Compare runs
B is compared with the previous run of the same agent.
10 steps · 12.4s · ≈$0.048
9 steps · 12.2s · ≈$0.044
steps 10 → 9 · cost −8% · duration −2% · outcome succeeded
What changed between them
Same app version, agent version, prompt version and model.
Step by step
= same (4)≠ changed (4)+ only in B (0)− only in A (1)| A · run_f929f0f0 | B · run_c9f2a6d4 | |
|---|---|---|
| = | #2 claude-sonnet-4-5 | #2 claude-sonnet-4-5 |
| ≠ | #3 sentry.getIssue | #3 sentry.getIssueargs changed |
| = | #4 claude-sonnet-4-5 | #4 claude-sonnet-4-5 |
| ≠ | #5 gitlab.searchCode | #5 gitlab.searchCodeargs changed |
| ≠ | #6 gitlab.readFile | #6 gitlab.readFileargs changed |
| = | #7 claude-sonnet-4-5 | #7 claude-sonnet-4-5 |
| ≠ | #8 jira.createIssue429 Too Many Requests | #8 jira.createIssueargs, outcome changed |
| − | #9 jira.createIssue | — |
| = | #10 slack.postMessage | #9 slack.postMessage |
System spans are left out; steps are aligned on the tool called (or LLM call), then compared on arguments and outcome.