[AdaWorldAPI]

D-ONNX-RELALG-CAP S0: convergence amendment, HOLD FOR OWNER RECONCILIATION

Status: draft, documentation only, NOT canonical S0 / NOT merge-ready.

Critical P0: three S0 competitors

#1744, #1745 and #1750 all add .claude/plans/d-onnx-relalg-cap-v1.md. Do not merge all three, overwrite another session, or silently select an owner. #1744 has 14 open points and source-backed limitations; #1750 has CC2's 19 open points and board reconciliation. Propose that one authorized S0 owner incorporate #1745's distinct Tensor / DR / Hybrid physical-competition thesis, holographic focus contract and LogicalShape/FocusWindow/PhysicalView/ResultDemand separation. Coordinate via EXO-50.

Existing implementation correction

The substrate is not lacking numerical methods and Palette256 capabilities. Existing crates/bgz-tensor, bgz17, helix and sibling AdaWorldAPI/ndarray already supply calibrated codebooks, FisherZ/DistanceLut, 256² tables, Burn CompiledAttention/CompiledLinear, SIMD tanh and ARM NEON. See #1749 for source-backed inventory and limitations. Do not build a new palette codec or general tensor AST.

Proposed architecture

ONNX keeps ModelProto/GraphProto and versioned operator semantics. Contract-owned set algebra + RelAlg (only eligible Boolean/selection subset) handle relational planning; existing ndarray/Burn numerical methods own value math. Optional witnessed global-to-local focus can select DR or Hybrid work. Tensor, DR, Hybrid are legal physical strategies, not new mathematical laws. Existing compiled palette methods are an additional approximate, model-specific candidate until exactly admitted.

Blocking safety: Burn caches attention by d_head and compiled linear by shape. Neither lookup alone proves model/layer/weight/codebook identity. Compiled attention uses registered distance/assignments, not current Q/K values. Shape-only transparent ONNX MatMul replacement is forbidden absent a new independent admission gate.

Limits: DR64³/DR256³ are proposals, not new global IDs or mandatory allocations. 1/16 quadratic and 1/64 cubic focus ratios differ. SQL Bag/K3, ADMISSION B, and frame/focus/scope authority survive. D-DRMC certified-law and independent-checker limits remain explicit.

Independently owned companion draft plans

• #1746: S1/S2 ONNX intake and typed value dispatch.
• #1747: S3a–c Cartesian/holographic/DR256³.
• #1748: S3d three-arm exact physical comparison plus optional compiled Palette quality frontier.
• #1749: S4/S5 existing BGZ/Helix/ndarray reuse, identity-safe compiled admission, ONNX export/codegen.

No executable code, no ABI or board mutation, no tests or benchmarks performed in this PR. Original 12-stage DR³ plan remains an open documentation dependency; no auto-merge.

[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: 4ff87d4a-fdd0-4149-ab5e-2757bf950a30

• 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.

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

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

[AdaWorldAPI]

P0 cross-session review: competing S0 documents, do not merge independently

External review found #1744 and this PR both add .claude/plans/d-onnx-relalg-cap-v1.md. They are not equivalent. #1744 has source-backed ONNX/RelAlg limitations, OP-ONNX-1..14, three Opus reviews, and board closeout that needs to be reconciled with merged #1742. This PR has the explicitly proposed Tensor/DR/Hybrid peer physical plans, holographic focus invariants, and LogicalShape/FocusWindow/PhysicalView/ResultDemand separation.

Recommended owner decision (NOT executed): make #1744 the candidate canonical S0; bring this PR's convergence concepts into that S0 by owner-approved amendment. Do not independently merge both S0 files, overwrite #1744, or auto-close this PR before its unique claims are retained. Coordinate the choice via EXO-50.

I've amended only our separate owned draft plan files:

• docs(onnx) S1–S2: semantic intake and minimal RelAlg polyfill plan #1746 S1–S2: typed value-math dispatch (ONNX GraphProto remains graph; no Rel::MatMul AST), STRICT_NATIVE vs proposed whole-model delegated fallback; mixed subgraph delegation still OPEN and must not invent a scheduler.
• docs(dr3) S3a–S3c: Cartesian reference, holographic focus, DR256³ validation plan #1747 S3a–c: existing R∘S ternary iteration space vs proposed DR³ interface; OP-ONNX-4 intra-register trailing_zeros enumeration vs forbidden cross-table gather; focus-closure tests.
• docs(onnx) S3d: Tensor × DR × Hybrid crossover + compiled Palette quality arm #1748 S3d: per-tile T/D/H crossover, amortized prepacking and incremental updates, optional real ARM NEON benchmark (not measured).
• docs(onnx) S4–S5: reuse BGZ/Helix/ndarray with identity-safe ONNX binding #1749 S4–S5: typed value methods, lab(drmc): certificate keeps egglog's ProofStore + ProofId, not its rendering; L1 board records #1742 proof DAG/independent-checker boundary, export portability and no general certification.

No executable implementation, ABI or shared board changes. #1744 remains untouched. Operator and current owners make the canonical S0 decision.

[AdaWorldAPI]

Independent Optimum offline frontend PR #1751: architecture note (no S0 ownership change)

Created #1751 as a separate, documentation-only plan for reusing Hugging Face Optimum-ONNX offline model export and calibration/reference workflows to build a non-destructive, identity-bound existing BGZ/Palette256 sidecar.

Boundary: #1751 O0–O3 = model preparation/export/calibration/artifact provenance; #1746 = ONNX native semantic intake and typed operator admission; #1749 = identity-safe native CompiledLinear/CompiledAttention method dispatch and numerical quality gates; #1748 = Tensor/DR/Hybrid and optional quality-gated compiled Palette crossover; #1747 DR64³ remains optional. No new AST, tensor runtime, palette codec, or hidden approximation.

This does not resolve the three-owner S0 conflict (#1744/#1745/#1750). No contested S0 file or board ledger has been modified. For CC1/CC2 review, see EXO-50.

[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 2026-10-23T22:13:12.961396Z	9271dea	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.