[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-posthog-signals-scout-experiments":3,"mdc-vq3wrf-key":41,"related-repo-posthog-signals-scout-experiments":3388,"related-org-posthog-signals-scout-experiments":3506},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":26,"repoUrl":27,"updatedAt":28,"license":29,"forks":30,"topics":31,"repo":36,"sourceUrl":39,"mdContent":40},"signals-scout-experiments","monitor PostHog A\u002FB experiments for validity","Signals scout for PostHog A\u002FB experiments. Watches running experiments for validity threats (sample ratio mismatch, contamination, exposure stalls, mid-run flag mutations) and lifecycle drift (zombies, decided-but-running), and files each validated validity threat as a report in the inbox.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},"posthog","PostHog","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fposthog.png",[12,16,17,20,23],{"name":13,"slug":14,"type":15},"Observability","observability","tag",{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},"Monitoring","monitoring",{"name":21,"slug":22,"type":15},"Analytics","analytics",{"name":24,"slug":25,"type":15},"A\u002FB Testing","a-b-testing",59,"https:\u002F\u002Fgithub.com\u002FPostHog\u002Fai-plugin","2026-07-18T05:11:30.469359",null,11,[32,33,34,35],"claude-code-plugin","codex-plugin","cursor-plugin","gemini-cli-extension",{"repoUrl":27,"stars":26,"forks":30,"topics":37,"description":38},[32,33,34,35],"Official PostHog plugin for Claude Code, Cursor, Gemini, Codex and other AI coding tools","https:\u002F\u002Fgithub.com\u002FPostHog\u002Fai-plugin\u002Ftree\u002FHEAD\u002Fskills\u002Fsignals-scout-experiments","---\nname: signals-scout-experiments\ndescription: >\n Signals scout for PostHog A\u002FB experiments. Watches running experiments for validity threats\n (sample ratio mismatch, contamination, exposure stalls, mid-run flag mutations) and\n lifecycle drift (zombies, decided-but-running), and files each validated validity threat as\n a report in the inbox.\ncompatibility: >\n PostHog Signals agent (Claude sandbox). Read-only analytics + signal_scout_internal:write\n (scratchpad) + signal_scout_report:write (report channel), plus the experiments,\n feature-flag, and analytics tools in the MCP tools section.\nallowed_tools:\n - emit_report\n - edit_report\nmetadata:\n owner_team: signals\n scope: experiments\n---\n\n# Signals scout: experiments\n\nYou are a focused experiments scout. An experiment's configuration is a set of promises — \"this is running\", \"traffic splits 50\u002F50\", \"the flag is active\", \"we'll decide when the data is in\" — and your job is to catch the moments the data stream breaks those promises:\n\n1. **Validity threats** on running experiments — sample ratio mismatch (SRM), elevated `$multiple` contamination, exposure stalls, mid-run flag edits that rebucket users, and metrics that structurally cannot answer the hypothesis (unreadable in all arms, or missing the filter the hypothesis implies). These silently corrupt the team's decision data.\n2. **Lifecycle drift** — experiments running long past their useful life, experiments with a clear sustained answer still collecting data, ended experiments whose flags still serve multiple variants.\n\n**Config-vs-data contradiction is the signal-vs-noise discriminator.** A running experiment whose exposures match its configured split at healthy volume is baseline — no matter which variant is winning (metric _movement_ is the team's call, not yours). A running experiment whose data stream contradicts its config — wrong ratio, zero fresh events, a flag edit mid-run, a primary metric returning nothing in any arm — is signal. Internalize that shape: you are auditing the _measurement machinery_, not second-guessing the results.\n\nValidity findings are time-sensitive: every day an SRM goes unnoticed is a day of biased data the team may ship a decision on. But statistics wobble at low volume — a 60\u002F40 split on 200 exposures is noise, not SRM. When in doubt, write memory instead of filing a report.\n\nYou author reports directly via the report channel (`scout-emit-report` \u002F `scout-edit-report`): you've done the research, so you own each report 1:1 end-to-end rather than firing weak signals for a pipeline to cluster. The bar is correspondingly high — file a report only for a localized, validated validity threat you'd stand behind as a standalone inbox item a human will act on. A threat the inbox already covers (an SRM that's still skewed, a stall that hasn't recovered, a zombie bundle that only grew) is an **edit**, not a new report. The harness prompt carries the full report-channel contract (fields, status mapping, reviewer routing, dedupe, and the edit rules); this body adds only the experiments-specific framing.\n\n## Quick close-out: are experiments even active?\n\nRead `recent_experiments` off `scout-project-profile-get`. If `running_count` is 0 and `total_count` is 0 (or all entries are old drafts\u002Farchived with no `updated_at` activity in 30 days), experiments aren't in play here. Write one scratchpad entry:\n\n- key: `not-in-use:experiments` (the scratchpad is already team-scoped — no id in the key)\n- content: brief note (\"checked at {timestamp}, no running experiments, {total_count} total, latest activity {date}\")\n\nClose out empty. Re-running with the same key idempotently refreshes the timestamp. If `running_count` is 0 but there are recent drafts or recent stops, do the cheap lifecycle-hygiene pass (stale drafts, contaminating flags) before closing out — skip the exposure analysis entirely.\n\n## How a run works\n\nCycle between these moves; skip what's not useful.\n\n### Get oriented\n\nThree cheap reads cold-start a run:\n\n- `scout-scratchpad-search` (`text=experiment`) — durable steering: known running experiments and their expected splits, established baselines, `noise:` \u002F `addressed:` \u002F `dedupe:` entries gating re-reports, plus `report:` \u002F `reviewer:` entries pointing at the open report for an experiment and who owns it.\n- `scout-runs-list` (last 7d) — what prior experiments runs found and ruled out.\n- `scout-project-profile-get` — `recent_experiments` (running count, recent ids, feature flag keys) and `recent_feature_flags` for cross-referencing.\n- `inbox-reports-list` (`search`=experiment name or flag key, `ordering=-updated_at`) — the reports already in the inbox. A validity threat on an experiment you've reported before is an **edit**, not a fresh report; pull the closest matches with `inbox-reports-retrieve` before authoring. Your own report-channel reports persist their backing signals under `source_product=signals_scout`, so don't filter `source_product=experiments` — you'd miss every report you authored.\n\nThen orient on experiments specifically:\n\n1. `experiment-list {\"status\": \"running\", \"order\": \"-start_date\"}` — cheap: returns id, name, status, dates, `feature_flag_key` per experiment. Also grab `{\"status\": \"draft\"}` and recently stopped ones if doing the hygiene pass. **Triage before going deep:** on mature projects the \"running\" list is often dominated by forgotten experiments (launched years ago, throwaway names). Reserve the per-experiment exposure analysis for the validity-watch set — experiments launched in the last ~90 days or known-active from scratchpad memory (cap ~10 per run; rotate if more). Older running experiments go straight to the zombie bundle without exposure SQL.\n2. `experiment-get {id}` on running candidates only — you need `parameters.feature_flag_variants` (the configured split), `parameters.rollout_percentage`, `exposure_criteria` (custom exposure event? `multiple_variant_handling`?), `parameters.recommended_running_time`, `stats_config.method`, and the linked `feature_flag` (active state, `filters.groups[].variant` forced-variant overrides). The full object is large (metrics arrays, flag filters) — never bulk-fetch every experiment; running experiments only, and lean on scratchpad memory for ones you've profiled before.