[svanlabs]

What generated this?

Generated-by: claude-code/claude-sonnet-5-5

Motivation

Closes #971.

The newest-hands loop of the history download drops a failed page without a log line: fetch_page returns a bare None on any 5xx and the loop breaks (crates/apps/bot/src/history/download.rs). A read-only probe on 2026-10-12 (GET /me/hand-history?limit=1, one per key, approved by the operator) gave HTTP 200 for SvanBotV10 and HTTP 500 after about 11 s for the other four keys. Their history stores had stopped growing one after another over eight days (newest row: Svanar 10-04 23:40, 13,147 live hands behind; SuraGunnar 10-10 18:59; Svanism 10-11 05:10; SurSvan 10-11 21:24) and the log held only the hourly season refresh.

Solution

One change in crates/apps/bot/src/history/download.rs:

• fetch_page returns Result<Value, String>: Err("HTTP 500") for a 5xx, the status after the 429 and other retries, a 200 body that is not JSON, or the last transport error. The key travels only in the request header, never in these strings.
• The newest-hands loop logs the reason as a warning through note_newest_failure, at most once an hour per bot (the loop runs hourly once every bot is done, every minute while one is not, so the limiter only matters for the one-minute case). A page that comes back clears the bot's hour.
• The backfill call site ignores the reason as before; it already names a repeated failure.

Tests: a_server_error_page_comes_back_with_its_status (a local listener answering 500) and a_failed_newest_page_is_reported_once_an_hour_per_bot. The new API does not exist on origin/main, so before the change they cannot compile; to show they assert behaviour, I disabled the fix (no log line, empty reason) and both failed (left 0, right 1; left "", right "HTTP 500"), then restored it. The stage-1 call site that connects the two is covered by reading, not by a test.

Log-only change off the decision path: no paired simulation, no golden change. The cause of the 500 is on the venue's side and is not addressed here. Not released: the installed build is 0e3d584.

Checklist

• scripts/check.sh full passes on this exact commit (0d96c89), run in a worktree on the shared warm target/dev:

test.py: 899 passed, 0 failed, 4 ignored in 35 binaries; build 0.3s, run 4.5s (6 jobs x 2 threads, profile gate, peak test RSS 195 MB)
== ok (full, 178 s)
check.sh full: exit 0

• This is one change.
• Not a behaviour change on the decision path: no paired simulation; the golden snapshot is untouched.
• Not a performance change.
• No document names the changed behaviour.
• The AI-provenance rule was followed: Generated-by: is in the commit footer and in the section above.

🤖 Generated with Claude Code