project devlogs / benchdb
BenchDB Spot performance regressions with benchmark histories across commits and machines, fleet trends, and history-aware CI reports
GitHub ↗ Website ↗
By week By project Choose project BenchDB ↓ AgentsView msgvault roborev kata Kenn Forge docbank Ghosthub kwt BenchDB Fotobank Source code measured at 14 snapshots, from 57,507 lines on 2026-08-27 to 99,180 lines on 2026-11-17. Full measurements are in the table below. 0 49,590 99,180 2026-08-27 2026-11-17 Source code measured at 14 snapshots, from 57,507 lines on 2026-08-27 to 99,180 lines on 2026-11-17. Full measurements are in the table below. 0 49.6K 99.2K 2026-08-27 2026-11-17 +41,673 lines since 2026-08-27
Measurements & method SCC code lines (nonblank, noncomment) in tracked source files, including tests, SQL, HTML, and CSS. Excludes prose, shell scripts, data/config files, vendored dependencies, build output, generated and minified files. Counts the checked-out tree, not accumulated additions.
Snapshots follow the repository’s default branch. Lines measure codebase size, not work completed. Historical counts can fall when code is removed.
Measured with scc 3.7.0. Updated 2026-11-17.
Source line measurements for BenchDB Date Source lines Files Commit 2026-11-17 99,180 510 a1a9cac 2026-11-15 99,180 510 a1a9cac 2026-11-08 98,733 504 1118d92 2026-11-01 97,224 502 63a6e7c 2026-10-25 81,020 438 a2df008 2026-10-18 79,907 433 4af184a 2026-10-11 65,466 345 d6aa11a 2026-10-04 63,581 325 50e07b8 2026-09-27 61,686 307 c40b7ff 2026-09-20 61,474 305 e2965b0 2026-09-13 63,637 303 5828e15 2026-09-06 62,118 290 beb85a7 2026-08-30 61,738 290 8662f51 2026-08-27 57,507 274 e571d74
Project timeline Newest first 117 merged PRs · 0 releases · 11 weeks in the devlog
2026-11-02 to 2026-11-08
6 PRs merged (5 Kenn, 1 community) Weekly diff: +2,082 / −57 lines
Charts can now fix a series baseline from a selected stable range. Benchmarks can now use x as the unit for dimensionless ratios. 2026-10-26 to 2026-11-01
33 PRs merged (33 Kenn, 0 community) Weekly diff: +23,458 / −2,419 lines
BenchDB can now index, verify, and store benchmark archives in S3 or Google Cloud Storage. Managed archives can now be transferred and audited, with startup coordination to prevent competing storage jobs. Reevaluate saved benchmark reports when new history arrives, preserving their original snapshots and links. History changes recheck only affected pull request runs. Baseline searches now use less CPU and avoid loading benchmark samples until they are needed. Remote archive reports now show read costs, network traffic, and storage reclamation status. 2026-10-19 to 2026-10-25
8 PRs merged (8 Kenn, 0 community) Weekly diff: +2,121 / −584 lines
By default, benchmark alerts now require regressions in consecutive runs, reducing alerts for one-off slowdowns. Submitted tolerances now determine benchmark statuses, and series filters now respect those statuses. Archives now work under directory names that contain an equals sign. 2026-10-12 to 2026-10-18
27 PRs merged (27 Kenn, 0 community) Weekly diff: +20,668 / −2,731 lines
Filter benchmarks by status and see how far each one moved, with clearer charts and sortable rows on narrow screens. Benchmark change flags now require a move greater than 2% that holds for two results, with at least ten earlier results for comparison. Alerts check every finalized run and say what regressed, where, and by how much. See every machine in Benchmarks and share links that preserve your sort and axis settings. The dashboard has better support for screen readers and low vision, and the Runs page works better on phones. 2026-10-05 to 2026-10-11
17 PRs merged (17 Kenn, 0 community) Weekly diff: +7,385 / −4,093 lines
Run pages now show CI reports with regressions listed first. Filter runs by whether they need attention, and scan the updated Runs page more easily. Scope the dashboard to one project, see regressions first, and check the deployed build's commit. Benchmark times now use the clearest unit: s, ms, µs, or ns. Compare now opens with the benchmark list visible and hides empty result fields. 2026-09-28 to 2026-10-04
12 PRs merged (11 Kenn, 1 community) Weekly diff: +3,152 / −293 lines
BenchDB now saves completed benchmark reports, making repeat reads less expensive. You can now deploy BenchDB in a standard container using published release images. Configurable timing and memory tolerances can hide benchmark changes too small to matter. Saved run reports now honor the selected baseline, and individual results can opt out of statistical history. Benchmark charts are easier to find and share, and chart cards now load the complete result history. 2026-09-21 to 2026-09-27
2 PRs merged (2 Kenn, 0 community) Weekly diff: +392 / −123 lines
You can now serve BenchDB under a configured URL path. 2026-09-14 to 2026-09-20
1 PR merged (1 Kenn, 0 community) Weekly diff: +8,244 / −10,696 lines
BenchDB now passes huma-check's API contract checks. 2026-09-07 to 2026-09-13
3 PRs merged (3 Kenn, 0 community) Weekly diff: +3,001 / −252 lines
Search commits and browse older benchmark runs. Open diagnostic profiles from benchmark results and comparisons. Comparison charts now show both selected results. 2026-08-31 to 2026-09-06
3 PRs merged (3 Kenn, 0 community) Weekly diff: +564 / −95 lines
Regression detection now uses a two-sigma threshold by default. CI reports now clearly flag incomplete coverage. Baseline comparisons now prefer complete CI runs. 2026-08-24 to 2026-08-30
5 PRs merged (5 Kenn, 0 community) Weekly diff: +85,129 / −2,256 lines
Benchmark results now include fleet context, making it easier to compare a machine with the wider fleet. Benchmark machines now have dedicated places in the UI. New documentation explains when BenchDB is the right tool and adds a product guide and reference site.