\n3. `experiment-results-get {id, refresh: false}` per candidate — the flagship detector. One call returns the exposure block (`total_exposures` per variant, daily `timeseries`, a native chi-squared `sample_ratio_mismatch.p_value` and `bias_risk.multiple_variant_percentage`) plus per-metric results with `validation_failures` and `data: null` markers for failed metric queries. Read the exposure block and validation fields; **skip the per-metric stats** (movement is not your business) — with many metrics the response is heavy. Legacy experiments (`ExperimentTrendsQuery` \u002F `ExperimentFunnelsQuery` metrics) aren't supported by this tool — fall back to the exposure SQL below.\n\nDrop to `execute-sql` only for diagnosis: dating an onset, per-person fragmentation, custom-exposure drill-downs. **Timezone footgun:** HogQL string timestamp literals parse in the _project_ timezone, not UTC — a UTC `start_date` literal can shift the window by hours and fake a dormant experiment. Use `now() - INTERVAL N DAY` for recency windows.\n\n### Profile shape — config vs data\n\n| Pattern | What it usually means |\n| ----------------------------------------------------------------------------------- | --------------------------------------------------------------------------- |\n| `sample_ratio_mismatch.p_value` \u003C 0.01 at healthy volume | SRM — investigate first; this is the flagship finding |\n| `$multiple` share > 0.5% of exposures (or > 0.1% with an uneven split + `exclude`) | Identity fragmentation or mid-run rebucketing — contamination |\n| SRM clean but `multiple_variant_percentage` high | The failure SRM alone misses — surviving arms balance, excluded users don't |\n| Primary metric `data: null` or `validation_failures` in all arms, exposures healthy | Metric machinery broken — measuring nothing while burning decision time |\n| Running experiment, zero exposures in 48h after a healthy baseline | Dormant — flag call removed from code, or upstream broke |\n| Running experiment, zero exposures ever, launched > 24h ago | Broken wiring — wrong SDK method, flag at 0%, custom exposure misconfigured |\n| Flag `filters` edited after `start_date` | Mid-run mutation — post-edit data may be contaminated |\n| Running far past `recommended_running_time` with flat exposure accumulation | Zombie — P3 recommendation to decide or end |\n| Stopped experiment, flag still active serving multiple variants weeks later | Lingering contamination + flag debt — P3 hygiene |\n| Ratio matches split, volume healthy, no recent flag edits | Baseline — leave it alone regardless of metric movement |\n\n### Explore\n\nPatterns to watch — starting points, not a checklist.\n\n#### Sample ratio mismatch (SRM)\n\nFor each running experiment launched > 24h ago, read `exposures.sample_ratio_mismatch.p_value` off `experiment-results-get` — PostHog runs the chi-squared itself (`$multiple` excluded). p \u003C 0.01 at healthy volume is the flag; cite the p-value and per-variant `total_exposures` vs the `expected` counts in the finding.\n\nTwo caveats before trusting a clean p-value:\n\n- It tests against the **current** configured split. If variants were redistributed mid-run, post-edit balance can look clean while pre-edit data is contaminated — check the flag history (below) whenever `feature_flag.version` is high.\n- It says nothing about `$multiple` — read `bias_risk.multiple_variant_percentage` as its own check (below).\n\nWhen the tool can't serve the experiment (legacy metrics) or you need to date an onset, fall back to the exposure SQL. Default exposure event:\n\n```sql\nSELECT\n properties.$feature_flag_response AS variant,\n count() AS exposures,\n count(DISTINCT person_id) AS persons\nFROM events\nWHERE event = '$feature_flag_called'\n AND properties.$feature_flag = '\u003Cflag-key>'\n AND timestamp >= toDateTime('\u003Cstart_date>', 'UTC')\nGROUP BY variant\nORDER BY exposures DESC\n```\n\nIf `exposure_criteria.exposure_event` is set, the experiment uses a custom exposure event — query that event name instead and read the variant from `properties.$feature\u002F\u003Cflag-key>` (a different property; the default's `$feature_flag_response` won't exist there).\n\nReading the output:\n\n- Rows with variant `false`, `''`, or null are evaluations that didn't bucket — exclude from the ratio, but note their share (a large share suggests release-condition issues).\n- The `$multiple` row is its own check (below) — exclude it from the ratio, matching PostHog's own SRM test.\n- **Sample-size gate:** per variant, the 2σ noise band on an expected share `p` with `n` total bucketed exposures is roughly `±2·sqrt(p·(1-p)\u002Fn)`. On 50\u002F50 that's ±7pp at n=200, ±2.2pp at n=2,000, ±0.7pp at n=20,000. Flag SRM only when the observed share sits **> 3σ** from expected — at 10k exposures, 53\u002F47 against a 50\u002F50 config clears that bar; at 300 exposures, 60\u002F40 doesn't. Below ~1,000 bucketed exposures total, don't call SRM at all; write a `pattern:` memory and recheck next run.\n\nA confirmed SRM is report-worthy on its own (the data is biased no matter the cause), but the finding lands much harder with a suspected cause. Cheap follow-ups: check `persons` vs `exposures` per variant (a high events-per-person skew in one variant suggests bots hashing to one bucket); check `feature-flags-activity-retrieve` for flag edits after launch (rebucketing); check whether the skew started at launch (wiring) or at a specific date (a change — find it in the activity log).\n\n#### `$multiple` contamination\n\nUsers counted under `$multiple` saw more than one variant — identity fragmentation (`identify()` after flag evaluation, `reset()` mid-session, cross-device), bootstrap vs `\u002Fdecide` disagreement, or a mid-run flag edit that rebucketed users. Read `bias_risk.multiple_variant_percentage` off `experiment-results-get`:\n\n- **> 0.5%** sustained — worth surfacing; with `multiple_variant_handling = \"exclude\"` (the default when `exposure_criteria` doesn't set it) these users are dropped, and on an **uneven** split the drop is asymmetric, biasing results (then even > 0.1% matters).\n- **Predictable mechanism check:** a flag with `bucketing_identifier: distinct_id` and `ensure_experience_continuity: false` on an experiment whose audience crosses an identity transition (new-user targeting, signup\u002Flogin flows) re-buckets every anonymous-to-identified user — `$multiple` grows steadily from day one, and the excluded users are non-randomly the exact population under study. Read both fields off `experiment-get`'s `feature_flag`; when this shape matches, the finding is strong even with clean SRM.\n- A sudden **step-change** in the `$multiple` timeseries dates a rebucketing event — cross-check `feature-flags-activity-retrieve {id: \u003Cfeature_flag_id>}` for a `filters` diff at that date. A variant zeroed mid-run with `parameters.excluded_variants` set is a deliberate arm-drop (a product feature), but it still rebuckets that arm's users — frame it as a deliberate change with statistical side effects, not a mystery mutation.\n- To dig into fragmentation: per-person variant counts —\n\n```sql\nSELECT person_id,\n count(DISTINCT properties.$feature_flag_response) AS variants_seen,\n count(DISTINCT distinct_id) AS distinct_ids\nFROM events\nWHERE event = '$feature_flag_called'\n AND properties.$feature_flag = '\u003Cflag-key>'\n AND properties.$feature_flag_response NOT IN ('$multiple', 'false', '')\n AND timestamp >= toDateTime('\u003Cstart_date>', 'UTC')\nGROUP BY person_id\nHAVING variants_seen > 1\nLIMIT 50\n```\n\n#### Metric machinery broken (not metric movement)\n\nVariant win\u002Floss is the team's call — but a metric that **cannot produce an answer** is a machinery fault, and the experiment burns calendar time measuring nothing. From `experiment-results-get`, with healthy exposures:\n\n- A primary metric row with `data: null` (its query failed) or `validation_failures` in **all** arms (e.g. baseline-mean-is-zero on a funnel whose conversion event never fires in control) — the headline result is unreadable.\n- A metric whose definition contradicts the stated hypothesis — the description names a condition (\"tagged with X\", \"for product Y\") the metric's event\u002Fproperties don't filter on, so the measured signal is dominated by unrelated traffic. Confirm with one SQL count comparing filtered vs unfiltered volume before claiming this.