FICTIONAL PRACTICE PACK: morning worklist
Version: 2026-10-07. All people, IDs, dates and records are made up.
Guide: https://animasai.co/service-backlog-to-morning-worklist

Open requests.csv and updates.csv in a spreadsheet. These are separate exported copies, not connections to a work system. There are nine input rows (seven requests and two updates). Keep the originals.

For this exercise only, review time is 2026-10-06T09:00:00-05:00 and an export older than 24 hours is stale. A newer record with a timestamp can supersede an older one for the same stable request ID. Complete requires an explicit staff completion record. Draft ready does not mean sent. Do not infer an order when timestamps are missing. Real teams must choose their own freshness and status rules.

Use review-blank.csv for the proposal. Copy this instruction into the AI tool you already use and provide only the two fictional exports:

Prepare a proposed morning worklist from these two copies. Apply the exercise rules above. Cite every source_row and request_id. Account for all nine input rows, including duplicates and completed items. Keep conflicts, stale exports and missing IDs visible as holds. Do not invent dates, resolve conflicting states without evidence, or treat a draft as sent. Leave reviewer and checked_at blank. Do not send, assign, close or edit live records.

Compare the proposal with expected-review.csv. Wording can differ; the dispositions and source-row coverage should agree. Expect six review entries: one recorded complete, one draft ready, one open request, and three holds. A3 and A4 appear together; A1 and U1 appear together; A5 and U2 appear together. Every input row appears exactly once across the entries. No row is silently dropped.

The answer key is an editorial example, not the output of a measured AI run. Do not upload the key as an input if you want to assess the proposal. A neat match does not establish reliability on real data. Check every inclusion and omission against current records before assigning work.