Read the existing planning guides.
The four public HTML resources below retain their October 1 filenames. The pack’s readme records an October 10 copy update. They describe planning questions and evaluation methods; they do not supply a production installer or establish current release acceptance. Check the package and model actually supplied for your evaluation.
Start with deployment if you own the device rollout, security if you review data paths, model approval if you select the weights, and the report template if you own the acceptance decision. Use the guides together when the same person covers more than one role.
Record actual observations in blank templates.
The CSV worksheet is a place to record task conditions, timings and outcomes. Define the expected result before a run, retain failed attempts and include time spent checking and correcting. For an accepted-task comparison, the denominator is work a reviewer actually accepted, rather than every answer a model produced.
The public ZIP combines the dated guides and templates. Its engineering snapshot concerns an earlier development report. That historical record should not be presented as the readiness of a newly supplied package or performance on your machine.
Check revisions, tables and diagrams against real documents.
Nine tasks use selected pages from two NASA Systems Engineering Handbook editions. The pack includes five text extracts, four source images, edition and page references, file hashes, reuse notes and an editable record format. It is a small systems-engineering document exercise that can be used with any suitable system.
The tasks include a real conflict between a table and a diagram in the same edition, as well as questions the supplied excerpts cannot answer. Import only the source files and question prompts into the system being evaluated; keep the reference key separate. The disclosed AI-assisted editorial review checks the reference readings. No application or model results are included.
The Python 3 checker uses the standard library to validate source hashes, citations and review-record structure. It reports recorded reviewer judgments; it does not decide whether an answer is correct. The full NASA handbooks and excluded third-party material are not redistributed. NASA is credited as the source and does not endorse Qwerty.
Calculate the cost of accepted work.
The offline calculator and editable workbook provide three blank scenarios for manual work, a local setup and a cloud or hosted setup. Include hardware allocation, software, support, administration, preparation, primary work, retries, review, API usage, energy and other attributable costs in the same reporting window.
Unknown inputs remain unknown. A component can be marked not applicable with a reason; zero accepted tasks leaves the per-task cost undefined. The pack contains formulas and calculation source, with no prefilled prices or product results. The offline tool makes no network requests and stores no entries in browser storage.
Try small synthetic reference packs.
The new October 16 examples are fictional plain-text documents with explicit section and clause references. They may be copied for evaluation. No customer files, real agency rules or operational military instructions are included. The contract fixtures have not received practitioner review.
The expected-findings CSV is a hand-authored illustrative key. It shows what the supplied text says, including a missing answer and a revision conflict. It is not model output, an executed demonstration or a product benchmark. Keep your actual trial answers in a separate record.
| Pack | What it lets you inspect | Workflow |
|---|---|---|
| Training-library revisions A/B | Required field, current owner, superseded schedule and missing archive address. | Technical-reference evaluation |
| Office-policy revisions 1/2 | Changed review schedule, preserved source labels and an unsupported disclosure question. | Policy-source evaluation |
| Agreement, amendment and playbook | A changed invoice period, a retention preference and an absent renewal term. | Contract-playbook evaluation |
Move from a reference exercise to an evaluation.
Choose one task and a reviewer who knows what acceptable work means. Confirm the supported Apple Silicon Mac, candidate package, model, file support, access rules and permitted connections. For a disconnected requirement, test the complete workflow and lifecycle rather than inferring readiness from a local answer.
Plain-text fixtures do not test scans, diagrams, long PDFs, language quality or access isolation. Add a rights-cleared representative corpus when the configuration and review process are ready. Record current source versions, package changes and unresolved requirements so another reviewer can repeat the test.
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
- NASA Systems Engineering Handbook — Rev1 official record · Reviewed 2026-10-17
- NASA Systems Engineering Handbook — Rev2 official record · Reviewed 2026-10-17
- NASA scientific and technical information reuse terms · Reviewed 2026-10-17