\n\nBoth are report-worthy: the team thinks they're collecting evidence and they aren't. A treatment-only conversion event legitimately reads ~zero in control — that's expected, not a fault (the control-arm `not-enough-metric-data` failure alone doesn't qualify).\n\n#### Exposure stall \u002F dormant experiment\n\nA running experiment should accrue exposures continuously. Read the per-variant `exposures.timeseries` off `experiment-results-get` (cumulative daily counts — a flat tail is the stall shape), or by SQL. **Query the experiment's actual exposure event**: default experiments use `$feature_flag_called`, but if `exposure_criteria.exposure_event` is set, query that event name instead (filtering on `properties.$feature\u002F\u003Cflag-key>` rather than `$feature_flag`) — running the default query against a custom-exposure experiment returns zero rows and fakes a stall:\n\n```sql\nSELECT toDate(timestamp) AS day, count() AS exposures\nFROM events\nWHERE event = '$feature_flag_called' -- or exposure_criteria.exposure_event\n AND properties.$feature_flag = '\u003Cflag-key>'\n AND timestamp >= toDateTime('\u003Cstart_date>', 'UTC')\nGROUP BY day ORDER BY day\n```\n\n- **Zero ever, launched > 24h ago** — broken wiring: the SDK method used doesn't record `$feature_flag_called` (bulk accessors like `getAllFlags()` don't), the flag is at 0% rollout or inactive, or a custom exposure event is missing its `$feature\u002F\u003Cflag-key>` property. Check `experiment-get`'s flag state before filing a report — a **paused** experiment (flag deactivated, status \"paused\") legitimately has no fresh exposures. And before diagnosing a custom-exposure experiment as dormant, confirm with both signals: the custom event by `$feature\u002F\u003Cflag-key>` **and** `$feature_flag_called` for the flag — if the flag is being called but the custom event never fires, the break is in the custom event wiring, not the experiment.\n- **Healthy baseline then a cliff to ~zero** — the flag-reading call was removed from code, or an upstream deploy broke the path. Date the cliff; cross-check `advanced-activity-logs-list` and `feature-flags-activity-retrieve` around it.\n- **Asymptotic plateau after weeks** (e.g. +4 exposures over 100 days) — the eligible audience is exhausted; the experiment is done recruiting. Fold into the zombie check.\n\n#### Mid-run flag mutation\n\n`feature-flags-activity-retrieve {id: \u003Cfeature_flag_id>}` returns the flag's edit history with diffs. Scan for changes **after** the experiment's `start_date`:\n\n- Variant `rollout_percentage` redistribution (e.g. 50\u002F50 → 70\u002F30) — rebuckets users, creates `$multiple`, biases everything after the edit. Report-worthy.\n- Overall rollout **decrease** — test users fall back to default UX; post-edit data is mixed. Worth surfacing. (Rollout **increase** is the one safe mid-run change — skip.)\n- Release-condition tightening, bucketing-key change, variant key rename — all rebucket.\n- `active` flips date pause\u002Fresume windows — context for stalls, usually deliberate.\n\nAlso `advanced-activity-logs-list {scopes: [\"Experiment\"], item_ids: [\u003Cid>]}` for experiment-level edits (exposure criteria swaps, metric changes near a decision point).\n\n#### Lifecycle drift (zombie \u002F decided \u002F lingering flags)\n\nCheap hygiene pass over the full list — P3 recommendations, not anomalies; bundle them into one finding rather than one per experiment:\n\n- **Zombie:** running well past its useful life — exposures far above `parameters.recommended_sample_size` (often the cleaner test; `recommended_running_time` can be 0\u002Fabsent), or > 60 days with a plateaued exposure curve. The data is as good as it will get; recommend deciding. For high-stakes calls, `experiment-timeseries-results` (needs `metric_uuid` + `fingerprint` from the experiment's `metrics` array) shows whether the primary metric has been stable for weeks — a sustained flat answer strengthens \"decide now\".\n- **Stopped but contaminating:** `end_date` set weeks ago, linked flag still `active` with a multivariate split (no variant shipped to 100%). Users still see random variants of a concluded test; recommend ship-variant or flag cleanup.\n- **Stale drafts:** drafts untouched > 30 days — lowest priority, mention only in a bundle, never alone.\n\n### Save memory as you go\n\nWrite a scratchpad entry whenever you observe something a future run should know. Encode the category in the key prefix — `pattern:`, `noise:`, `addressed:`, `dedupe:`, `report:`, `reviewer:`:\n\n- key `pattern:experiments:running-inventory` — _\"Running: `new-checkout` (id 42, flag `new-checkout`, 50\u002F50, launched 2026-05-20, ~1.2k exposures\u002Fday, default exposure event); `pricing-v2` (id 57, 33\u002F33\u002F33, launched 2026-06-01, custom exposure event `pricing_page_viewed`).\"_\n- key `pattern:experiments:new-checkout` — _\"Baseline ~1.2k exposures\u002Fday, observed split 50.3\u002F49.7 on 18k exposures at 2026-06-08, `$multiple` 0.2%. Healthy; recheck ratio only if volume or flag version changes.\"_\n- key `noise:experiments:pricing-v2-forced-ios` — _\"Flag has a forced-variant release condition (iOS → test) — deliberate per config; per-variant ratio will never match the nominal split. Don't call SRM on the aggregate; compare within the random cohort only.\"_\n- key `dedupe:experiments:42-srm` — _\"Filed SRM on `new-checkout` (id 42) 2026-06-09: 56\u002F44 on 22k exposures, started at flag v7 edit 2026-06-05. If still skewed next run, skip; if team reset\u002Frelaunched, watch the fresh data instead.\"_ One stable key per issue — update it in place, don't mint a dated variant.\n- key `addressed:experiments:31-zombie` — _\"Recommended ending `old-onboarding` (id 31, running 140 days) on 2026-05-15; team aware. Don't re-file unless it's still running in 30 days.\"_\n- key `report:experiments:new-checkout` — _\"Report `019f0a96-…` covers the `new-checkout` (id 42) SRM. Edit it (append_note the fresh numbers) while the skew persists and the report is still live; if it was resolved and the experiment later re-skews, that's a fresh report.\"_\n- key `reviewer:experiments:new-checkout` — _\"`new-checkout` owned by `alice` (GitHub login) — route its reports there.\"_\n\nBy run #5 you should know every running experiment's expected split, exposure baseline, exposure-event type, and which quirks are deliberate — so a real contradiction stands out immediately and cheaply.\n\n### Decide\n\nFor a candidate that clears the bar, the call is **edit an existing report, author a new one, remember, or skip** — use judgment, these are the rails:\n\n- **Search the inbox first.** The `report:experiments:\u003Cslug>` scratchpad pointer is the reliable path (it holds the `report_id` — `inbox-reports-retrieve` it directly); with no pointer, `inbox-reports-list` by the specific experiment name **and** flag key (`ordering=-updated_at`), not a broad word like `experiment` (which matches hundreds of unrelated UX reports).\n- **Edit** (`scout-edit-report`) when a still-live report already covers the same experiment issue — an SRM still skewed, a stall that hasn't recovered, a `$multiple` trend still climbing. `append_note` the fresh numbers, or rewrite the title\u002Fsummary on a report you authored. This is the default when a match exists. `edit-report` can't change status, so if the matched report is `resolved` \u002F `suppressed` \u002F `failed`, don't append (it won't resurface) — author a fresh report for the relapse and repoint the `report:` key.