Crate cp_validator 

Validator protocol for empirical quality verification (CP-015).

Any Canon node can opt into validation. Validators download content from Arweave, index it locally, run test queries against it, measure retrieval quality, and publish OpenSkill Bayesian ratings for content contributors. These ratings help other nodes prioritize what to download from Arweave.

Validators also test live search peers by connecting over Tor, sending test queries, and rating peer quality for circuit selection.

The approach adapts Templar/Covenant (tplr.ai), which uses empirical quality verification for decentralized pre-training. OpenSkill Bayesian ratings provide a robust, game-resistant ranking that converges even with sparse observations. Window-based coordination ensures validators test the same content without requiring blockchain synchronization.

Data flow:

1. Load test query corpus (from Arweave or local file)
2. Compute current validation window
3. Select test queries deterministically from corpus
4. For each contributor: download from Arweave, index, run queries, measure metrics
5. Rank contributors by composite score, update OpenSkill ratings pairwise
6. For each peer: connect over Tor, send queries, measure quality + latency
7. Rank peers by composite score, update OpenSkill ratings pairwise
8. Publish results to Arweave

Re-exports

pub use corpus::TestCorpus;

pub use corpus::TestQuery;

pub use evaluation::composite_score;

pub use evaluation::evaluate_contributor;

pub use evaluation::evaluate_local_graph;

pub use evaluation::evaluate_peer;

pub use evaluation::mrr;

pub use evaluation::ndcg_at_k;

pub use evaluation::peer_composite_score;

pub use evaluation::precision_at_k;

pub use evaluation::update_contributor_ratings;

pub use evaluation::update_peer_ratings;

pub use evaluation::EvaluationResult;

pub use evaluation::PeerQualityMetrics;

pub use evaluation::PeerTestResult;

pub use evaluation::QualityMetrics;

pub use rating::pairwise_update;

pub use rating::Rating;

pub use rating::DEFAULT_MU;

pub use rating::DEFAULT_SIGMA;

pub use window::select_test_queries;

pub use window::ValidationWindow;

Modules

corpus

Test query corpus for validator evaluations.

evaluation

Evaluation pipeline for content contributors and live peers.

rating

Bayesian rating implementation using skillratings’ Weng-Lin model.

window

Window-based coordination for validators.

Structs

ContributorRating

OpenSkill Bayesian rating for a content contributor.

PeerRating

OpenSkill Bayesian rating for a live search peer.

Validator

The main validator that runs evaluation loops.

ValidatorConfig

Configuration for the validator.

Enums

ValidatorError

Errors specific to the validator protocol.

Functions

prioritize_downloads

Prioritize Arweave downloads by contributor rating.