# Autonomous Product Laboratory

Webdev Agent OS v12 adds a governed product-discovery system that reduces uncertainty before full production commitment.

## Discovery chain

1. Record research evidence with a visible source and evidence level.
2. Express falsifiable problem, customer, value, usability, feasibility, viability, growth, or pricing hypotheses.
3. Generate multiple competing product concepts instead of accepting the first idea.
4. Build prototypes at the cheapest fidelity capable of answering the current question.
5. Require reviewer approval before consequential experiments start.
6. Record results with evidence references, sample size, observed metric, and interpretation.
7. Score competing concepts and prototypes through a transparent tournament.
8. Require a human product decision before a winner is promoted.

## Autonomy boundary

The laboratory can organize evidence, calculate scores, generate alternatives, identify missing proof, and recommend experiments. It cannot fabricate interviews, market data, analytics, customer behavior, statistical significance, or production outcomes.

External research, user testing, paid traffic, customer contact, spending, and production promotion remain explicitly controlled activities.

## Evidence levels

- **Hypothesis:** proposed but unobserved.
- **Anecdotal:** limited testimony or informal observation.
- **Observed:** directly seen in a project or workflow.
- **Validated:** repeated or independently confirmed.
- **Proven:** substantial repeated evidence with successful reuse.

## Product decision rule

A high score is a recommendation, not an authorization. Reviewers retain the final decision and must document rationale, accepted risk, and the evidence actually considered.