Alan-Jowett / LexonGraph / 27390011587

82%

push

github

Add section-4 benchmark suite assets and runner (#93)

* Add section-4 benchmark suite for streaming clustering evaluator

Implement the approved leaf-stage readiness work for lexongraph-streaming-clustering-evaluator by adding a section-4 suite layer that can:
- generate zip-backed benchmark corpus assets and benchmark profiles
- compute deterministic exact-neighbor ground truth
- harvest deterministic real-world corpora from block-store sources
- run comparative screening across candidates
- emit suite-level reports with scale-tier and build-time-per-vector data

Also add CLI entry points for suite asset generation and suite execution, plus validation coverage for suite generation, harvesting, zip-direct consumption, screening, reporting, and CLI execution.

Traceability:
- REQ-STREAM-EVAL-031..037
- DSG-STREAM-EVAL-020..026
- VAL-STREAM-EVAL-025..031

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

* Add checked-in synthetic section-4 suite assets

Add a repository-owned synthetic benchmark suite under the streaming clustering evaluator crate, including the suite spec, generated manifest, generated benchmark profiles, and zip-backed synthetic corpora.

Also add regression coverage that loads the checked-in manifest, verifies the committed assets exist, and executes the suite from those assets.

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

* Normalize section-4 asset paths across platforms

Normalize Windows-style relative paths in checked-in section-4 manifests and block-store corpus references so synthetic suite assets run correctly on Linux CI as well as Windows.

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

* Address PR review findings for section-4 assets

Fix substantive review findings by validating section-4 profile ids, emitting portable forward-slash paths in checked-in suite assets, validating harvested embeddings with the shared embedding validator, and rejecting inv... (continued)

791 of 1017 new or added lines in 3 files covered. (77.78%)

10383 of 12480 relevant lines covered (83.2%)

1066.91 hits per line