Maze

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me the unmoderated testing tool I actually need instead of Maze.

Be honest about the split first: **the software is a week and the panel is the product.** If you have no way to recruit testers, build this and buy participants elsewhere; if you have a customer list, this is entirely worth building.

STACK
- Node 20+ with Fastify, server-rendered HTML
- SQLite through better-sqlite3, WAL mode
- A small vanilla JS recorder on the tested page
- Caddy in front

THE DATA MODEL
- studies: id, name, target_url, tasks_json, questions_json, status, max_participants
- participants: id, study_id, token, started_at, finished_at, device, source, consent_at
- task_results: id, participant_id, task_index, outcome, duration_ms, path_json, misclicks — outcome is success, gave up, or wrong destination
- answers: id, participant_id, question_index, value
- interactions: id, participant_id, kind, selector, x_pct, y_pct, at_ms — sampled, not every mouse move

RUNNING A TASK
- The participant gets a token URL, reads a task, and is dropped on the target with a small unobtrusive bar showing the task and a 'done' and a 'give up' button
- **The 'give up' button matters more than the success button.** Without it, everybody who fails simply closes the tab and your completion rate is a lie
- Success is a destination or an event you declare in advance, verified on the server rather than self-reported
- A short question after each task, one line, optional

RECORDING WITHOUT BUILDING A SESSION REPLAY
- Clicks with a stable selector and a percentage position, page changes, scroll depth at intervals, and time per step. That is enough for every conclusion this tool produces
- **Do not record keystrokes and do not record input values.** You are testing strangers on a real site; capturing what they type is capturing their data
- Cap the event rate and the payload

THE ANALYSIS THAT IS ACTUALLY USEFUL
- Completion rate, median time and misclick count per task. Three numbers
- The most common path and the most common wrong turn, which is where the design problem actually is
- A click map per screen from the percentage positions, which needs no library
- **Show the sample size next to every number, and refuse to draw a chart under about a dozen participants.** Percentages from five people are the single most common misuse of this category and they end arguments with false confidence

RECRUITING, WHICH YOU HAVE TO SOLVE SOMEHOW
- An email invite to your own users with a token per person
- An intercept on your own site — one in fifty visitors, capped, never twice to the same person
- Consent and incentive stated before the first task, and recorded

WHAT MATTERS MOST
The give-up button and the sample size beside every figure. The first makes the data true; the second stops the data being misused.