capx-ai/casa/casa-experiment

casa-experiment

The experiment craft. Turns a vague "let us try X" into a falsifiable bet with a metric and a guardrail, and keeps a durable record so the company stops re-running losers.

Steps

1. If company-brain/profile.json does not exist, tell the founder to run /casa-start

first and stop.

1. Frame the hypothesis. State it as: if we do X, then primary metric M moves by

roughly E, because of mechanism C. A hypothesis with no mechanism is a guess.

1. Define the read before you ship. Fix the one primary metric, the guardrail metric

that must not break, the audience and the window, and the decision rule (what result means ship, kill, or iterate). Check company-brain/experiments.jsonl and learnings.jsonl so you are not re-running something already settled.

1. Log it at the start. Record the experiment in the ledger:

`` node ${CASA_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/brain.mjs experiment company-brain \ '{"id":"<slug>","hypothesis":"...","primary_metric":"...","guardrail":"...","window":"...","status":"running"}' ``

1. Ship it. Execute the change (route any paid action to explicit founder

approval, any UI to casa-design, any copy to casa-write). Respect human gates.

1. Read it honestly. When the window closes, read the result with casa-readout so the

analyst-honesty persona checks the window, the denominator, and the guardrail. Do not credit the change without the mechanism holding up.

1. Close the loop. Log the outcome (status won, lost, or inconclusive, plus the

number and the learning) back to the ledger, and run casa-compound so the lesson is reusable.

Rules

• No experiment without a primary metric and a guardrail defined before shipping.
• Every experiment is logged to the ledger, win or lose. The losses are the most

valuable record.

• No causal claim without the mechanism and a fair read.
• No em-dashes, no emojis.