[AdaWorldAPI]

This revises the D-TFLP capstone (#1785). It makes no code, kernel or ABI change.

What changed

• GPT's six docs(D-TFLP-0): TFLite models on the native polyfill — int8 codes as Palette256 #1785 corrections are folded in:
  • Exactness is measured against pinned TFLM. A Rust port is not its own oracle.
  • Mish is fused only when the subgraph has been validated.
  • Conv padding equals the input zero point, and a non-zero per-tensor weights_offset is refused.
  • Softmax is a table plus a per-row algorithm.
  • TABLE is a binding-time choice that must be measured.
  • Parity is a claim about a pinned kernel and build.
• D-TFLP-4 is rescoped from the probes(D-TFLP 6.5): softmax fixture replay + certified preparation gate #1800 lab evidence:
  • The source GetReciprocal is the default: 13 of 14 known calibrations and all 8 held-out ones require it.
  • The exact reciprocal is allowed only behind a PROVED receipt.
  • The probes(D-TFLP 6.5): softmax fixture replay + certified preparation gate #1800 replay fixtures become the CI oracle, so CI needs no TFLM build.
• Prepared-object identity is split into two keys (from probes(D-TFLP 6.5): softmax fixture replay + certified preparation gate #1800 and probes(D-TFLP 6.5): Codec8 semantic admission + DT64/Quack/D-NAM handoff #1801):
  • a numerical cache key, (mult, shift), for sharing tables;
  • a semantic receipt, which decides refusal.
• D-TFLP-2 measures a 65,536-entry table against fixed-point SIMD and keeps whichever is faster, per target. This merges v1 open point 5, which the DQM reuse plan also cites.
• D-TFLP-5a (new) is a TFLite intake bake-off following the docs(D-ONNX-INTAKE-0): Burn onnx-ir vs Tract cold frontend capstone, DT64/D-NAM gates #1802 pattern.
• New section on boundaries with DT64-5 (docs(DT64-5): native DTO behavior and ONNX/time-series lowering gates #1798), DR256/DQM (docs(quack): audit DuckDB vector techniques against DR landing zones #1787–D-DR256-WHERE: capstone, DR64³ inventory, P2 Quack-path lab, evidence reconciliation, D-NFR-1 ownership correction #1792) and ndarray SMB tenant schema reference #376/plan(causaledge64-mailbox-rename-soa-v1): composition spec + 10 epiph… #377.
• v1 is marked as revised where v2 corrects it. Updated: STATUS_BOARD, INTEGRATION_PLANS, a new board entry and the regenerated entries index.

Open

• Who issues the generation identity (EXO-152).
• The binding store (D-NFR S5).
• Where Receipt lives in the contracts.
• Accumulator wrapping.
• Full-runtime parity.
• Target models.
• Validating the Mish subgraph.
• Five TFLM kernels that have not been read yet.

---

Generated by Claude Code

[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: 6b537f9f-a677-4571-9b90-1ed39c8ee305

• 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.}

[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 2026-10-26T21:15:20.704596Z	f57c35d	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.