Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

README.md

Usage

A standalone documentation site is not yet published. This folder is the placeholder for end-user usage docs; until it grows, the in-product onboarding inside the web dashboard is the authoritative guide.

The release-decision loop

Every experiment moves through the same eight phases (CF-01 → CF-08). The web UI walks you through them; the same phases are encoded as Skills under skills/ so the coding agent can drive them too.

intent → hypothesis → exposure → measurement → analysis → decision → learning
Phase What you do Where
CF-01 Intent Turn a vague goal into a measurable outcome UI: experiment Setup tab
CF-02 Hypothesis Write a falsifiable claim ("if we do X, metric Y will move by Z") UI: experiment Setup tab
CF-03 / CF-04 Exposure Decide who sees what — feature flag + rollout FeatBit feature flag (managed in FeatBit)
CF-05 Measurement Pick the primary metric + guardrails, instrument the events UI: experiment Setup tab + your application code
CF-06 Analysis Run Bayesian A/B (or Thompson-sampling bandit) over collected data UI: experiment Run tab → Analyze
CF-07 Decision Frame the result as CONTINUE / PAUSE / ROLLBACK CANDIDATE / INCONCLUSIVE UI + activity log
CF-08 Learning Capture a structured postmortem so the next iteration starts from evidence UI: experiment Learning tab

Choosing a data source

When you create an experiment in Expert Setup, the Data Source step lets you pick how analysis pulls per-variant statistics:

Mode What it does
FeatBit Managed (default) Auto-pull from the bundled track-service (ClickHouse-backed). Easiest path.
Customer Managed Endpoint — single Call your own HTTPS endpoint that returns experiment statistics on demand. Use this when you already have a data warehouse and don't want to ingest events twice.
Customer Managed Endpoint — per-metric Same as above, but route different metrics to different endpoints.
Paste manually Type per-variant totals directly in the Primary metric / Guardrails steps.
External / other Free-text note. No live fetch — record only.

The Customer Managed Endpoint contract (request / response shape, auth, retries) is implemented in modules/web/src/lib/stats/customer-endpoint-client.ts.

Two chat modes

Inside an experiment, the chat panel can drive a coding agent through the workflow. Toggle persists per-browser:

  • Managed (default) — web proxies to FeatBit's hosted Managed Agents. No client-side install.
  • Local Claude Code — run npx @featbit/experimentation-claude-code-connector on your own machine; the chat panel connects to it on 127.0.0.1:3100 and uses your local Claude Code CLI.

API contracts

For programmatic access — analyse a run, list running experiments, query per-variant statistics — the canonical contracts are documented in AGENTS.md under "Key API contracts".


Found a gap? Open an issue or PR — this folder is intentionally a stub waiting for production usage docs.