docs(D-SAG-3): OpenPsi × Rubikon × CARE wisdom-frontier research and ONA/Hyperon probes

[AdaWorldAPI]

Why

Follow-up to merged #1509 / D-SAG-1 (based directly on its merge commit 42fa2a3). #1509 proved one-hop SoftwareGraph projection and ordinary Quack execution for real MethodTable entries. This draft asks whether that same substrate can host a falsifiable wisdom-seeded epistemic frontier and support motivating, deliberating about, and auditing investigations without introducing a second graph engine.

Research synthesis: OpenPsi demands + Heckhausen/Gollwitzer Rubikon + Kanban lifecycle + CE64 test-local Maturation3 (7→0) + typed prerequisite popcount + calibrated Shannon EIG + CARE reference-conditioned executable scoring + independent causal ablation + ONA/Hyperon comparison.

Included

1. D-SAG-2 research/implementation plan — concrete substrate mapping, stage-0 recursive frontier, masked missing/refuted/known distinctions, schemas as proposals, explicit MultiHopBlocked, OpenPsi→Rubikon→permission→receipt loop, CARE gate, pre-registered metrics, 16 numbered falsifiers and stop conditions.
2. Annotated .arxiv literature shelf — 13 source treatments with links, short paraphrased excerpts/takeaways, claim-vs-transfer distinction and disconfirming experiments: CARE (2606.14581v5), Hyperon (2310.18318), metagraph folding (2012.01759), Shannon, Oudeyer/Kaplan intrinsic motivation, causal EIG (2103.15229; 2203.02016), NARS self, ONA MeTTa-NARS, OpenPsi incl. upstream open PR feat: 6 EmbedAnything patterns — auto-detect, chunker, tensor bridge + practices #117, Rubikon, modern PLN.
3. Claude Code execution prompt — actionable P0 source census → P1 scalar oracle/disable-red probes → P2 legal Quack parity → P3 pinned foreign oracles → P4 scheduler ablation → P5 CARE/Rubikon simulated actions → P6 SAG optimization proposals. Includes hard acceptance gates, nonclaims, exact starting paths and runtime links.

Architectural decisions explicitly not made

• CE64 bits 50..52 currently mean plasticity in production; planner: residue fixed point + Maslov Pyramid of Cognition probe (D-RESIDUE-FP-0, D-MATURATION3-0) #1494's 7..0 maturation is test-local and provenance-gated. No silent reassignment. CE64 59..63 remain epistemic readings, not desire or action.
• Shannon entropy/EIG need a declared calibrated probability distribution, never blind NARS f/c→probability coercion.
• Derived premises inherit independent observation roots; no manufactured evidence or self-support, including reopened cycles.
• CARE paper = reference-conditioned policy comparison and auditable pre-outcome gate; causal ablation/Pearl tests are a separate proposed oracle.
• SoftwareGraph one-hop and D-SAG-1: software-awareness graph — lance-graph's own methods, programs and evidence as Quack-queried rows #1509 H1 obligations still apply. No arbitrary traversal, new Quack opcode, second registry, unverified SIMD fused terminal, change to authorization, or executed optimization.
• ONA v0.9.3 exposes a MeTTa-NARS integration seam; not official CARE parity. OpenPsi PR feat: 6 EmbedAnything patterns — auto-detect, chunker, tensor bridge + practices #117 is an open proposal, not merged behavior.

First falsifier

With one independent scalar oracle and identical fixtures, compare plain fixed-point hydration and stage-0-seeded popcount priority over: A→B→C, independent vs duplicate roots, grounded/ungrounded cycles, after-anchor ablation, 0/1/3 missing typed prerequisites, refuted vs unknown, stale schema seal and storage-order permutations. Forbid 0 missing → automatic certification. Require guard-disable RED and restore evidence before any engine/attention benchmark.

Only after this: matched-budget FIFO, confidence, popcount, naive entropy (negative control), calibrated EIG, OpenPsi demand and CARE-gated challenger arms; freeze inference rules when varying scheduler, and vice versa. Count total work.

