[AdaWorldAPI]

The plan for D-STAUNEN-5, and its first two phases, P0 and P1. P2–P7 stay gated.

Question: can a novel stimulus trigger a local structural adaptation that changes a recursively computed prediction, and can the system then learn when that adaptation was useful?

Plan

• One binary relation edge over at most 64 ids; its closure reach is the prediction.
• A declared set of dormant candidate edges is the only structural freedom.
• A novel stimulus (an observed pair reach misses) licenses one activation, under the contract it shares with Atlas's update_mask (brian2-atlas dev 9ec52f9): fixed mask shape, 0/1 entries, committed state.
• NotACandidate, InsideRound, Stale, NoClosingCandidate and novelty gating are the cube's own rules. P7 never demands them from Atlas.
• Two competing hypotheses, scored on held-out pairs from a hidden graph G*: H1 parsimony, H2 novelty × eligibility.
• Later evidence consolidates or reverts each update. A selector is judged against the best fixed hypothesis under a pre-registered paired criterion.

Inherited kill condition (from the EXO-50 board references). D-CE64-STAUNEN-0 (#1395) measured that raw surprise is captured by noise: 0.649 resolved against round-robin's 0.964. So:

• H2 runs raw and habituated, and raw is expected to lose;
• a round-robin arm is added, and any policy that does not beat it on held-out pairs has failed;
• outcomes are pooled as NARS revision over root-disjoint proofs (nars::support::DerivationLedger), not Fisher-Z addition.

Open with the operator: is a source root the observer or one report?

P0: fixtures and oracle

• A breadth-first reference closure that shares no code with the cube.
• A seeded episode generator. G* lives only in World; the Episode that later phases hand to the policies has no field for it.
• The oracle labels every novel stimulus one-edge-closable (and lists the candidates that close it) or not. Every stream carries a planted pair that needs two withheld edges, for P3's NoClosingCandidate.
• Observation noise: pairs G* does not reach, presented exactly like real stimuli and labelled only in World. They include distractor traps (closable by one distractor alone) in at least 35 of 39 seeded streams.

P1: resident closure, Cube::closure

• Semi-naive over row masks built from the relation's resident lanes: delta' = (delta ∘ edge) andnot full, then full |= delta'.
• The difference is mask_andnot in the cube. Nothing is added to the cube (CubeStats::relations_added stays 0) and no tuple is built.
• Refuses a map wider than 64 (ClosureWidth) and a fixpoint that does not empty its delta within n+1 rounds (FixpointBound).

Tests (tests/staunen5.rs, 9):

• The fixed fixture {(0,1),(1,2),(2,0),(2,3)} has exactly 12 pairs: (0,0), (1,1), (2,2) in, (3,3) out.
• 400 random cyclic and acyclic graphs up to local position 63, bit for bit against the oracle.
• The longest chain and the 64-ring.
• In every case, rows composed equals |edge⁺|: each closure pair enters the delta exactly once.
• P0 generator checks, including the noise labels.

Guards broken on purpose, each turning the suite red:

Change	How it fails
D1: drop andnot	FixpointBound on cycles
D2: compose from full	same closure, but 24 rows composed against 12, so caught on work, not rounds
D3: drop the union	wrong closure
D4: stop counting added relations	the counter test
D5: drop noise generation	the noise-label test

Third comparator: xlog-cube's closure needed a nonempty r(V,V) fixture first. That is AdaWorldAPI/xlog#5. Comparing all three closures on the same graphs, which completes P1's acceptance, follows once this PR and xlog#5 are on their main branches.

Board: INTEGRATION_PLANS.md and STATUS_BOARD.md.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

[coderabbitai]

Important

• Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 1c4ac718-05c8-4e48-91f1-39caf3c35449

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

• X
• Mastodon
• Reddit
• LinkedIn

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-01T11:44:50.497374Z	729e802	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.