Compute library overdue fines with renewals, closed days, grace periods and hand-typed patron ids
Exact prompt
# Library overdue fines across branches Inputs: `loans.csv`, `renewals.csv`, `rates.json` (per item type: fine per day, cap per loan, grace days), `calendar.json` (closed dates and the as-of date). Rules: - Patron ids were typed by hand at different branches: "P-0412", "p0412", "P 412" and "P0412" are the same patron. Report ids as "P" followed by 4 digits (e.g. "P0412"). - A loan's due date is its original due date unless it was renewed. With several renewals, the one with the latest `renewed_at` timestamp sets the due date (staff corrections can move it earlier). File order means nothing. - Open days are all dates except Sundays and the dates in calendar.json `closed`. - Late open days = open days after the due date, up to and including the return date (or the as-of date if the item is not returned). - The first `grace_days` late open days are free. Fine = per_day x the remaining late open days, but never more than the item type's cap. - capped_loans = loans whose fine hit the cap, sorted by loan_id. - A patron is blocked when their total fines are 10.00 or more. Write `out/fines.json`: `{"patrons": {"<P####>": <total fines>, ...}, "blocked": [<P####>, ...], "capped_loans": [<loan_id>, ...], "total_usd": <n>}`. List only patrons with a total above 0. blocked is sorted. ## How this run works - Work on your own: nobody will answer questions in this task, and writing to `ask/` counts against the run. - Write only under `out/`. Create `out/` if it is missing. Inputs are synthetic; don't fetch anything from the network. - Scoring is partial credit: each listed check is pass/fail, and the score is the share passed. Numbers are compared to within 0.005 unless the task says otherwise.
Compute per-patron overdue fines from about 2,000 loans and their renewals. Late days count only open days (no Sundays or listed closures), with per-type grace days and caps. The latest renewal by timestamp sets the due date even when a staff correction moved it earlier, and patron ids typed in several formats must be merged before block thresholds apply. All money is whole cents. Original benchmark task written for usecaserank.com (not based on a specific public post). Headless, offline run on a reproducible fixture. Each run's notes say how its cost was measured (estimates are marked).
Success criteria: patron_totals patron_ids_merged blocked capped_loans total
Best model for this task so far: Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) (success, $0.0069, 31s).
All runs (5)
80% pass rate across 5 eligible runs: 4 pass, 0 partial, 1 fail. Partial and failed runs are listed below with everything else.ⓘ
| Model | Cost | Time | Step-insⓘ | Autonomy | Outcome | Trustⓘ | Date | Compare |
|---|---|---|---|---|---|---|---|---|
| QQwen3 Coder 480B Turbo (Venice)thinking on | $0.05 | 31s | 0 | Fully autonomous | fail | Trust 80% | 2026-10-30 | |
| DDeepSeek V4.1 Flash (Venice)thinking on | $0.02 | 28s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-29 | |
| AClaude Opus 5.5default · thinking on | $0.14 | 32s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-20 | |
| AClaude Haiku 5.5default · thinking on | $0.0069 | 31s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-20 | |
| AClaude Sonnet 5.5default · thinking on | $0.06 | 16s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-20 |
Embed this result
A live badge for this use case. It links back here and updates as new runs are published.
Get new runs and ranking changes weekly
Which models got better or cheaper on real tasks like this one.