Status

• Directly based on D-SAG-1: software-awareness graph — lance-graph's own methods, programs and evidence as Quack-queried rows #1509 merged main SHA; compare reports 8 commits, 6 Markdown files, 0 unrelated changes (3 original files + 3 expanded source/plan/prompt files; 825 additions).
• Source-gated literature review, testable hypotheses, gaps, and reusable Claude Code prompt.
• No P0–P6 implementation or external/oracle runtime performed in this draft.
• No performance or cross-engine equivalence result asserted.
• CI/checks and reviewer decisions pending.

Research-only, regardless of GitHub review flag: source census, semantic oracle and CE64 semantic-layout decision remain open. No auto-merge or production feature claims. The PR's current review state is left unchanged.

Related: #1494 maturation, #1497 PLN/Shannon, #1498 Hyperon translation, #1500 Rubikon, #1509 SAG.

Second wave added: deferred hard problems, not hidden feature claims

Three new linked documents and two original handoffs updated.

• Hard-problems plan: four-state masks (known/refuted/missing/unread); alternative supports; root retraction with independent full recomputation; high-degree frontier memory bound; provenance/snapshot invalidation; OpenPsi tie/hysteresis/starvation; Rubikon permission races; graph ablation != interventional causal truth; ONA/Hyperon fieldwise trace matrix. Adds 18 proposed adversarial falsifiers A16–A33.
• Second-wave source excerpts: Friston et al. (Nature Communications 2026 / arXiv:2512.21129), Wei (arXiv:2408.06542), Catanach/Das (arXiv:2304.07949), Pathak (arXiv:1705.05363), Burda (arXiv:1810.12894), McSherry et al. differential dataflow (CIDR 2013), Beckers actual causality (2021), and existing planner: residue fixed point + Maslov Pyramid of Cognition probe (D-RESIDUE-FP-0, D-MATURATION3-0) #1494/docs: OpenCog PLN × Shannon attention research map and falsifiers #1497/docs: Rubicon volition source excerpts and falsifiable agent mapping #1500/D-SAG-1: software-awareness graph — lance-graph's own methods, programs and evidence as Quack-queried rows #1509 caveats. Eight annotated treatments, not claims of executed replication.
• Red-team Claude Code extension: first implement only A16–A19 after a source audit in a separate future implementation PR; fix an independent oracle and negative/silence twins before performance claims.

New first-risk priority: correct deletion/retraction and unread-vs-false semantics. A simple popcount speedup is not useful if withdrawing one independent observation leaves a stale learned consequence, or an unread directory edge is mistaken for an empty permission set.

Explicit nonclaims: no Rust experiments, outside-runtime tests, incremental engine, authorization integration, CE64 migration, new opcode, cryptographic audit signature, ONA-Hyperon parity or causal-awareness result were produced by these documentation commits. Any benchmark, SIMD gain or implementation is still pending.

[coderabbitai]

Warning

Review limit reached

Your organization has reached its usage spending cap. Adjust your spending cap in the billing tab.

Next included review available in 31 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available. Your 62 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Essentials
• Run ID: ce8d5eaa-2ff9-49bd-a22f-8beeb4ec8e14

---

📥 Commits

Reviewing files that changed from the base of the PR and between 42fa2a3 and d8040d2.

---

📒 Files selected for processing (6)

• .arxiv/2026-10_openpsi_rubicon_care_wisdom_frontier_source_excerpts.md
• .arxiv/2026-10_wisdom_frontier_active_inference_delta_and_misspecification.md
• .claude/plans/sag-wisdom-frontier-hard-problems-v1.md
• .claude/plans/sag-wisdom-frontier-ona-hyperon-v1.md
• .claude/prompts/sag-wisdom-frontier-adversarial-expansion.md
• .claude/prompts/sag-wisdom-frontier-claude-code.md

---

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

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

---

_{Comment @coderabbitai help to get the list of available commands.}

[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 2026-10-12T19:26:21.339006Z	406248a	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.

[AdaWorldAPI]

Superseded by #1523. Every commit from this branch is merged there unchanged except for the alignment edits.

Changes for this PR:

• The Proposal {…} scheduler record is renamed SchedulerDecision, because it selects work and changes nothing.
• "Any optimization is a proposal" now reads ProposedMethod.
• The missing mask in the plan and in the prompt now excludes unread prerequisites, matching your own hard-problems H1. The two had contradicted each other.

Closing in favour of the capstone.

---

Generated by Claude Code