\n- **Author** (`scout-emit-report`) only when nothing live covers it. A good report names the experiment id and flag key, quantifies the contradiction (observed vs expected split with exposure counts, `$multiple` percentage, days dormant), passes the sample-size gate, and dates the onset — ideally tied to a flag version or activity-log entry. Set `priority` (P0–P4) + `priority_explanation` — validity threats on a live decision (SRM, mid-run mutation, contamination) are P2, stalls P2–P3 by blast radius, lifecycle hygiene P3; it's the report's importance in the inbox, your call to make. Set `suggested_reviewers` via `scout-members-list` (objects — a `{github_login}` or `{user_uuid}`, not bare strings; cache under `reviewer:experiments:\u003Cslug>`); left empty the report reaches no one. A validity threat is an investigation a human confirms, not a one-line change → `actionability=requires_human_input` and `repository=NO_REPO` (NO_REPO is what stops `priority`+reviewers from spawning a pointless repo-selection sandbox). After authoring, write the `report:experiments:\u003Cslug>` pointer with the `report_id` so the next run edits instead of duplicating.\n- **Remember** if below the bar but worth carrying forward (a ratio drifting but inside the noise band, `$multiple` creeping at 0.3%, a plateau that needs one more week); **skip** with a one-line note if a `noise:` \u002F `addressed:` \u002F `dedupe:` entry or an existing report already covers it.\n\nSibling scouts share memory — the feature-flags scout owns non-experiment flag wiring, and the generalist (which ran an experiment-integrity lens before this specialist existed) may hold `dedupe:general:experiment-*` scratchpad entries; honor them like your own. When a prior run already covered a topic, default to edit-or-skip: the same fact twice in the inbox costs more than missing one finding for one tick.\n\n### Close out\n\nSummarize the run in one paragraph: which experiments you checked, which reports you authored or edited, what you remembered, and what you ruled out. The harness saves it as the run summary; future runs read it via `scout-runs-list`. Don't write a separate \"run metadata\" scratchpad entry. \"All running experiments healthy\" is a real, useful outcome.\n\n## Disqualifiers (skip these)\n\n- **Launched \u003C 24h ago** — exposure precomputation lags ~15 min and day-one volume is unrepresentative; zero or skewed exposures right after launch are not findings yet.\n- **Ratio claims below the sample-size gate** — no SRM call under ~1,000 bucketed exposures, and never inside the 3σ band. Low-volume splits wobble; that's variance.\n- **Metric movement** — a variant winning, losing, or wobbling is the team's decision surface, not a scout finding. Only flag metric _machinery_ (validity), with one exception: a long-stable answer on a zombie feeds the \"decide now\" recommendation.\n- **Paused experiments with no fresh exposures** — that's what pause means. Check flag `active` before calling a stall.\n- **Rollout increases mid-run** — the safe change; new users enter cleanly.\n- **Forced-variant release conditions** (`filters.groups[].variant` set) — deliberate non-random assignment; aggregate ratios won't match the nominal split by design. Note it once in `noise:` memory.\n- **Declared A\u002FA, placebo, or engine-validation experiments** (name\u002Fdescription says A\u002FA, placebo, validation, identical variants) — long runtimes and null results are the point; skip lifecycle \"decide now\" nudges. SRM checks still fully apply — a skewed A\u002FA is exactly the kind of machinery fault these exist to catch. Note the intent once in `noise:` memory.\n- **Holdout-enrolled experiments** — the holdout slice shifts effective ratios; read `holdout_id` before judging a split.\n- **Bucketing failures** (`$feature_flag_response` = false\u002Fempty) counted as variants — exclude from ratios; only their _share_ trending up is interesting.\n- **Experiments already concluded with a conclusion set** — the team decided; lingering _flag_ state is the only thing left worth checking.\n\nWhen in doubt, write a memory entry instead of filing a report.\n\n## MCP tools\n\nDirect calls (read-only):\n\n- `experiment-list` — cheap candidate discovery: id, name, status (draft \u002F running \u002F paused \u002F stopped), dates, `feature_flag_key`. Filter by `status`; start here.\n- `experiment-results-get` — **the flagship detector**: exposure block (`total_exposures`, daily `timeseries`, native `sample_ratio_mismatch.p_value`, `bias_risk.multiple_variant_percentage`) plus per-metric `validation_failures` \u002F `data: null`. Heavy response with many metrics — read the exposure + validation fields, skip the per-metric stats. New-engine experiments only; pass `refresh: false`.\n- `experiment-get` — full config for a candidate: `parameters.feature_flag_variants` (configured split), `parameters.rollout_percentage`, `recommended_sample_size`, `parameters.excluded_variants`, `exposure_criteria` (custom `exposure_event`, `multiple_variant_handling`, `filterTestAccounts`), `stats_config.method`, `holdout_id`, linked `feature_flag` (active, `version`, `bucketing_identifier`, `ensure_experience_continuity`, `filters.groups[].variant` overrides), `metrics` (each with `uuid` + fingerprint). Large response — candidates only.\n- `experiment-stats` — project-wide velocity aggregate (launched \u002F completed last 30d, active count). Cheap context for the hygiene pass.\n- `experiment-timeseries-results` — day-by-day per-variant results for one metric (`metric_uuid` + `fingerprint` from the metrics array). Use sparingly, for the zombie \"decide now\" check.\n- `feature-flag-get-definition` \u002F `feature-flags-activity-retrieve` — flag state and edit-history diffs; the latter is how you date mid-run mutations.\n- `advanced-activity-logs-list` (`scopes: [\"Experiment\"]`) — experiment-level edit timeline.\n- `execute-sql` against `events` — exposure analysis. Properties: `$feature_flag` (flag key) + `$feature_flag_response` (variant, incl. `$multiple`) on `$feature_flag_called`; `$feature\u002F\u003Cflag-key>` on custom exposure events.\n- `read-data-schema` — confirm a custom exposure event and its properties exist before aggregating over them.\n\nInbox & reviewer routing:\n\n- `inbox-reports-list` \u002F `inbox-reports-retrieve` — the reports already in the inbox; check before authoring so you edit instead of duplicating (`ordering=-updated_at`).\n- `inbox-report-artefacts-list` — a comparable report's artefact log, where the routed `suggested_reviewers` live (the report record doesn't expose them) — reviewer precedent.\n- `scout-members-list` — this project's members with their resolved `github_login`, to route `suggested_reviewers` to an experiment's owner (wrap as a `{github_login}` object, or pass the member's `{user_uuid}` and let the server resolve; null `github_login` → try the next owner). The in-run roster; the org-scoped resolver tools aren't available in a scout run.\n\nHarness-level:\n\n- `scout-project-profile-get` \u002F `scout-scratchpad-search` \u002F `scout-runs-list` \u002F `scout-runs-retrieve` — orientation + dedupe.\n- `scout-emit-report` \u002F `scout-edit-report` — author a report \u002F edit an existing one (the report-channel contract is in the harness prompt).\n- `scout-scratchpad-remember` \u002F `scout-scratchpad-forget` — remember \u002F prune stale memory keys.\n\n## When to stop\n\n- No experiments in use → `not-in-use:` entry, close out empty.\n- All running experiments match their config (ratio in band, fresh exposures, no post-launch flag edits) → close out empty; refresh `pattern:` baselines if stale.\n- Candidates all gated by `noise:` \u002F `addressed:` \u002F `dedupe:` entries, or an existing inbox report → edit-or-skip and close out.\n- You've filed (or edited) reports for what's solid → close out. One sharp validity report beats a laundry list of P3 hygiene nits.\n\n\"Looked but found nothing meaningful\" is a real outcome.\n",{"data":42,"body":50},{"name":4,"description":6,"compatibility":43,"allowed_tools":44,"metadata":47},"PostHog Signals agent (Claude sandbox). Read-only analytics + signal_scout_internal:write (scratchpad) + signal_scout_report:write (report channel), plus the experiments, feature-flag, and analytics tools in the MCP tools section.