[CodexVeritas]

Summary

Recurring failures in the run-bot-aib-tournament.yaml workflow over the last few main-site forecasting windows (Oct 30th and Nov 2nd 2026):

• Skip 44553 and 44882. Bots conclude both questions have already resolved (44553: research reports a Sept 2026 Nigeria NBS inflation figure below 15%; 44882: its Sept 24th 2026 deadline has passed but the question is still open). The resolved-question parser guard then refuses the forecast, so each affected bot fails and retries every run in the window. 44553 alone failed for 42 bots on Nov 2nd.
• Don't fail the workflow on 4922 and 41139. z-ai GLM models return None content on these China-related questions, and Moonshot Kimi models get content-filter 400s on 4922. Bots still forecast them; errors on them just no longer fail the job.
• Raise the Sonnet 5 bot timeout to 160s (from the 60s default for Claude models), matching the other Claude bots. Its long main-site responses timed out on all 5 predictions for some questions in both windows.

Testing

• code_tests/unit_tests/test_run_bots.py passes.
• Checked that the bot dict builds and the Sonnet 5 LLM gets timeout=160.

🤖 Generated with Claude Code