Set acceptance questions first
Write the decisions the pilot must support before opening the sample. Can the documents support the target task? Are required fields present? Can tables be linked? Is source documentation sufficient for project review? A visually polished sample is less useful than one selected to answer these concrete questions.
Request an honest sample design
Choose representative slices and include difficult cases, known gaps, and source concentration. Ask whether the sample was randomly selected, curated, or constrained by access. Record its relationship to any proposed larger collection; a small handpicked set cannot establish the distribution of material that has not been reviewed.
Review artifacts and documentation together
Inspect original files, derived outputs, dictionaries, manifests, and coverage summaries side by side. Check identifiers and trace a few labels back to evidence. Record issues by severity and distinguish a repairable formatting defect from missing provenance or content that cannot support the intended model task.
Make the next decision explicit
Conclude with accepted scope, rejected elements, open questions, and requirements for another sample. Do not equate pilot fit with a cleared or fully available larger delivery. Send Fiund the acceptance plan and findings so follow-up sourcing conversations address specific evidence gaps rather than repeating a general request for more data. Keep a short decision log for the pilot with one row per issue, supporting artifact references, and the condition that would resolve it. This turns a general impression into reviewable next steps. A second sample can then target unanswered questions rather than merely presenting more examples of already accepted easy cases.
Review checklist
- Write decision questions before review
- Record how the sample was selected
- Trace labels back to original evidence
- Separate accepted scope from unresolved issues