\n",[45,46],"emit_report","edit_report",{"owner_team":48,"scope":49},"signals","experiments",{"type":51,"children":52},"root",[53,61,67,102,127,132,160,167,212,234,246,252,257,264,269,420,425,616,659,665,884,890,895,902,944,949,991,996,1099,1128,1133,1216,1245,1256,1305,1434,1525,1531,1550,1584,1597,1603,1659,1711,1815,1821,1844,1902,1915,1921,1926,2020,2026,2067,2268,2273,2279,2291,2579,2592,2598,2610,2616,2780,2785,2791,2796,3148,3153,3243,3248,3314,3320,3377,3382],{"type":54,"tag":55,"props":56,"children":57},"element","h1",{"id":4},[58],{"type":59,"value":60},"text","Signals scout: experiments",{"type":54,"tag":62,"props":63,"children":64},"p",{},[65],{"type":59,"value":66},"You are a focused experiments scout. An experiment's configuration is a set of promises — \"this is running\", \"traffic splits 50\u002F50\", \"the flag is active\", \"we'll decide when the data is in\" — and your job is to catch the moments the data stream breaks those promises:",{"type":54,"tag":68,"props":69,"children":70},"ol",{},[71,92],{"type":54,"tag":72,"props":73,"children":74},"li",{},[75,81,83,90],{"type":54,"tag":76,"props":77,"children":78},"strong",{},[79],{"type":59,"value":80},"Validity threats",{"type":59,"value":82}," on running experiments — sample ratio mismatch (SRM), elevated ",{"type":54,"tag":84,"props":85,"children":87},"code",{"className":86},[],[88],{"type":59,"value":89},"$multiple",{"type":59,"value":91}," contamination, exposure stalls, mid-run flag edits that rebucket users, and metrics that structurally cannot answer the hypothesis (unreadable in all arms, or missing the filter the hypothesis implies). These silently corrupt the team's decision data.",{"type":54,"tag":72,"props":93,"children":94},{},[95,100],{"type":54,"tag":76,"props":96,"children":97},{},[98],{"type":59,"value":99},"Lifecycle drift",{"type":59,"value":101}," — experiments running long past their useful life, experiments with a clear sustained answer still collecting data, ended experiments whose flags still serve multiple variants.",{"type":54,"tag":62,"props":103,"children":104},{},[105,110,112,118,120,125],{"type":54,"tag":76,"props":106,"children":107},{},[108],{"type":59,"value":109},"Config-vs-data contradiction is the signal-vs-noise discriminator.",{"type":59,"value":111}," A running experiment whose exposures match its configured split at healthy volume is baseline — no matter which variant is winning (metric ",{"type":54,"tag":113,"props":114,"children":115},"em",{},[116],{"type":59,"value":117},"movement",{"type":59,"value":119}," is the team's call, not yours). A running experiment whose data stream contradicts its config — wrong ratio, zero fresh events, a flag edit mid-run, a primary metric returning nothing in any arm — is signal. Internalize that shape: you are auditing the ",{"type":54,"tag":113,"props":121,"children":122},{},[123],{"type":59,"value":124},"measurement machinery",{"type":59,"value":126},", not second-guessing the results.",{"type":54,"tag":62,"props":128,"children":129},{},[130],{"type":59,"value":131},"Validity findings are time-sensitive: every day an SRM goes unnoticed is a day of biased data the team may ship a decision on. But statistics wobble at low volume — a 60\u002F40 split on 200 exposures is noise, not SRM. When in doubt, write memory instead of filing a report.",{"type":54,"tag":62,"props":133,"children":134},{},[135,137,143,145,151,153,158],{"type":59,"value":136},"You author reports directly via the report channel (",{"type":54,"tag":84,"props":138,"children":140},{"className":139},[],[141],{"type":59,"value":142},"scout-emit-report",{"type":59,"value":144}," \u002F ",{"type":54,"tag":84,"props":146,"children":148},{"className":147},[],[149],{"type":59,"value":150},"scout-edit-report",{"type":59,"value":152},"): you've done the research, so you own each report 1:1 end-to-end rather than firing weak signals for a pipeline to cluster. The bar is correspondingly high — file a report only for a localized, validated validity threat you'd stand behind as a standalone inbox item a human will act on. A threat the inbox already covers (an SRM that's still skewed, a stall that hasn't recovered, a zombie bundle that only grew) is an ",{"type":54,"tag":76,"props":154,"children":155},{},[156],{"type":59,"value":157},"edit",{"type":59,"value":159},", not a new report. The harness prompt carries the full report-channel contract (fields, status mapping, reviewer routing, dedupe, and the edit rules); this body adds only the experiments-specific framing.",{"type":54,"tag":161,"props":162,"children":164},"h2",{"id":163},"quick-close-out-are-experiments-even-active",[165],{"type":59,"value":166},"Quick close-out: are experiments even active?",{"type":54,"tag":62,"props":168,"children":169},{},[170,172,178,180,186,188,194,196,202,204,210],{"type":59,"value":171},"Read ",{"type":54,"tag":84,"props":173,"children":175},{"className":174},[],[176],{"type":59,"value":177},"recent_experiments",{"type":59,"value":179}," off ",{"type":54,"tag":84,"props":181,"children":183},{"className":182},[],[184],{"type":59,"value":185},"scout-project-profile-get",{"type":59,"value":187},". If ",{"type":54,"tag":84,"props":189,"children":191},{"className":190},[],[192],{"type":59,"value":193},"running_count",{"type":59,"value":195}," is 0 and ",{"type":54,"tag":84,"props":197,"children":199},{"className":198},[],[200],{"type":59,"value":201},"total_count",{"type":59,"value":203}," is 0 (or all entries are old drafts\u002Farchived with no ",{"type":54,"tag":84,"props":205,"children":207},{"className":206},[],[208],{"type":59,"value":209},"updated_at",{"type":59,"value":211}," activity in 30 days), experiments aren't in play here. Write one scratchpad entry:",{"type":54,"tag":213,"props":214,"children":215},"ul",{},[216,229],{"type":54,"tag":72,"props":217,"children":218},{},[219,221,227],{"type":59,"value":220},"key: ",{"type":54,"tag":84,"props":222,"children":224},{"className":223},[],[225],{"type":59,"value":226},"not-in-use:experiments",{"type":59,"value":228}," (the scratchpad is already team-scoped — no id in the key)",{"type":54,"tag":72,"props":230,"children":231},{},[232],{"type":59,"value":233},"content: brief note (\"checked at {timestamp}, no running experiments, {total_count} total, latest activity {date}\")",{"type":54,"tag":62,"props":235,"children":236},{},[237,239,244],{"type":59,"value":238},"Close out empty. Re-running with the same key idempotently refreshes the timestamp. If ",{"type":54,"tag":84,"props":240,"children":242},{"className":241},[],[243],{"type":59,"value":193},{"type":59,"value":245}," is 0 but there are recent drafts or recent stops, do the cheap lifecycle-hygiene pass (stale drafts, contaminating flags) before closing out — skip the exposure analysis entirely.",{"type":54,"tag":161,"props":247,"children":249},{"id":248},"how-a-run-works",[250],{"type":59,"value":251},"How a run works",{"type":54,"tag":62,"props":253,"children":254},{},[255],{"type":59,"value":256},"Cycle between these moves; skip what's not useful.",{"type":54,"tag":258,"props":259,"children":261},"h3",{"id":260},"get-oriented",[262],{"type":59,"value":263},"Get oriented",{"type":54,"tag":62,"props":265,"children":266},{},[267],{"type":59,"value":268},"Three cheap reads cold-start a run:",{"type":54,"tag":213,"props":270,"children":271},{},[272,328,339,364],{"type":54,"tag":72,"props":273,"children":274},{},[275,281,283,289,291,297,298,304,305,311,313,319,320,326],{"type":54,"tag":84,"props":276,"children":278},{"className":277},[],[279],{"type":59,"value":280},"scout-scratchpad-search",{"type":59,"value":282}," (",{"type":54,"tag":84,"props":284,"children":286},{"className":285},[],[287],{"type":59,"value":288},"text=experiment",{"type":59,"value":290},") — durable steering: known running experiments and their expected splits, established baselines, ",{"type":54,"tag":84,"props":292,"children":294},{"className":293},[],[295],{"type":59,"value":296},"noise:",{"type":59,"value":144},{"type":54,"tag":84,"props":299,"children":301},{"className":300},[],[302],{"type":59,"value":303},"addressed:",{"type":59,"value":144},{"type":54,"tag":84,"props":306,"children":308},{"className":307},[],[309],{"type":59,"value":310},"dedupe:",{"type":59,"value":312}," entries gating re-reports, plus ",{"type":54,"tag":84,"props":314,"children":316},{"className":315},[],[317],{"type":59,"value":318},"report:",{"type":59,"value":144},{"type":54,"tag":84,"props":321,"children":323},{"className":322},[],[324],{"type":59,"value":325},"reviewer:",{"type":59,"value":327}," entries pointing at the