← Back to the recipe
Run data
Does 'think hard' actually help? Same pass rate, but it tested its own work
Claude Code 2.1.302 · 23 Oct 2026 · 6 runs · costs as Claude Code reports them. Raw summary (JSON)
Prompt
Implement billingDates() in src/schedule.js. The spec is in the comment above it.
Runs
| Run | Model | Cost | Turns | Time | npm test passes | month-end dates: clamps, leap years, intervals (hidden test) | count, until and skip together (hidden test) | rejects bad input with RangeError (hidden test) | same dates in any server time zone (hidden test, Chicago and Auckland) | added tests for billingDates | ran tests after last edit | Diff |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Plain #1 | sonnet-5-5 | $0.057 | 7 | 23s | pass | pass | pass | pass | pass | fail | fail | diff |
| Plain #2 | sonnet-5-5 | $0.053 | 6 | 20s | pass | pass | pass | pass | pass | fail | fail | diff |
| Plain #3 | sonnet-5-5 | $0.060 | 8 | 22s | pass | pass | pass | pass | pass | fail | fail | diff |
| Think hard #1 | sonnet-5-5 | $0.093 | 10 | 33s | pass | pass | pass | pass | pass | pass | pass | diff |
| Think hard #2 | sonnet-5-5 | $0.081 | 10 | 34s | pass | pass | pass | pass | pass | pass | pass | diff |
| Think hard #3 | sonnet-5-5 | $0.099 | 10 | 35s | pass | pass | pass | pass | pass | pass | pass | diff |