:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.
Python ab-testingai-analyticsanalyticscdpdata-warehouseexperimentsfeature-flagsjavascriptproduct-analyticspythonreactsession-replaysurveystypescriptweb-analytics
import { HogFlowAction } from '~/cdp/schema/hogflow'
import { CyclotronJobInvocationHogFlow } from '~/cdp/types'
import { findNextAction } from '../hogflow-utils'
import { ActionHandler, ActionHandlerOptions, ActionHandlerResult } from './action.interface'
type Action = Extract<HogFlowAction, { type: 'random_cohort_branch' }>
export class RandomCohortBranchHandler implements ActionHandler {
execute({
invocation,
action,
}: ActionHandlerOptions<Extract<HogFlowAction, { type: 'random_cohort_branch' }>>): ActionHandlerResult {
const nextAction = getRandomCohort(invocation, action)
return { nextAction, result: { assigned_cohort: nextAction.id } }
}
}
export function getRandomCohort(invocation: CyclotronJobInvocationHogFlow, action: Action): HogFlowAction {
// Programmatically-authored nodes can be stored without their cohorts array (the API doesn't
// require it on lenient saves); assign nothing and fall through the continue edge instead of
// crashing the run.
const cohorts = Array.isArray(action.config.cohorts) ? action.config.cohorts : []
if (cohorts.length === 0) {
return findNextAction(invocation.hogFlow, action.id)
}
const total = cohorts.reduce((sum, cohort) => sum + cohort.percentage, 0)
// A zero or NaN total has no proportions to split by, so fall through like the empty case.
if (!(total > 0)) {
return findNextAction(invocation.hogFlow, action.id)
}
// Percentages act as relative weights, scaled by whatever they actually add up to. Summing to 100
// is the same thing; not summing to 100 (an even N-way split of a count that doesn't divide 100,
// say) still gives every cohort its intended share, rather than piling the shortfall onto the last
// cohort or making later cohorts unreachable.
const random = Math.random() * total
let cumulativePercentage = 0
for (const [index, cohort] of cohorts.entries()) {
cumulativePercentage += cohort.percentage
if (random <= cumulativePercentage) {
return findNextAction(invocation.hogFlow, action.id, index)
}
}
// Unreachable except for floating-point drift in the cumulative sum, where the last cohort is
// the intended landing spot anyway.
return findNextAction(invocation.hogFlow, action.id, cohorts.length - 1)
}