open report for an experiment and who owns it.",{"type":54,"tag":72,"props":329,"children":330},{},[331,337],{"type":54,"tag":84,"props":332,"children":334},{"className":333},[],[335],{"type":59,"value":336},"scout-runs-list",{"type":59,"value":338}," (last 7d) — what prior experiments runs found and ruled out.",{"type":54,"tag":72,"props":340,"children":341},{},[342,347,349,354,356,362],{"type":54,"tag":84,"props":343,"children":345},{"className":344},[],[346],{"type":59,"value":185},{"type":59,"value":348}," — ",{"type":54,"tag":84,"props":350,"children":352},{"className":351},[],[353],{"type":59,"value":177},{"type":59,"value":355}," (running count, recent ids, feature flag keys) and ",{"type":54,"tag":84,"props":357,"children":359},{"className":358},[],[360],{"type":59,"value":361},"recent_feature_flags",{"type":59,"value":363}," for cross-referencing.",{"type":54,"tag":72,"props":365,"children":366},{},[367,373,374,380,382,388,390,394,396,402,404,410,412,418],{"type":54,"tag":84,"props":368,"children":370},{"className":369},[],[371],{"type":59,"value":372},"inbox-reports-list",{"type":59,"value":282},{"type":54,"tag":84,"props":375,"children":377},{"className":376},[],[378],{"type":59,"value":379},"search",{"type":59,"value":381},"=experiment name or flag key, ",{"type":54,"tag":84,"props":383,"children":385},{"className":384},[],[386],{"type":59,"value":387},"ordering=-updated_at",{"type":59,"value":389},") — the reports already in the inbox. A validity threat on an experiment you've reported before is an ",{"type":54,"tag":76,"props":391,"children":392},{},[393],{"type":59,"value":157},{"type":59,"value":395},", not a fresh report; pull the closest matches with ",{"type":54,"tag":84,"props":397,"children":399},{"className":398},[],[400],{"type":59,"value":401},"inbox-reports-retrieve",{"type":59,"value":403}," before authoring. Your own report-channel reports persist their backing signals under ",{"type":54,"tag":84,"props":405,"children":407},{"className":406},[],[408],{"type":59,"value":409},"source_product=signals_scout",{"type":59,"value":411},", so don't filter ",{"type":54,"tag":84,"props":413,"children":415},{"className":414},[],[416],{"type":59,"value":417},"source_product=experiments",{"type":59,"value":419}," — you'd miss every report you authored.",{"type":54,"tag":62,"props":421,"children":422},{},[423],{"type":59,"value":424},"Then orient on experiments specifically:",{"type":54,"tag":68,"props":426,"children":427},{},[428,462,536],{"type":54,"tag":72,"props":429,"children":430},{},[431,437,439,445,447,453,455,460],{"type":54,"tag":84,"props":432,"children":434},{"className":433},[],[435],{"type":59,"value":436},"experiment-list {\"status\": \"running\", \"order\": \"-start_date\"}",{"type":59,"value":438}," — cheap: returns id, name, status, dates, ",{"type":54,"tag":84,"props":440,"children":442},{"className":441},[],[443],{"type":59,"value":444},"feature_flag_key",{"type":59,"value":446}," per experiment. Also grab ",{"type":54,"tag":84,"props":448,"children":450},{"className":449},[],[451],{"type":59,"value":452},"{\"status\": \"draft\"}",{"type":59,"value":454}," and recently stopped ones if doing the hygiene pass. ",{"type":54,"tag":76,"props":456,"children":457},{},[458],{"type":59,"value":459},"Triage before going deep:",{"type":59,"value":461}," on mature projects the \"running\" list is often dominated by forgotten experiments (launched years ago, throwaway names). Reserve the per-experiment exposure analysis for the validity-watch set — experiments launched in the last ~90 days or known-active from scratchpad memory (cap ~10 per run; rotate if more). Older running experiments go straight to the zombie bundle without exposure SQL.",{"type":54,"tag":72,"props":463,"children":464},{},[465,471,473,479,481,487,489,495,497,503,505,511,512,518,520,526,528,534],{"type":54,"tag":84,"props":466,"children":468},{"className":467},[],[469],{"type":59,"value":470},"experiment-get {id}",{"type":59,"value":472}," on running candidates only — you need ",{"type":54,"tag":84,"props":474,"children":476},{"className":475},[],[477],{"type":59,"value":478},"parameters.feature_flag_variants",{"type":59,"value":480}," (the configured split), ",{"type":54,"tag":84,"props":482,"children":484},{"className":483},[],[485],{"type":59,"value":486},"parameters.rollout_percentage",{"type":59,"value":488},", ",{"type":54,"tag":84,"props":490,"children":492},{"className":491},[],[493],{"type":59,"value":494},"exposure_criteria",{"type":59,"value":496}," (custom exposure event? ",{"type":54,"tag":84,"props":498,"children":500},{"className":499},[],[501],{"type":59,"value":502},"multiple_variant_handling",{"type":59,"value":504},"?), ",{"type":54,"tag":84,"props":506,"children":508},{"className":507},[],[509],{"type":59,"value":510},"parameters.recommended_running_time",{"type":59,"value":488},{"type":54,"tag":84,"props":513,"children":515},{"className":514},[],[516],{"type":59,"value":517},"stats_config.method",{"type":59,"value":519},", and the linked ",{"type":54,"tag":84,"props":521,"children":523},{"className":522},[],[524],{"type":59,"value":525},"feature_flag",{"type":59,"value":527}," (active state, ",{"type":54,"tag":84,"props":529,"children":531},{"className":530},[],[532],{"type":59,"value":533},"filters.groups[].variant",{"type":59,"value":535}," forced-variant overrides). The full object is large (metrics arrays, flag filters) — never bulk-fetch every experiment; running experiments only, and lean on scratchpad memory for ones you've profiled before.",{"type":54,"tag":72,"props":537,"children":538},{},[539,545,547,553,555,561,563,569,571,577,579,585,586,592,594,599,601,607,608,614],{"type":54,"tag":84,"props":540,"children":542},{"className":541},[],[543],{"type":59,"value":544},"experiment-results-get {id, refresh: false}",{"type":59,"value":546}," per candidate — the flagship detector. One call returns the exposure block (",{"type":54,"tag":84,"props":548,"children":550},{"className":549},[],[551],{"type":59,"value":552},"total_exposures",{"type":59,"value":554}," per variant, daily ",{"type":54,"tag":84,"props":556,"children":558},{"className":557},[],[559],{"type":59,"value":560},"timeseries",{"type":59,"value":562},", a native chi-squared ",{"type":54,"tag":84,"props":564,"children":566},{"className":565},[],[567],{"type":59,"value":568},"sample_ratio_mismatch.p_value",{"type":59,"value":570}," and ",{"type":54,"tag":84,"props":572,"children":574},{"className":573},[],[575],{"type":59,"value":576},"bias_risk.multiple_variant_percentage",{"type":59,"value":578},") plus per-metric results with ",{"type":54,"tag":84,"props":580,"children":582},{"className":581},[],[583],{"type":59,"value":584},"validation_failures",{"type":59,"value":570},{"type":54,"tag":84,"props":587,"children":589},{"className":588},[],[590],{"type":59,"value":591},"data: null",{"type":59,"value":593}," markers for failed metric queries. Read the exposure block and validation fields; ",{"type":54,"tag":76,"props":595,"children":596},{},[597],{"type":59,"value":598},"skip the per-metric stats",{"type":59,"value":600}," (movement is not your business) — with many metrics the response is heavy. Legacy experiments (",{"type":54,"tag":84,"props":602,"children":604},{"className":603},[],[605],{"type":59,"value":606},"ExperimentTrendsQuery",{"type":59,"value":144},{"type":54,"tag":84,"props":609,"children":611},{"className":610},[],[612],{"type":59,"value":613},"ExperimentFunnelsQuery",{"type":59,"value":615}," metrics) aren't supported by this tool — fall back to the exposure SQL below.",{"type":54,"tag":62,"props":617,"children":618},{},[619,621,627,629,634,636,641,643,649,651,657],{"type":59,"value":620},"Drop to ",{"type":54,"tag":84,"props":622,"children":624},{"className":623},[],[625],{"type":59,"value":626},"execute-sql",{"type":59,"value":628}," only for diagnosis: dating an onset, per-person fragmentation, custom-exposure drill-downs. ",{"type":54,"tag":76,"props":630,"children":631},{},[632],{"type":59,"value":633},"Timezone footgun:",{"type":59,"value":635}," HogQL string timestamp literals parse in the ",{"type":54,"tag":113,"props":637,"children":638},{},[639],{"type":59,"value":640},"project",{"type":59,"value":642}," timezone, not