# Formative usability route for a micro-firm

Use this route when fewer than six eligible workers are available, when a
workflow is still being shaped, or when the randomized readiness gate is on
HOLD. This is a usability rehearsal, not a smaller trial.

This route uses the printable/editable templates and requires no software
installation. It still requires an approved existing agent interface for the
copy-and-run rehearsal; it adds no agent or API integration.

## What it may learn

- whether one to five workers understand the protocol and its boundaries;
- whether the copy-and-run steps work in the approved agent interface;
- where people ask for help and how many support minutes they need;
- whether the quality rubric, consent language, stop route, and rollback card
  are comprehensible; and
- which parts should be revised before a comparative evaluation.

## What it cannot learn

Do not calculate or publish a randomized effect estimate, protocol-versus-agent
contrast, confidence interval, productivity claim, or evidence-status upgrade.
Do not call it a successful pilot. It cannot separate protocol value from the
agent, worker selection, learning, task difficulty, or facilitator support.

## Safe rehearsal

1. Use synthetic or fully redacted documents only.
2. Complete governance, consent, data, incident, and rollback checks even though
   randomization is not used.
3. Give the observer
   `formative-observation-and-failure-recognition-sheet.md`. Let each volunteer
   try the frozen copy-and-run protocol on one or two non-operational examples.
4. Ask the worker to think aloud about confusing steps, but do not score their
   performance. Then show the sheet's fixed deliberately bad output and observe
   whether the worker recognises unsupported claims, prohibited actions, the
   approval boundary, and the stop/help route.
5. Record only coded usability issues, help-request count, participant minutes,
   optional facilitator minutes, burden rating, the unsummed recognition notes,
   and any safety concern.
6. Treat a missed defect as a material/support design finding, not employee
   failure. Revise the materials transparently and record what changed.
7. End with one of: **Continue** bounded formative learning, **revise** the
   materials or workflow, or **stop** because the route is unsuitable. These
   process-navigation choices do not authorize rollout or an impact claim.

Do not run tools/pilot-summary.js on these records. A micro-firm may later join a
multi-company evaluation with shared governance and compatible task definitions,
but pooling and analysis require a separate prespecified protocol.