Bosca / Experiments

Bosca Experiments

Ship behind a flag.
Prove it with data.

One system for feature flags and A/B tests — typed flags with deterministic targeting, experiments that measure the difference, automated rollouts with guardrails, and results computed from your own analytics.

How it works

Flag it, test it, decide.

01

Flag it

Wrap any change in a typed flag — boolean, percentage, string, or JSON — with ordered targeting rules that decide who sees which variation. Everyone is bucketed deterministically, so a person always gets the same answer.

02

Test it

Point an experiment at one of the flag's rules and it measures one variation against another. Assignment is sticky per person, and exclusion layers keep overlapping experiments from colliding.

03

Decide & roll out

Results come straight from your analytics with real statistics and a clear ship / don't-ship call. A rollout policy then ramps the winner step by step, halting the moment a guardrail regresses.

What makes it powerful

Release control and accurate measurement

Everything you need to change software safely — typed flags, precise targeting, and gradual rollouts — plus the statistics to know whether a change actually worked.

Typed flags

Boolean, percentage, string, or JSON variations — a palette of typed values, each validated against the flag's type, with one marked default.

Ordered targeting

Rules evaluated top to bottom, with conditions on segments, principals, profile and device attributes, and even other flags.

Deterministic bucketing

A stable hash of the user plus a per-flag salt places everyone in the same variation every time. Regenerate the salt to reshuffle.

Sticky assignment

Each person — a signed-in principal or an anonymous install — keeps the same variation across sessions and devices.

Exclusion layers

Group experiments so a user enrolled in one is held out of the others, keeping concurrent tests from interfering.

Automated rollouts

Manual, scheduled, or data-adaptive ramps advance the winning variation step by step toward full traffic.

Guardrails that halt

Guardrail goals block a ship — or halt a rollout in progress — the moment they regress past the threshold you set.

Real statistics

Frequentist or Bayesian analysis over your analytics warehouse, with lift, confidence, CUPED variance reduction, and sample-ratio checks.

One system

Flags and experiments, the same machinery

A flag owns the typed variations, the targeting rules, and the deterministic bucketing. An experiment is just a measurement lens over one of those rules — so what you test is exactly what you ship, with no drift between the two.

  • An experiment doesn't own its variants — it measures the variations the flag already serves.
  • The same deterministic bucketing decides both who sees a flag and who lands in an experiment.
  • Exclusion layers partition traffic so concurrent experiments never contaminate each other.
  • Results read from the same analytics warehouse the rest of the platform uses — no separate metrics pipeline.
one flag, one rule
Flag checkout-cta — variations control, treatment
Rule mobile-users — 50 / 50 rollout
Experiment measures that rule
Rollout ramps the winner to 100%

Change software with confidence

Flags to release safely, experiments to measure accurately, rollouts to ramp automatically, and statistics you can trust — all reading from the analytics you already collect. The docs cover flags, experiments, exclusion layers, rollouts, and results end to end.