UTC — a UTC ",{"type":54,"tag":84,"props":644,"children":646},{"className":645},[],[647],{"type":59,"value":648},"start_date",{"type":59,"value":650}," literal can shift the window by hours and fake a dormant experiment. Use ",{"type":54,"tag":84,"props":652,"children":654},{"className":653},[],[655],{"type":59,"value":656},"now() - INTERVAL N DAY",{"type":59,"value":658}," for recency windows.",{"type":54,"tag":258,"props":660,"children":662},{"id":661},"profile-shape-config-vs-data",[663],{"type":59,"value":664},"Profile shape — config vs data",{"type":54,"tag":666,"props":667,"children":668},"table",{},[669,688],{"type":54,"tag":670,"props":671,"children":672},"thead",{},[673],{"type":54,"tag":674,"props":675,"children":676},"tr",{},[677,683],{"type":54,"tag":678,"props":679,"children":680},"th",{},[681],{"type":59,"value":682},"Pattern",{"type":54,"tag":678,"props":684,"children":685},{},[686],{"type":59,"value":687},"What it usually means",{"type":54,"tag":689,"props":690,"children":691},"tbody",{},[692,711,737,758,785,798,811,837,858,871],{"type":54,"tag":674,"props":693,"children":694},{},[695,706],{"type":54,"tag":696,"props":697,"children":698},"td",{},[699,704],{"type":54,"tag":84,"props":700,"children":702},{"className":701},[],[703],{"type":59,"value":568},{"type":59,"value":705}," \u003C 0.01 at healthy volume",{"type":54,"tag":696,"props":707,"children":708},{},[709],{"type":59,"value":710},"SRM — investigate first; this is the flagship finding",{"type":54,"tag":674,"props":712,"children":713},{},[714,732],{"type":54,"tag":696,"props":715,"children":716},{},[717,722,724,730],{"type":54,"tag":84,"props":718,"children":720},{"className":719},[],[721],{"type":59,"value":89},{"type":59,"value":723}," share > 0.5% of exposures (or > 0.1% with an uneven split + ",{"type":54,"tag":84,"props":725,"children":727},{"className":726},[],[728],{"type":59,"value":729},"exclude",{"type":59,"value":731},")",{"type":54,"tag":696,"props":733,"children":734},{},[735],{"type":59,"value":736},"Identity fragmentation or mid-run rebucketing — contamination",{"type":54,"tag":674,"props":738,"children":739},{},[740,753],{"type":54,"tag":696,"props":741,"children":742},{},[743,745,751],{"type":59,"value":744},"SRM clean but ",{"type":54,"tag":84,"props":746,"children":748},{"className":747},[],[749],{"type":59,"value":750},"multiple_variant_percentage",{"type":59,"value":752}," high",{"type":54,"tag":696,"props":754,"children":755},{},[756],{"type":59,"value":757},"The failure SRM alone misses — surviving arms balance, excluded users don't",{"type":54,"tag":674,"props":759,"children":760},{},[761,780],{"type":54,"tag":696,"props":762,"children":763},{},[764,766,771,773,778],{"type":59,"value":765},"Primary metric ",{"type":54,"tag":84,"props":767,"children":769},{"className":768},[],[770],{"type":59,"value":591},{"type":59,"value":772}," or ",{"type":54,"tag":84,"props":774,"children":776},{"className":775},[],[777],{"type":59,"value":584},{"type":59,"value":779}," in all arms, exposures healthy",{"type":54,"tag":696,"props":781,"children":782},{},[783],{"type":59,"value":784},"Metric machinery broken — measuring nothing while burning decision time",{"type":54,"tag":674,"props":786,"children":787},{},[788,793],{"type":54,"tag":696,"props":789,"children":790},{},[791],{"type":59,"value":792},"Running experiment, zero exposures in 48h after a healthy baseline",{"type":54,"tag":696,"props":794,"children":795},{},[796],{"type":59,"value":797},"Dormant — flag call removed from code, or upstream broke",{"type":54,"tag":674,"props":799,"children":800},{},[801,806],{"type":54,"tag":696,"props":802,"children":803},{},[804],{"type":59,"value":805},"Running experiment, zero exposures ever, launched > 24h ago",{"type":54,"tag":696,"props":807,"children":808},{},[809],{"type":59,"value":810},"Broken wiring — wrong SDK method, flag at 0%, custom exposure misconfigured",{"type":54,"tag":674,"props":812,"children":813},{},[814,832],{"type":54,"tag":696,"props":815,"children":816},{},[817,819,825,827],{"type":59,"value":818},"Flag ",{"type":54,"tag":84,"props":820,"children":822},{"className":821},[],[823],{"type":59,"value":824},"filters",{"type":59,"value":826}," edited after ",{"type":54,"tag":84,"props":828,"children":830},{"className":829},[],[831],{"type":59,"value":648},{"type":54,"tag":696,"props":833,"children":834},{},[835],{"type":59,"value":836},"Mid-run mutation — post-edit data may be contaminated",{"type":54,"tag":674,"props":838,"children":839},{},[840,853],{"type":54,"tag":696,"props":841,"children":842},{},[843,845,851],{"type":59,"value":844},"Running far past ",{"type":54,"tag":84,"props":846,"children":848},{"className":847},[],[849],{"type":59,"value":850},"recommended_running_time",{"type":59,"value":852}," with flat exposure accumulation",{"type":54,"tag":696,"props":854,"children":855},{},[856],{"type":59,"value":857},"Zombie — P3 recommendation to decide or end",{"type":54,"tag":674,"props":859,"children":860},{},[861,866],{"type":54,"tag":696,"props":862,"children":863},{},[864],{"type":59,"value":865},"Stopped experiment, flag still active serving multiple variants weeks later",{"type":54,"tag":696,"props":867,"children":868},{},[869],{"type":59,"value":870},"Lingering contamination + flag debt — P3 hygiene",{"type":54,"tag":674,"props":872,"children":873},{},[874,879],{"type":54,"tag":696,"props":875,"children":876},{},[877],{"type":59,"value":878},"Ratio matches split, volume healthy, no recent flag edits",{"type":54,"tag":696,"props":880,"children":881},{},[882],{"type":59,"value":883},"Baseline — leave it alone regardless of metric movement",{"type":54,"tag":258,"props":885,"children":887},{"id":886},"explore",[888],{"type":59,"value":889},"Explore",{"type":54,"tag":62,"props":891,"children":892},{},[893],{"type":59,"value":894},"Patterns to watch — starting points, not a checklist.",{"type":54,"tag":896,"props":897,"children":899},"h4",{"id":898},"sample-ratio-mismatch-srm",[900],{"type":59,"value":901},"Sample ratio mismatch (SRM)",{"type":54,"tag":62,"props":903,"children":904},{},[905,907,913,914,920,922,927,929,934,936,942],{"type":59,"value":906},"For each running experiment launched > 24h ago, read ",{"type":54,"tag":84,"props":908,"children":910},{"className":909},[],[911],{"type":59,"value":912},"exposures.sample_ratio_mismatch.p_value",{"type":59,"value":179},{"type":54,"tag":84,"props":915,"children":917},{"className":916},[],[918],{"type":59,"value":919},"experiment-results-get",{"type":59,"value":921}," — PostHog runs the chi-squared itself (",{"type":54,"tag":84,"props":923,"children":925},{"className":924},[],[926],{"type":59,"value":89},{"type":59,"value":928}," excluded). p \u003C 0.01 at healthy volume is the flag; cite the p-value and per-variant ",{"type":54,"tag":84,"props":930,"children":932},{"className":931},[],[933],{"type":59,"value":552},{"type":59,"value":935}," vs the ",{"type":54,"tag":84,"props":937,"children":939},{"className":938},[],[940],{"type":59,"value":941},"expected",{"type":59,"value":943}," counts in the finding.",{"type":54,"tag":62,"props":945,"children":946},{},[947],{"type":59,"value":948},"Two caveats before trusting a clean p-value:",{"type":54,"tag":213,"props":950,"children":951},{},[952,972],{"type":54,"tag":72,"props":953,"children":954},{},[955,957,962,964,970],{"type":59,"value":956},"It tests against the ",{"type":54,"tag":76,"props":958,"children":959},{},[960],{"type":59,"value":961},"current",{"type":59,"value":963}," configured split. If variants were redistributed mid-run, post-edit balance can look clean while pre-edit data is contaminated — check the flag history (below) whenever ",{"type":54,"tag":84,"props":965,"children":967},{"className":966},[],[968],{"type":59,"value":969},"feature_flag.version",{"type":59,"value":971}," is high.",{"type":54,"tag":72,"props":973,"children":974},{},[975,977,982,984,989],{"type":59,"value":976},"It says nothing about ",{"type":54,"tag":84,"props":978,"children":980},{"className":979},[],[981],{"type":59,"value":89},{"type":59,"value":983}," — read ",{"type":54,"tag":84,"props":985,"children":987},{"className":986},[],[988],{"type":59,"value":576},{"type":59,"value":990}," as its own check (below).",{"type":54,"tag":62,"props":992,"children":993},{},[994],{"type":59,"value":995},"When the tool can't serve the experiment (legacy metrics) or you need to date an onset, fall back to the exposure SQL. Default exposure event:",{"type":54,"tag":997,"props":998,"children":1003},"pre",{"className":999,"code":1000,"language":1001,"meta":1002,"style":1002},"language-sql shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","SELECT\n properties.