Skip to content

Turn a meeting transcript into tracked action items (owners, relative due dates, reassignments, cancellations) plus a 120-word summary

Exact prompt

# Task: turn the meeting transcript into tracked action items `transcript.md` is the transcript of a team meeting held on Tuesday 2026-03-10. Produce two files in this directory: ## 1. `actions.json` A JSON object with one entry for every ticket ID (`OPS-<number>`) mentioned in the transcript, and no other keys: `{"OPS-101": {"owner": "<full name>", "due": "YYYY-MM-DD" or null, "status": "open" | "done" | "cancelled"}, ...}`. - `owner`: the person's full name exactly as written in the attendee list. "I'll take it" / "I'll own it" / "I fixed it" makes the speaker the owner. - When the transcript changes an owner, due date or status later on, the LATEST statement wins. - `status`: `done` if it is said to be done/fixed/finished, `cancelled` if it is dropped/cancelled, otherwise `open`. A done or cancelled item keeps its last due date (or null). - `due`: null if no due date is ever given. Resolve relative dates against the meeting date (Tuesday 2026-03-10), with weeks running Monday-Sunday: - "by/on <weekday>" or just "<weekday>": the first such day strictly after the meeting date (so "by Friday" = 2026-03-13 and "by Tuesday" = 2026-03-17). - "next <weekday>": that weekday in the following calendar week (Mon 2026-03-16 - Sun 2026-03-22). - "end of this week" / "end of next week": the Friday of the meeting week / of the following week. - "in N days" / "in N weeks": the meeting date plus N days / plus 7xN days. - "end of the month": the last day of the meeting's month; "end of <month name>": the last day of that month in 2026. - "the <Nth>": that day of the meeting's month if it is after the meeting date, otherwise of the next month. - Dates mentioned that are not given as the due date (for example when something expires) are not due dates. ## 2. `summary.md` A short status summary for people who missed the meeting: at most 120 words (counted as whitespace-separated tokens), mentioning the ID of every ticket that is still open, and not mentioning the ID of any cancelled ticket. The grader compares `actions.json` with the correct answers field by field (every ticket must be fully correct) and checks `summary.md` against the three rules above. No network.

Later statements override earlier ones (owner hand-off, earlier due date, drop, done), and relative dates ('by Friday' vs 'next Friday', 'in 10 days', 'the 17th', an expiry date that is not a due date) must be resolved with the stated rules. 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: 1. tickets_complete: exactly the 10 ticket IDs. 2. owners_correct. 3. due_dates_correct. 4. statuses_correct. 5. summary_word_limit: summary.md 1-120 whitespace tokens. 6. summary_mentions_open: every open ticket ID appears. 7. summary_omits_cancelled: no cancelled ticket ID appears.

Best model for this task so far: Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) (success, $0.0056, 20s).

Run this on another model

All runs (4)

100% pass rate across 4 eligible runs: 4 pass, 0 partial, 0 fail. Partial and failed runs are listed below with everything else.ⓘ

Runs for this use case
ModelCostTimeStep-insⓘAutonomyOutcomeTrustⓘDateCompare
OGPT-6.1 Soldefault · thinking off$0.0438s0Fully autonomouspassTrust 80%2026-10-22
AClaude Haiku 5.5default · thinking on$0.005620s0Fully autonomouspassTrust 80%2026-10-19
AClaude Opus 5.5default · thinking on$0.1119s0Fully autonomouspassTrust 80%2026-10-19
AClaude Sonnet 5.5default · thinking on$0.0512s0Fully autonomouspassTrust 80%2026-10-19

Embed this result

A live badge for this use case. It links back here and updates as new runs are published.

UseCaseRank badge for Turn a meeting transcript into tracked action items (owners, relative due dates, reassignments, cancellations) plus a 120-word summary

Get new runs and ranking changes weekly

Which models got better or cheaper on real tasks like this one.