What buyers think

The agents in a normal run use your product. Evaluators judge what it says: skeptical buyers on a budget who read your landing and pricing pages and decide whether they'd buy. They need no browser, so even hundreds cost cents. Each one must name at least two concrete objections before giving a score, and leans towards "no".

Your pages

npx debutly perceive --icp "Owners of small marketing agencies" \
  --url https://staging.example.com/ --url https://staging.example.com/pricing \
  --competitors https://rival.example/pricing,https://other.example/pricing

With --competitors, evaluators judge you next to what your buyers would compare you with, read from those companies' own pricing pages.

Two versions: which do buyers pick?

npx debutly perceive --icp "…" --variant A=https://staging.example.com/pricing --variant B=pricing-b.md

Each variant is an address or a text file. Every evaluator sees both and picks one, or neither. The report says which won, with a 95% interval, and says so when the lead could be noise.

Before anything is built

npx debutly perceive --icp "…" --file pitch.md

A pitch, a landing page draft or a one-pager in a text file is enough.

After a run

Add --evaluators 50 (or "evaluators": 50) to a normal run, and evaluators judge the pages the agents saw.

The report

launch-readiness.md ranks the blockers and the objections buyers raised most, compares variants, and suggests fixes. It never shows a conversion rate: it is simulated evidence, and says so.

With --store, each prediction (the winner, the top objection, the share who would buy) is saved. Record what really happened with npx debutly predictions record <id> <actual>, and npx debutly predictions accuracy shows how often the evaluators were right.