Alan-Jowett / LexonGraph / 27358638606

81%

push

github

Implement streaming clustering evaluator crate (#84)

* Implement streaming clustering evaluator crate

Traceability:
- Requirements baseline: docs/specs/rust-streaming-clustering-evaluator-crate/requirements.md
- Design baseline: docs/specs/rust-streaming-clustering-evaluator-crate/design.md
- Validation baseline: docs/specs/rust-streaming-clustering-evaluator-crate/validation.md

Summary:
- add the new lexongraph-streaming-clustering-evaluator crate to the workspace
- implement the evaluator-owned benchmark profile, candidate adapter, campaign runner, scoring, reporting, and CLI surfaces
- add executable spec-validation coverage for comparative execution, determinism, leaf membership scoring, failure taxonomy, and deferred-goal reporting

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address PR review feedback

- validate probe and leaf-membership cluster ids against K
- preserve profile software identity in provenance output
- add regression coverage for out-of-range cluster ids
- fix workspace member formatting and SPDX header coverage

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address additional PR review feedback

- sanitize candidate-derived artifact file names
- tighten benchmark profile validation for corpus IDs, entity corpus references, ranking weights, and gate minima
- preserve original states in invalid-transition fixture errors
- add regression coverage for the new validation and artifact-safety cases

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Handle profile-size overflow in evaluator validation

- use checked_mul for leaf_size * declared_final_cluster_count
- return a deterministic InvalidConfiguration error on overflow

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Stabilize evaluator regression test selection

- select the target MetricAtLeast gate by gate_id instead of fixture index
- avoid brittle coupling between the reg... (continued)

1038 of 1276 new or added lines in 2 files covered. (81.35%)

8297 of 9880 relevant lines covered (83.98%)

208.48 hits per line