its expected output are inREADME.md.Board
entries/2026-11-03-d-bag-fold-witnessed-existential.md.ISS-MULTIHOP-PRESENCE-ARM-UNRULEDinISSUES.md.entries_index.py --writeandfruits_index.py --writewere regenerated.Validation
Ran
cargo +1.99.0 run --bin v1_prefix_vs_presenceand--bin v2_witnessed_existential: 0 mismatches against the oracle. Source is514948203.The probe needs rustc 1.99, because ndarray 0.17.2 requires it. It also needs the sibling
../ndarraycheckout.Limitations
OrderedLaneWitness.Implementation is not authorized. It waits on an operator ruling between arm P and arm L.
Generated by Claude Code