Show combined averages with clear test hardware #9
Loading…
Reference in a new issue
No description provided.
Delete branch "feat/combined-dataset-averages"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Dataset-specific rankings and repeated runs made the benchmark site harder to scan. Each suite now has one combined ranking per engine, averaging only the newest run per model and dataset. Older measurements remain archived; missing metrics are excluded rather than counted as zero.
The dataset selector, best-run toggle and run-count labels are removed. Hardware labels identify the CPU, shared-memory iGPU and RTX 5050 with 8 GB VRAM without machine names. The seven RTX 4070 Ti runs are excluded from pages, downloads and public evidence routes, while their original records remain unchanged. Source runs and metric coverage remain available in average details and exports. Coverage, protocols and precision can differ between models; the averages are not a pooled evaluation.
Proven with 60 Bun tests, type checks, a production build verifying 355 published results and excluded evidence, plus desktop/mobile browser checks of every populated engine, newest-run selection, computed means, exports and hardware labels.
kReview review
Verdict: no findings
No specific defects were verified in the reviewed changes.
kReview checked only the PR diff, metadata, and supplied repository context at
b64592fc0c. It reports only issues supported by this material.Estimated cost: ~$0.10 total (164.6k in / 1.2k out); this run ~$0.05 (83.6k in / 326 out) / gpt-6.1-terra.
Run time: 1:55 total; this run 0:12.