Settle a property manager's security deposits on move-out with interest, final-month proration, allowed deductions and a statement deadline
Exact prompt
# Settle a property manager's security deposits on move-out with interest, final-month proration, allowed deductions and a statement deadline Inputs: `tenancies.csv` (tenancy_id, unit, move_in, move_out, monthly_rent, deposit, final_month_paid, statement_sent), `charges.csv` (charge_id, tenancy_id, type, item, amount), `move-in-report.csv` (unit, item: conditions already recorded when the tenant moved in) and `deposit-rules.json` (interest_rate_pct, cleaning_cap, statement_days). All tenants and units are synthetic; amounts are in dollars. These rules are the record; where `leasing-note.txt` disagrees, the rules win. Rules: - Interest: deposit × interest_rate_pct / 100 × whole months held / 12, rounded half-up to the cent. Whole months held = the number of monthly anniversaries of move_in (same day number in a later month) that fall on or before move_out. - Final month: the rent owed for the move-out month is monthly_rent × (move_out's day number) / (number of days in that month), rounded half-up to the cent; final_rent_due = that amount − final_month_paid (negative means the tenant overpaid and is owed the difference). - Deductions: a charge of type wear is never deducted. A damage charge is deducted only if its item is not listed for the same unit in move-in-report.csv. Cleaning charges are deducted up to cleaning_cap in total per tenancy. unpaid_rent charges are always deducted. - Late statement: if statement_sent is more than statement_days days after move_out, nothing from charges.csv may be deducted for that tenancy (deductions = 0); final_rent_due still applies. - Balance: refund = deposit + interest − deductions − final_rent_due. If that is negative, report refund 0 and balance_due = the shortfall; otherwise balance_due is 0. - Write `out/deposits.json`: `{"tenancies": {"<tenancy_id>": {"interest": <dollars>, "final_rent_due": <dollars>, "deductions": <dollars>, "refund": <dollars>, "balance_due": <dollars>}, ...}, "total_refunded": <sum of refund>}`. ## 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.
Settle the security deposits of synthetic tenants who moved out of a managed building. Interest accrues by whole monthly anniversaries, the final month's rent is prorated by the move-out day, normal wear and conditions on the move-in report are never deducted, cleaning is capped per tenancy, and a statement sent after the deadline voids every deduction. Statements land exactly on and one day past the deadline. A leasing note gets the proration, interest, wear and deadline rules wrong. 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: interest final_month deductions late_statement balance
Best model for this task so far: Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) (success, $0.0047, 17s).
All runs (3)
100% pass rate across 3 eligible runs: 3 pass, 0 partial, 0 fail. Partial and failed runs are listed below with everything else.ⓘ
| Model | Cost | Time | Step-insⓘ | Autonomy | Outcome | Trustⓘ | Date | Compare |
|---|---|---|---|---|---|---|---|---|
| AClaude Opus 5.5default · thinking on | $0.15 | 30s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-24 | |
| AClaude Haiku 5.5default · thinking on | $0.0047 | 17s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-24 | |
| AClaude Sonnet 5.5default · thinking on | $0.06 | 13s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-24 |
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.