Tools provided
Tools the agent exposes — your AI client calls them automatically when it needs them.
- run_full — Orchestrated path: design an A/B test — hypothesis, primary and guardrail metrics, required sample size, estimated duration — or read out a running or finished test with p-value, effect size, guardrail check and a ship or iterate decision.
- analyze — Run the pure deterministic statistical readout over provided experiment data with no model call, returning the statistics and the decision.
- design_test — Design the test from provided inputs: required sample size per variant, total sample, target rate, estimated duration and the design findings.
- compute_sample_size — Compute only the required sample size per variant and in total for the provided design inputs.
- estimate_duration — Estimate how many days the test needs at the given daily traffic and variant count. Reports not-estimable rather than inventing traffic.
- check_guardrails — Run the two-proportion z-test on every observed guardrail metric with the Bonferroni-corrected alpha and return each result.
- check_peeking — Check the observed sample size against the design requirement — the peeking and early-stopping guard that decides whether a call can be made at all.
- get_decision — Return the ship, iterate, hold, continue or inconclusive decision along with its statistical rationale.
- list_capabilities — List the agent's static capabilities: the available tools and the statistical defaults for alpha, power and minimum detectable effect. Useful for discovery.
- plan_inputs — Plan the inputs for a tool: returns the clarifying questions, the JSON schema and a ready-to-edit example.
- discover_intent — Understand your goal and co-design the exact input through clarifying questions before running.
- open_form — Opens the guided form to collect the test design or the observed counts. The entry point for running the agent on your own numbers.
- run_form — Internal: invoked by the guided form when the user submits. Not for direct use.
