eval-triage v2 is out.
It still does the core thing: reads Inspect AI .eval logs and ranks samples by monitor suspicion score so you stop manually hunting for flagged trajectories.
What’s new:
> --compare for side-by-side run comparison. Drop two logs and see which samples scored higher, which ones flipped, and where the monitors disagreed.
> --config for config files. Set your usual filters, exports, and thresholds once instead of repeating flags every time.
Still the same workflow otherwise. Rank by suspicion, filter with --top / --min-score / --only-unparsed, and export to JSON, CSV, or markdown.
If you’re iterating on agent evals or control experiments in Inspect and ControlArena, the comparison mode is the useful addition. One command instead of opening multiple logs and grepping.
https://t.co/c1BezRFErE
the market doesn’t wait, neither should you.
a coin spikes at 3am and you’re asleep, you wake to a drop. you see a lot of theses that are hard to check. you set a stop-loss in your head and forget it by lunch.
been looking at the @MerrymenAI agent for this. it can run 24/7 while you’re not watching, and the limits you set stay active even when you’re offline. it only trades inside those rules, and rejected trades stay visible so you can still see what it tried to do.
create your Merrymen AI agent today👇 🚀
@ms_ShogunR Hi , hope you’re doing well! Just wanted to follow up on the wallet address I sent last week regarding the events that happened. I understand you might be busy, so I just thought to check in today .I just hope that you do still have me in mind thank you i really appreciate it 🙏