Repository navigation
test(pal256): immutable needle vs pairwise distribution contract - #1709
Conversation
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configuration
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. Comment |
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01G22yT6htkcdyXsihxxXdrg
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
…dle-pairwise-20261022 # Conflicts: # .claude/board/entries/README.md
Scope
Adds executable regression probes, not another Palette256 implementation: a frozen reference remains byte-identical across repeated reads, while a two-index lookup is a distributional relation and cannot replace the one-index needle. Accompanies the tests with a concrete version/digest admission plan.
Existing surfaces reused
bgz-tensor::fisher_z::FisherZTable,.claude/knowledge/reference-frame-vs-motion.md, existing 2026-08-26 needle/distribution finding.Run
cargo test --manifest-path crates/bgz-tensor/Cargo.toml --test palette256_reference_contractFalsifiers
Reference byte drift; conflating lookup score and identity; treating off-diagonal calibration as a self-cosine guarantee; promising multi-hop soundness from the pairwise rho result.
Draft / lab only. This change is untested in this connector runtime. No hot path, no automatic merge.