An LLM agent that triages post-silicon debug logs: it extracts a failure signature, retrieves similar prior failures, checks known escalation patterns, and commits to a bin with a calibrated confidence. Every tool call it makes is shown below as it runs.
Pick a failure above and run the triage.
Step 2 of the agent is search_prior_failures — BM25-style retrieval over these
records. It scores each prior by how many query terms it shares, weighting rare terms more
heavily than common ones, then reports how strong the best match is. That strength is what
caps the final confidence. Type below to run the real retrieval function; no model is
involved, so the same input always gives the same scores.
| ID | Bin | Title | Step | Status |
|---|