Developer adoption pilot

Status: pending participants; no observed usability improvement claimed.

Protocol

Recruit 3–5 external TypeScript developers with no prior AgentPlat experience. Obtain authorization before contacting anyone. Use synthetic proposal data and anonymous participant IDs. Participation is voluntary; collect only task timing, errors, documentation questions and facilitator interventions, not credentials or model content. Agree with participants on storage and deletion of notes.

Use the persistent collaboration guide on a clean checkout at a recorded commit. Record OS, Node/pnpm/PostgreSQL versions, Docker availability and whether prerequisites were installed beforehand. Record prerequisite setup separately from the timed task; do not hide failed setup attempts.

Start the 45-minute clock when the participant begins the documented install commands with prerequisites available. Ask them to:

1. Start the API and complete the two-agent proposal demo.
2. Locate the human revision decision and approved proposal version.
3. Run recovery and identify retained work and the operation identity.
4. Explain what is persistent, what is mock, and which protections and identity checks the application must configure.

Allow documentation lookup. Log help verbatim; corrective facilitator help makes the primary unaided-completion endpoint false. Stop at 45 minutes and record incomplete work. Do not omit failures or replace participants to improve results. Do not use real model calls in the timed path.

Acceptance and analysis

The exploratory target is at least 80% unaided completion of all four tasks within 45 minutes: 3/3, 4/4 or 4/5, depending on final cohort size. Freeze cohort size before sessions. Report counts and individual durations, not a population productivity claim. There is no measured baseline comparison in this protocol.

Record blockers, change the guide or example, rerun automated checks and repeat blocked tasks. Keep initial and follow-up outcomes separate. Technical checks may pass while this pilot remains pending.

Observation template

Copy this table for each anonymous participant; leave observations empty until an actual session. Do not populate it with synthetic success data.

Aggregate report: planned cohort, attempted sessions, unaided completions, assisted completions, incomplete sessions, durations, recurring blockers and follow-up results. Owner schedules sessions when participants are available.