Field guide

From working draft to inspectable evidence.

The field lab separates authoring, issued participant material, response evidence, analysis, and a candidate structure. That separation prevents later draft edits from quietly changing the meaning of earlier responses.

1. Define the research question. Choose open, closed, or hybrid. Write a bounded purpose, participant instructions, study-specific information, completion rules, and a non-identifying session-code process. The app does not supply consent or ethics approval.
2. Author cards and seed categories. Keep labels parallel, concrete, and understandable without surrounding navigation. Participant explanations are issued; private researcher notes are not. For closed or hybrid studies, check that category capacities can hold the required cards.
3. Pilot the exact experience. Try keyboard-only movement, a narrow screen, long labels, category limits, an unfinished save, and a restore. A pilot is where you find confusing labels and impossible rules before asking participants for effort.
4. Freeze an immutable edition. Freezing creates a participant-safe copy and a SHA-256 content fingerprint. Later draft changes do not mutate it. Create another edition for materially changed labels, cards, groups, or rules.
5. Issue the standalone file. Download one HTML file. A participant can open it directly, sort without an account or connection, save progress, restore the saved response, and download a final JSON response. Nothing is submitted automatically.
6. Reconcile response files. Import one or many JSON files. A batch commits only if every non-duplicate file validates. Exact duplicates do not add evidence. Higher-revision amendments replace the earlier file; same-or-lower conflicting revisions stop the batch.
7. Review before analysis. Inspect completion, duration, unsorted cards, all-in-one patterns, singleton patterns, labels, and comments. Flags are descriptive prompts, not proof of participant quality. Exclusion requires a reason and never deletes the file.
8. Inspect counts and method sensitivity. Read exact pair and placement denominators, raw labels, individual sorts, response-pair agreement, and all linkage merge steps. Compare single, complete, and average linkage before turning one cut into a candidate draft.

Folder pattern for a real study

study-name/
00-approved-study-material/
01-issued-edition/
02-returned-response-files/
03-project-checkpoints/
04-exported-analysis/
05-retention-and-deletion-record/

Keep the participant HTML and every returned response for one edition together. Retain the approved study information and your external consent, privacy, retention, and withdrawal process alongside the technical files.

What to test before fieldwork

Open the local field lab