Overview
Tool failure rate
≥ 2.0 %
1 of 50 results
Unreported tool outcomes are counted per session only, so the fleet figure is a lower bound. 1 calls never returned a result and are in neither figure.
Spend
≥ $3.14
16 responses carried no cost.
Worst stall
43.3 s
Blocked
≥ 5
Lower bound — 9 of 12 sessions predate this measurement and are not included.
What enforcement decided
- 5 Blocked refused before the call ran
- 3 Allowed no rule denied it
- 0 Redacted ran with arguments rewritten
- 0 Held for approval not an allow and not a block
- 0 Unclassified recorded, but carried no readable verdict
Worst repeat run ≥ 0 calls — back-to-back identical calls in one session. 4 sessions could not be read for this and are excluded.
12 sessions in this window. 3 excluded from the worst-stall and repeat figures because their events did not arrive in execution order. 1 excluded from the repeat figure because a tool call had no measurable identity. Both are normal for push sources.
Which tools fail most?
| Tool | Calls | Failure rate | Unanswered | Max duration |
|---|---|---|---|---|
| Bash | 22 | 1 | 960 ms | |
| Read | 21 | 0 | 245 ms | |
| Grep | 3 | 0 | 16 ms | |
| Glob | 3 | 0 | 210 ms | |
| Write | 2 | 0 | 8 ms |
Where is the spend going?
| Model | Requests | Cost | No cost reported | Input (uncached) | Cache reads | Cache writes | Output |
|---|---|---|---|---|---|---|---|
| claude-opus-5 | 52 | 16 | 136 | 2,042,222 | 201,973 | 33,316 |
What is going wrong?
No errors. 12 sessions checked.