Start with a question about the supplied files.
A legal team may need to find a clause, compare an amendment with an agreement, identify a dated event in correspondence or prepare a summary for review. These are document-work questions. A selected matter packet is different from a licensed case-law or legal-research collection.
Choose a task a qualified reviewer can score from the files. Keep the source set, question and acceptance criteria together. A source link helps the reviewer inspect an answer; it does not establish that a legal interpretation is correct or that all relevant authority has been considered.
- Law-firm teams: start with an authorized matter-document question and a clear source boundary.
- In-house teams: compare a supplied agreement against an approved review playbook.
- Legal IT and risk teams: inspect processing locations, provider routes, retention and the actual configuration.
Compare a fictional agreement with a supplied playbook.
The synthetic agreement gives a 45-day invoice review period in C1. Amendment A1 changes it to 30 days. The invented playbook P1 flags periods longer than 30. A reviewer should therefore compare all three documents before recording a finding. Reading C1 alone would miss the amendment.
A second check compares the agreement’s 90-day working-copy retention period in C2 against the playbook’s 60-day preference in P2. A third asks for a renewal term, which is absent. These are hand-authored illustrative reference findings, not model outputs or professionally approved legal recommendations. The fixtures have not received practitioner review.
Review each material statement.
Open the cited document and check the exact clause or passage. Confirm the document version, whether an amendment applies and whether the answer has omitted an exception. Distinguish an absent answer from an unfavorable answer. The reviewer should be able to correct the draft without having to guess which source the model used.
For a matter-summary task, include a conflicting date and a question that the packet cannot answer. Record source correctness, important-fact omissions, appropriate abstention and the time spent checking and editing. Do not describe an example finding as legal advice or assume a valid citation settles the legal conclusion.
Review the whole processing path.
The intended default is local inference, with separate approval of the purpose and data sent to optional external providers. A future package review must cover parsing, OCR where available, embeddings, indexes, logs, backups, diagnostics and exports. Local inference alone does not settle retention or confidentiality questions.
Include selected-file access and proposed actions in the review. Test whether the chosen configuration handles excluded or removed files as intended. Your organization controls device policies and decides what may be processed. Agree on a permitted evaluation corpus before introducing nonpublic matter material.
Confirm the supported Mac and the review owner.
Apple Silicon Macs are the planned first platform. A future evaluation will need to establish the supported operating system, hardware, package, model and file types. Bring mandatory integrations, different fleet platforms or language requirements into the fit discussion early.
A legal evaluation needs a qualified practitioner to establish expected findings and review professional claims. The public synthetic examples are preparation for that process. Actual document performance, review-time savings and matter separation must be demonstrated in the tested configuration before being claimed as results.
Describe the task without sending the matter.
Use the evaluation request to describe the document types, jurisdiction, device profile and intended work using nonsensitive information. The public form is not a matter-upload route. Start with a synthetic demonstration, then agree on ownership, handling and acceptance for any authorized evaluation.
An inquiry begins a fit discussion. It is distinct from an agreed evaluation, an accepted result or a paid pilot. The next decision should follow the evidence from a defined task rather than a general promise about AI for legal work.
Sources and further reading
- Qwerty product and selected-file workflow · Reviewed 2026-10-16
- Qwerty deployment and supported environment · Reviewed 2026-10-16
- Qwerty evaluation methods · Reviewed 2026-10-16