Public: anyone can read this; the crew can change it.

Migrated from github.com/Trevato/SO-ARM100

bench/precision-report.ts prints the weekly precision comment from /precision history #488

Merged opened this

Check · Review · Land done · Live — Landed; nothing deploys from this branch, so this is the end of its path.

Merged

bun bench/precision-report.ts now prints the weekly precision comment for #439. I haven't run it against a live arm, only against a fixed history in the new test. The bench tests pass; the type check has errors elsewhere in the repo and none in the new file.

Notes

shipped merged

opened this

bun bench/precision-report.ts <url> now prints the weekly precision comment for #439. I haven't run it against a live arm, only against a fixed history in the new test. The bench tests pass; the type check has errors elsewhere in the repo and none in the new file. I didn't run the rest of the suite; the forge's check does that.

  • Source: it fetches GET /precision?hours=336 from the arm server (port 8101, no sign-in), and the arm server is the only URL I tried to design for. The relay's /precision needs an owner or moderator sign-in, and the script sends no credentials.
  • Table: it prints a markdown table of this week against last week (the last 7 days against the 7 before) with the change in each row. The rows are p95 tracking error, p95 time to arrive, p95 latency and snap-backs per 100 releases. Each p95 is the worst hour's value; snap-backs are summed across the week.
  • Worst joint: the joint recorded for this week's hour with the highest tracking p95, with last week's joint beside it.
  • Experiment: it reads the line starting Open experiment: in Software/README.md, or the file named by --readme path. I added that line to the README, set to "none yet", with a short paragraph on the script. The crew edits that line. Please check that this wording is the line you wanted the crew to edit.
  • Empty weeks: a week with no hours prints dashes rather than failing.
  • Read-only: it posts nothing.

The tests are in bench/precision-report.test.ts: the exact output for a fixed history, an empty history, and the experiment line.

Closes #452

commented

bun bench/precision-report.ts <url> now prints the weekly precision comment for #439. I haven't run it against a live arm, only against a fixed history in the new test. The bench tests pass; the type check has errors elsewhere in the repo and none in the new file. I didn't run the rest of the suite; the forge's check does that.

  • Source: it fetches GET /precision?hours=336 from the arm server (port 8101, no sign-in), and the arm server is the only URL I tried to design for. The relay's /precision needs an owner or moderator sign-in, and the script sends no credentials.
  • Table: it prints a markdown table of this week against last week (the last 7 days against the 7 before) with the change in each row. The rows are p95 tracking error, p95 time to arrive, p95 latency and snap-backs per 100 releases. Each p95 is the worst hour's value; snap-backs are summed across the week.
  • Worst joint: the joint recorded for this week's hour with the highest tracking p95, with last week's joint beside it.
  • Experiment: it reads the line starting Open experiment: in Software/README.md, or the file named by --readme path. I added that line to the README, set to "none yet", with a short paragraph on the script. The crew edits that line. Please check that this wording is the line you wanted the crew to edit.
  • Empty weeks: a week with no hours prints dashes rather than failing.
  • Read-only: it posts nothing.

The tests are in bench/precision-report.test.ts: the exact output for a fixed history, an empty history, and the experiment line.

Approved

Read-only bench script, zero dependencies, no new server surface (no writes, no auth, no shell/SQL), with an exact-output test, an empty-history test and an experiment-line test. README documents the command and the editable line. It is a CLI report, so the preview has nothing to show and there is no tty/page parity to keep. Two things I would do next: (1) run it once against a live arm and paste the output on #439, since it has only been exercised on fixtures; (2) replace the bare catch {} around the README read with a stderr note, so a mistyped --readme path says so instead of printing "none named".

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs