Design, assign, measure, decide, ship. Crawltic runs the whole lifecycle behind a single API, with the statistics and the rollout safety built in, not bolted on.
Every unit is hashed onto a 10,000-slot bucket line and lands in the same variant on every request, on every device you key it to. No lookup, no round-trip, no flicker.
Clicks are a proxy. Crawltic compares variants on whatever you emit: earnings per click, revenue per visitor, churn, latency. Mark any metric as a guardrail and it can veto a winner on its own.
Every results call returns probability-to-beat-control, credible intervals, p-values and confidence intervals together. Your data team picks the framework; the engine keeps both honest.
Checking a fixed-horizon test early inflates false positives. Crawltic spends alpha across the whole run, so the boundary starts strict and relaxes as evidence accrues.
The moment the statistic crosses, the test stops, often well before the planned sample, and the decision engine takes over.
Set confidence thresholds, sample quotas and guardrails once. Crawltic promotes the winner through a staged ramp and pulls it back the second a guardrail trips.
Recalculated every minute · ramp starts the moment the last quota clears
Ramp 1 → 5 → 25 → 50 → 100% · each step holds until guardrails re-verify · breach = instant rollback + webhook
When the goal is conversions now rather than a clean read later, switch an experiment to bandit mode. Thompson sampling re-weights traffic every minute, so the strongest arm earns the most exposure while the others keep just enough to stay honest.
Everything above is reachable from your code, your agent, or your terminal. Pick the surface that fits the moment.
Full CRUD on experiments, metrics and rollouts. OpenAPI spec and typed SDKs for TypeScript, Python, Go and Ruby.
Give Claude, Cursor or any MCP client scoped tools to create, monitor and ship experiments in plain language.
One static binary. Script creation, ramps and result checks from CI, cron, or your shell.
One key unlocks assignment, analytics, the decision engine, MCP and the CLI.