Growthbook Experiments Tracker

Internal Apify Actor that fetches experiments from the GrowthBook REST API, runs a set of health checks on them, and posts a status digest to Slack. Built to keep the experiments used by apify-core and apify-web honest — following the practices from the "GrowthBook at Apify" wiki page.

What it reports

Each run covers three groups (all filterable by tags, owner, and project):

• Running experiments — the core status list, with owner, days running, traffic coverage, and exposed users. Experiments whose rule is not enabled in the production environment are marked staging only and exempt from the running-too-long check. Because GrowthBook records no date for the production switch, the tracker remembers it itself (named key-value store growthbook-experiments-state): for experiments first seen as staging-only, "days running" starts the day the production rule turns on; for experiments already in production when first tracked, it falls back to the phase start.
• Stale drafts — drafts created long ago but never launched (optional).
• Recently stopped — experiments stopped within the lookback window (optional).

Health checks

Check	Severity	Meaning
decided-but-running	🔴 critical	Result is documented (won/lost/inconclusive) but the experiment still runs — burning billed exposure events
running-too-long	🟠 warning	Current phase exceeds maxRunningDays (escalated wording at 100% traffic)
high-exposure	🟠 warning	More users exposed than maxExposedUsers
sample-ratio-mismatch	🟠 warning	SRM p-value < 0.001 in the latest analysis — bucketing is likely broken (e.g. missing hash attribute)
overdue-status-update	🟡 attention	The scheduled status update date has passed
stale-draft	🟡 attention	Draft older than staleDraftDays, never launched
stopped-without-conclusion	🟡 attention	Stopped recently with no documented winner/conclusion
missing-hypothesis	🔵 hygiene	Running without a hypothesis
no-tags	🔵 hygiene	Running without tags (convention: tag with team + surface)
no-owner	🔵 hygiene	No owner assigned

Input

See the Input tab / .actor/input_schema.json. The important ones:

• growthbookToken (required, secret) — GrowthBook REST API secret key (secret_...). Org-scoped, created under GrowthBook → Settings → API Keys.
• tags, owner, projectId — optional GrowthBook filters.
• includeStagingOnly — set false to hide running experiments not yet enabled in production (their production start is tracked either way).
• maxRunningDays, staleDraftDays, recentlyStoppedDays, maxExposedUsers — health-check thresholds.
• slackToken (secret) + slackChannel — when both set, the report is posted to Slack. Without them, the run just logs the report and fills the dataset. The bot needs chat:write (plus users:read and users:read.email to @mention experiment owners by their GrowthBook email) and must be a member of the channel.

Output

• Slack: one channel message (summary counts, issues sorted by severity with owner mentions, running-experiments list) and per-experiment detail cards in its thread.
• Dataset: one record per experiment with status, phase, exposure, result, and detected issues[] — usable by other integrations.

Development

npm install
npm test              # unit tests for the health checks
npm run lint
apify run             # local run; input from storage/key_value_stores/default/INPUT.json
apify push            # deploy

The local INPUT.json is gitignored-friendly — never commit real tokens.