[AdaWorldAPI]

What

This PR implements two operator rulings from 2026-10-30.

1. u8→i8, only through an explicit intent parameter.

• ReinterpretBits is a zero-sized marker type.
• U8x64::reinterpret_as_i8(ReinterpretBits) and I8x64::reinterpret_as_u8(ReinterpretBits) live in src/simd.rs and work on every backend.
• The operation reinterprets bits and converts nothing: 0x80 becomes -128 and 0xFF becomes -1, with no saturation.
• FACET STRIDE rule 3 in CLAUDE.md now names this as the single allowed crossing; a plain as i8 stays forbidden.
• This closes the facade gap that U6 recorded (the head-to-head probe's bits_as_i8 store/load stand-in).

2. Codec9 runtime scale and batch encoder, so the same codewords serve exact LUT[a,b] work and learning.

• Codec9Runtime<R> takes the scale as a value, checked once (it must be positive and finite).
• It calls the same encoder and decoder as Codec9<R, C>, so equal scales produce identical codewords.
• Rounding stays a type parameter limited to nearest rounding. Storage, polarity, Sym and NaN handling stay fixed.
• encode_touched_batch and decode_batch read and write the magnitude lane plus the presence plane:
  • A non-finite value is refused before anything is written.
  • Presence bits past the last lane are cleared.
  • A never-touched lane decodes to NaN.
• The DR512 law doc records which form to use where:
  • Compile-time Codec9 for codeword tables.
  • Codec9Runtime for learned scales.
  • A table built under one scale must be rebuilt when the scale changes.
• The batch loops are scalar for now. They have not been measured.

Tests

• New tests:
  • 8 in hpc::codec9, 12 in that module in total. One pairs Codec9Runtime against Codec9 over a 6,000-value sweep and all 512 codewords.
  • 4 in simd::reinterpret_bits_parity, one of which feeds 0x80 into dot4_u8_i8_add and expects −512 per lane.
• Disable runs, each turns its test red:
  • the runtime encoder ignores the scale;
  • tail presence bits are not cleared;
  • the batch writes before the finiteness check;
  • the reinterpret saturates;
  • the scale check is removed.
• Suites:
  • The lib suite passes: 2,676 tests.
  • The reinterpret tests pass on the AVX2 tier (config-v3).
  • The aarch64 lib build passes.
  • wasm-parity.sh and neon-parity.sh pass.
• Checks: clippy -D warnings and fmt are clean, and the facet-stride guard reports nothing on the diff.

---

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: d0f36181-3d93-4d56-98ca-b76b97ce7eda

• 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

_{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 2026-10-30T02:08:37.140625Z	7e6dd25	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.