Audit a household bank statement for forgotten subscriptions, double charges and true delivery spend
Exact prompt
# Audit a household bank statement for forgotten subscriptions, double charges and true delivery spend Inputs: `statement.csv` (txn_id, posted, description, amount), `merchants.csv` (merchant_id, match_text, category), `categories.csv` (category) and `last-used.csv` (merchant_id, last_used). The statement covers three whole calendar months. All people, merchants and transactions are synthetic; amounts are in dollars, a negative amount is money out (a debit) and a positive amount is money in (a credit). These rules are the household's; where `friend-tip.txt` disagrees, the rules win. Rules: - Matching: a transaction belongs to the merchant whose match_text appears in its description, ignoring letter case; at most one match_text appears in any description. A credit that matches no merchant (pay, transfers in) is ignored; every debit matches a merchant. - Forgotten subscriptions: a merchant in the 'subscriptions' category is forgotten when it has at least one debit in each of the three statement months and its last_used is blank or more than 60 days before the statement's last day; its annual cost is 12 × the amount of its most recent debit (as a positive number). - Double charges: take each merchant's debits in posted order (then txn_id order). A debit is a double charge when the debit just before it from the same merchant has exactly the same amount and was posted on the same day or the day before, unless a credit from that merchant of the same size is posted within 7 days after the double charge (on or before its posted date + 7). List the later debit of each pair. - Delivery spend: for each statement month, the debits from merchants in the 'delivery' category posted that month (orders, delivery fees, service fees and tips are separate lines and all count) minus the credits from those merchants posted that month. - Category spend: for each category in categories.csv, the category's debits minus its credits over the whole statement, as a positive dollar figure (double charges are still spend unless refunded). - Write `out/audit.json`: `{"forgotten_subscriptions": [<sorted merchant_ids>], "forgotten_annual_cost": <sum of their annual costs>, "double_charges": [<sorted txn_ids>], "double_charged_total": <sum of their amounts, as a positive number>, "delivery_by_month": {"<YYYY-MM>": <dollars>, ...}, "category_spend": {"<category>": <dollars>, ...}}`. - Roll-ups: Months in the statement with no delivery transactions are included at 0. Categories in categories.csv with no transactions are included at 0. ## 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.
Audit three months of a synthetic household's bank statement. Find subscriptions still charging every month that nobody has used for over 60 days, same-day or next-day double charges that were never refunded, and the true monthly delivery spend once fees, tips and refunds are counted. Some subscriptions changed price, one was last used exactly 60 days ago, and a refunded duplicate and a two-day repeat are decoys. The prompt says months and categories with no activity are listed at 0. A friend's tip gets the delivery, double-charge and roll-up 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: forgotten_subscriptions forgotten_annual_cost double_charges double_charged_total delivery_by_month category_spend
Best model for this task so far: Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) (success, $0.0067, 25s).
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 | $0.19 | 31s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-25 | |
| AClaude Haiku 5.5default · thinking on | $0.0067 | 25s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-25 | |
| AClaude Sonnet 5.5default · thinking on | $0.06 | 14s | 0 | Fully autonomous | pass | Trust 80% | 2026-10-25 |
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.