[status-page] Flag ferry runs slower than prior 7 successful runs#4788
[status-page] Flag ferry runs slower than prior 7 successful runs#4788ravwojdyla merged 3 commits intomainfrom
Conversation
Overlay an amber "!" at the top-right of any history tile whose wall time exceeds mean + 1σ of the preceding 7 successful runs. Baseline restricts to successes so failures/cancels/timeouts (early exits, hangs) don't poison the mean. Hovering any tile opens a small CSS tooltip with the run's sha, status, wall time, and — when flagged — the expected threshold. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e60bf522f3
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| const variance = | ||
| priorDurations.reduce((s, x) => s + (x - mean) ** 2, 0) / priorDurations.length; | ||
| const stddev = Math.sqrt(variance); | ||
| if (stddev === 0) return null; |
There was a problem hiding this comment.
When the prior 7 successful runs all have the same durationSeconds, stddev becomes 0 and this function returns null, which means isSlowRun will never flag the current run regardless of how much slower it is. That breaks the documented rule of flagging runs above mean+1σ, because in this case σ=0 should still yield a valid threshold at the mean (for example, seven 300s runs followed by a 600s run should be flagged).
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
🤖 Fixed in 54fe9a8 — removed the σ=0 short-circuit so the threshold collapses to the mean when all prior successful runs are identical. A strictly slower run now fires as expected.
Slow-run baseline was a hard requirement of 7 prior successful runs, so sparse or newer workflows never flagged anything. Take up to 7 but require at least 3 so the signal kicks in earlier. Tooltip now shows the actual sample size used. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
If all prior successful runs share the same duration, σ=0 and the old guard returned null, so a genuinely slower run never fired. Drop the short-circuit: threshold collapses to the mean, and any run strictly slower than a perfectly stable baseline is anomalous. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2 participants
Overlay an amber "!" at the top-right of any ferry history tile whose wall time exceeds mean + 1σ of the preceding successful runs. Baseline uses up to 7 samples and requires at least 3 so the signal kicks in on newer/sparser workflows; restricts to successes so failures/cancels/timeouts don't poison the mean. When σ=0 the threshold collapses to the mean. Hovering any tile opens a small CSS tooltip with the sha, status, wall time, and — when flagged — the expected threshold and sample size used.