# Get a product change through specialist review

Marlquist · Product-management work sample
All company, partner, vendor and product information is original synthetic material.

## Situation and outcome
Marlquist is a fictional US spending account and debit card with about 410,000 cardholders; its cards are issued by Brasterly Community Bank under a program agreement. Quibbleway is a new in-app flow where an AI assistant drafts a card dispute and suggests whether it is fraud or a merchant problem. That type sets the path: fraud disputes get provisional credit within 10 business days and merchant disputes do not. The pilot to 5% of cardholders is planned for Mon 26 Oct, the launch council meets on Wed 14 Oct, and the initiative owner wants your status by Mon 12 Oct. The PM who ran the launch-review intake is now on leave, and the intake agent has written a summary of where everything stands.

Your outcome is the **launch-review record** for the initiative owner: a corrected intake (the change as it now stands with unknowns marked, your routing answers with reasons, and the third-party record with its gaps), a findings tracker that keeps each reviewer's own severity, and a status record that says what is and is not cleared, gives options with the premises each needs, and names who decides what. A well-argued "not cleared" or "cannot be set yet, because these named people have not answered" is a valid result. The specialist determinations themselves, closing any finding, the partner bank's decision, building a fix, the ship call and measuring or accepting the AI risk are not part of this work.

The evidence period is 2026-07-01 through 2026-10-09 inclusive. Examine 55 source records from policies and an agreement excerpt, specs and release notes, the intake ticket, review findings and trackers, vendor documents, employee dogfood and channel messages. The records and the correction note stand in for the people; do not contact fictional organisations or seek private data. The AI summary is not evidence authority. Work with the supplied source IDs and function IDs.

## Your role
You are joining the team as Product Manager.

You own this intake from today: corrected packet, routing with trigger answers, third-party record, findings tracker and the clearance status for the initiative owner (cleared, or a not-cleared record with options and the escalation route). Propose the AI risk tier with its rationale for Model Risk to set, and name the agreement clause for Compliance to confirm. You may not set the tier, close or downgrade a finding, judge a legal or privacy question, contact the Bank directly or approve the pilot.

## About the team
A fintech product team taking an AI card-dispute feature through its launch-review intake before a pilot.

An intake agent drafts packets and routing answers and wrote the 9 Oct summary you audit; a digest agent summarises dogfood. Reviewers use AI to summarise but own their findings, severities and closures. People own routing answers, tiers, approvals and the pilot decision.

Each reviewing function owns its findings and closes only its own; Model Risk sets the AI tier; Compliance owns the partner bank's submission; the initiative owner and the launch council decide the pilot.

## Work to complete
1. Read sources/coverage-and-access.md before making claims about who has reviewed what. The policies and the agreement excerpt are the first six records; the request to you is the last.
2. Audit the AI summary against the original records, row by row. Keep what the records support, and change only what they do not, naming the record that shows it. A row that still stands is kept, with its reason.
3. Count by function, not by record. Say which functions actually stand behind a sign-off, a finding or a status, and whether a record is a reviewer's own words, a relay of them or an opinion from elsewhere. A repeated or relayed record is not a second observation.
4. Answer each trigger question in the intake standard with a reason, route what needs a review, and say what needs re-intake and why. Do not exempt a function on a guess.
5. Keep each reviewer's finding and severity as the reviewer wrote it, including where reviewers or records conflict; refer the conflict to its owner rather than resolving it yourself. Only the function that raised a finding can close it.
6. Mark what is unknown, take a rare but serious event seriously, and keep contrary evidence and real benefits visible without letting either set the status.
7. Read the correction note (U001) with the other sources and apply it. Carry each change through the intake, the tracker, the status record and the revision log, and keep the original wording of a record the correction note disputes.
8. Keep authority where it sits. Propose the AI tier and the agreement clause for the people who set and confirm them, put asks to owners as requests, and state that the status is not pilot approval.

## Time
You have 90 minutes of work time. The clock starts when you press Start.

After that comes a 10-minute break, then a written interview: up to eight typed questions about your own files, for about 15 minutes. If you submit early, your break starts straight away.

## Mid-task correction
There is no mid-task correction in this take-home. Everything you need is in the files you download at Start, including the correction note **updates/U001-correction-and-build-scope-v1.md**. Read it with the other sources, apply it where it changes your work, and cite it as U001.

## Tools and help
Use any tools and help you like: AI tools and agents, search, your own notes, friends or colleagues. You don't need a paid tool, an API key or an account.

Nothing you do in your own tools is watched. Only your files and your interview answers are reviewed. When you submit, you can say which tools or help you used. That is optional, and leaving it empty is never held against you.

## Submission
Hand in one ZIP with these files at its top level. Start from the copies in the templates folder.
- README.md: list every file you hand in.
- intake.md: the corrected intake, about 900 words at most. Give the change as it now stands with unknowns marked and the build the pilot would use, each trigger answer in the intake standard with its reason and what has been requested and said, what needs re-intake, the third-party record with its gaps, and the AI tier and agreement clause you propose or flag at your level.
- tracker.csv: required columns finding,reviewer,reviewer_severity,required_action,owner,due,status,source_ids. One row per finding, at most 25 rows, with the severity the reviewer gave (or "none given"), a named owner and a date from the records where one exists.
- status.md: the status record for the initiative owner, about 700 words at most. State clearly that it is not pilot approval. Give what is on file, what is open or unreviewed, options with their premises and dates, the escalation route, the decisions for others and your questions to named owners.
- evidence.csv: the template already has one row for each record, R001–R055, with the record ID and function ID filled in. Do not type IDs. For each row, fill in the item the record bears on (a review, a finding or a claim) and how it is used (supports, contrary, relay, rule, scope or context are suggested). You may add rows for a record that bears on more than one item, and rationale and evidence_version values as useful. Record the correction explicitly if you apply it, and cite the correction note as U001.
- revision-log.md: each error you found in the AI summary with the record check that shows it, the rows you kept, and what the correction changed, about 600 words at most. No private prompts or unrelated files.

Words are counted in the prose of each document you write; headings, tables, the evidence file and the README do not count.

Use plain text files only: .md, .txt, .csv or .json. Put your work in the files themselves; a link to a document somewhere else does not count. Your ZIP can be up to 5 MB, or 20 MB unzipped, with at most 39 files of up to 2 MB each. Programs, macros, password-protected ZIPs and ZIPs inside your ZIP are not accepted.

Upload your ZIP on your take-home page and wait for **Received and checked**. It normally appears within a few seconds of the upload finishing. If it has not appeared after a minute, upload the ZIP again; if it still does not appear, use **Help and support**. You can replace your ZIP until you submit. Submitting is final and starts your break.

If time runs out before you submit, the last ZIP marked **Received and checked** is handed in for you. If none was received, the team is told that you need support.

If something goes wrong, use **Help and support** on your take-home page. A technical problem is never counted against you. Writing to us does not pause the clock.

## What happens next
The written interview covers these parts of the job:
- Ties status to evidence
- Handles unknowns and serious risks
- Checks and fixes the AI draft
- Hands over a usable status
- Explains the reasoning
- Routes reviews and records clearance

An AI then drafts a review from your files and your answers. Each point quotes what it rests on. A person at the company reads the review and makes any decision. No one is hired, rejected or ranked automatically.
