Four disciplines. One system.
Feature management, safe rollouts, trusted measurement, and adaptive optimization usually mean four tools and four sources of truth. Traffical builds all four on a single primitive — the typed parameter — so every change gets the whole lifecycle.
Everything is a policy on a typed parameter
A flag, a ramp, an experiment, and a bandit are not four products — they are four policies assigning values to the same parameter. Move between them without re-instrumenting anything.
Feature management
A feature flag is just a boolean parameter under a policy — targeting, kill switches, and scheduled releases without a separate tool.
Graduates into a rollout or experiment without re-instrumenting.
Safe rollouts
Progressive delivery with automated ramps, health checks, and rollback on guardrail breach — reversibility built into the lifecycle.
Health gates reuse the same metrics your experiments trust.
Trusted measurement
Experiment evidence computed in your warehouse from the metric definitions your business already trusts — estimated lift with 95% intervals.
Every flag flip and ramp step is measurable — same events, same statistics.
Adaptive optimization
When there is no single winner, adaptive policies learn from rewards — Thompson Sampling, UCB1, epsilon-greedy, and contextual bandits.
Picks up where the experiment leaves off, governed by the same guardrails.
Built for production, documented in depth
The mechanics underneath the four disciplines — each one covered end to end in the docs.
The building block of everything
Numbers, strings, enums, JSON — typed, versioned values with safe defaults. One parameter drives web, mobile, backend, and algorithms; booleans are the degenerate case.
Docs →Concurrent experiments, clean data
Layers give mutual exclusivity within and orthogonal bucketing across — teams test at the same time without collisions, interaction effects, or sample pollution.
Docs →One lifecycle, fully audited
Flags, A/B tests, rollouts, and bandits are all policies assigning parameter values. Every lifecycle change lands in the decision log — who did what, when, and why.
Docs →Evidence where your data lives
Define assignments and facts as SQL; Traffical aggregates and runs significance tests inside Postgres, BigQuery, Snowflake, Databricks, or ClickHouse.
Docs →Peek freely, decide confidently
Anytime-valid sequential testing, CUPED variance reduction to reach significance sooner, and ratio metrics via the delta method — uncertainty stated, never hidden.
Docs →Sub-millisecond, no hot-path calls
SDKs resolve values in-process from a synced config bundle — works offline, works at the edge, with a server evaluation mode when decisions must stay central.
Docs →For your workflow — and your agents'
Config lives beside your code and syncs from the command line. Agents get the same surface, scoped by governance.
Parameters, events, and metrics live in .traffical/ YAML beside your code — push, pull, sync, reviewed in the PR.
Your coding agent sets up Traffical, parametrizes code, and wires events — Claude Code, Cursor, and 40+ agents.
Agents inspect evidence and decision logs, preview transitions, and propose changes — least-privilege, humans own approvals.
Create A/B tests on live pages without writing code, using a browser bookmarklet.
Inspect SDK state, current assignments, and exposure events live in your browser.
See all four on one of your changes
Bring one high-impact product change. We'll take it through parameterization, controlled release, measurement, and an evidence-based rollout — on your metrics, in your warehouse.