$feature_flag_response AS variant,\n count() AS exposures,\n count(DISTINCT person_id) AS persons\nFROM events\nWHERE event = '$feature_flag_called'\n AND properties.$feature_flag = '\u003Cflag-key>'\n AND timestamp >= toDateTime('\u003Cstart_date>', 'UTC')\nGROUP BY variant\nORDER BY exposures DESC\n","sql","",[1004],{"type":54,"tag":84,"props":1005,"children":1006},{"__ignoreMap":1002},[1007,1018,1027,1036,1045,1054,1063,1072,1081,1090],{"type":54,"tag":1008,"props":1009,"children":1012},"span",{"class":1010,"line":1011},"line",1,[1013],{"type":54,"tag":1008,"props":1014,"children":1015},{},[1016],{"type":59,"value":1017},"SELECT\n",{"type":54,"tag":1008,"props":1019,"children":1021},{"class":1010,"line":1020},2,[1022],{"type":54,"tag":1008,"props":1023,"children":1024},{},[1025],{"type":59,"value":1026}," properties.$feature_flag_response AS variant,\n",{"type":54,"tag":1008,"props":1028,"children":1030},{"class":1010,"line":1029},3,[1031],{"type":54,"tag":1008,"props":1032,"children":1033},{},[1034],{"type":59,"value":1035}," count() AS exposures,\n",{"type":54,"tag":1008,"props":1037,"children":1039},{"class":1010,"line":1038},4,[1040],{"type":54,"tag":1008,"props":1041,"children":1042},{},[1043],{"type":59,"value":1044}," count(DISTINCT person_id) AS persons\n",{"type":54,"tag":1008,"props":1046,"children":1048},{"class":1010,"line":1047},5,[1049],{"type":54,"tag":1008,"props":1050,"children":1051},{},[1052],{"type":59,"value":1053},"FROM events\n",{"type":54,"tag":1008,"props":1055,"children":1057},{"class":1010,"line":1056},6,[1058],{"type":54,"tag":1008,"props":1059,"children":1060},{},[1061],{"type":59,"value":1062},"WHERE event = '$feature_flag_called'\n",{"type":54,"tag":1008,"props":1064,"children":1066},{"class":1010,"line":1065},7,[1067],{"type":54,"tag":1008,"props":1068,"children":1069},{},[1070],{"type":59,"value":1071}," AND properties.$feature_flag = '\u003Cflag-key>'\n",{"type":54,"tag":1008,"props":1073,"children":1075},{"class":1010,"line":1074},8,[1076],{"type":54,"tag":1008,"props":1077,"children":1078},{},[1079],{"type":59,"value":1080}," AND timestamp >= toDateTime('\u003Cstart_date>', 'UTC')\n",{"type":54,"tag":1008,"props":1082,"children":1084},{"class":1010,"line":1083},9,[1085],{"type":54,"tag":1008,"props":1086,"children":1087},{},[1088],{"type":59,"value":1089},"GROUP BY variant\n",{"type":54,"tag":1008,"props":1091,"children":1093},{"class":1010,"line":1092},10,[1094],{"type":54,"tag":1008,"props":1095,"children":1096},{},[1097],{"type":59,"value":1098},"ORDER BY exposures DESC\n",{"type":54,"tag":62,"props":1100,"children":1101},{},[1102,1104,1110,1112,1118,1120,1126],{"type":59,"value":1103},"If ",{"type":54,"tag":84,"props":1105,"children":1107},{"className":1106},[],[1108],{"type":59,"value":1109},"exposure_criteria.exposure_event",{"type":59,"value":1111}," is set, the experiment uses a custom exposure event — query that event name instead and read the variant from ",{"type":54,"tag":84,"props":1113,"children":1115},{"className":1114},[],[1116],{"type":59,"value":1117},"properties.$feature\u002F\u003Cflag-key>",{"type":59,"value":1119}," (a different property; the default's ",{"type":54,"tag":84,"props":1121,"children":1123},{"className":1122},[],[1124],{"type":59,"value":1125},"$feature_flag_response",{"type":59,"value":1127}," won't exist there).",{"type":54,"tag":62,"props":1129,"children":1130},{},[1131],{"type":59,"value":1132},"Reading the output:",{"type":54,"tag":213,"props":1134,"children":1135},{},[1136,1156,1168],{"type":54,"tag":72,"props":1137,"children":1138},{},[1139,1141,1147,1148,1154],{"type":59,"value":1140},"Rows with variant ",{"type":54,"tag":84,"props":1142,"children":1144},{"className":1143},[],[1145],{"type":59,"value":1146},"false",{"type":59,"value":488},{"type":54,"tag":84,"props":1149,"children":1151},{"className":1150},[],[1152],{"type":59,"value":1153},"''",{"type":59,"value":1155},", or null are evaluations that didn't bucket — exclude from the ratio, but note their share (a large share suggests release-condition issues).",{"type":54,"tag":72,"props":1157,"children":1158},{},[1159,1161,1166],{"type":59,"value":1160},"The ",{"type":54,"tag":84,"props":1162,"children":1164},{"className":1163},[],[1165],{"type":59,"value":89},{"type":59,"value":1167}," row is its own check (below) — exclude it from the ratio, matching PostHog's own SRM test.",{"type":54,"tag":72,"props":1169,"children":1170},{},[1171,1176,1178,1183,1185,1191,1193,1199,1201,1206,1208,1214],{"type":54,"tag":76,"props":1172,"children":1173},{},[1174],{"type":59,"value":1175},"Sample-size gate:",{"type":59,"value":1177}," per variant, the 2σ noise band on an expected share ",{"type":54,"tag":84,"props":1179,"children":1181},{"className":1180},[],[1182],{"type":59,"value":62},{"type":59,"value":1184}," with ",{"type":54,"tag":84,"props":1186,"children":1188},{"className":1187},[],[1189],{"type":59,"value":1190},"n",{"type":59,"value":1192}," total bucketed exposures is roughly ",{"type":54,"tag":84,"props":1194,"children":1196},{"className":1195},[],[1197],{"type":59,"value":1198},"±2·sqrt(p·(1-p)\u002Fn)",{"type":59,"value":1200},". On 50\u002F50 that's ±7pp at n=200, ±2.2pp at n=2,000, ±0.7pp at n=20,000. Flag SRM only when the observed share sits ",{"type":54,"tag":76,"props":1202,"children":1203},{},[1204],{"type":59,"value":1205},"> 3σ",{"type":59,"value":1207}," from expected — at 10k exposures, 53\u002F47 against a 50\u002F50 config clears that bar; at 300 exposures, 60\u002F40 doesn't. Below ~1,000 bucketed exposures total, don't call SRM at all; write a ",{"type":54,"tag":84,"props":1209,"children":1211},{"className":1210},[],[1212],{"type":59,"value":1213},"pattern:",{"type":59,"value":1215}," memory and recheck next run.",{"type":54,"tag":62,"props":1217,"children":1218},{},[1219,1221,1227,1229,1235,1237,1243],{"type":59,"value":1220},"A confirmed SRM is report-worthy on its own (the data is biased no matter the cause), but the finding lands much harder with a suspected cause. Cheap follow-ups: check ",{"type":54,"tag":84,"props":1222,"children":1224},{"className":1223},[],[1225],{"type":59,"value":1226},"persons",{"type":59,"value":1228}," vs ",{"type":54,"tag":84,"props":1230,"children":1232},{"className":1231},[],[1233],{"type":59,"value":1234},"exposures",{"type":59,"value":1236}," per variant (a high events-per-person skew in one variant suggests bots hashing to one bucket); check ",{"type":54,"tag":84,"props":1238,"children":1240},{"className":1239},[],[1241],{"type":59,"value":1242},"feature-flags-activity-retrieve",{"type":59,"value":1244}," for flag edits after launch (rebucketing); check whether the skew started at launch (wiring) or at a specific date (a change — find it in the activity log).",{"type":54,"tag":896,"props":1246,"children":1248},{"id":1247},"multiple-contamination",[1249,1254],{"type":54,"tag":84,"props":1250,"children":1252},{"className":1251},[],[1253],{"type":59,"value":89},{"type":59,"value":1255}," contamination",{"type":54,"tag":62,"props":1257,"children":1258},{},[1259,1261,1266,1268,1274,1276,1282,1284,1290,1292,1297,1298,1303],{"type":59,"value":1260},"Users counted under ",{"type":54,"tag":84,"props":1262,"children":1264},{"className":1263},[],[1265],{"type":59,"value":89},{"type":59,"value":1267}," saw more than one variant — identity fragmentation (",{"type":54,"tag":84,"props":1269,"children":1271},{"className":1270},[],[1272],{"type":59,"value":1273},"identify()",{"type":59,"value":1275}," after flag evaluation, ",{"type":54,"tag":84,"props":1277,"children":1279},{"className":1278},[],[1280],{"type":59,"value":1281},"reset()",{"type":59,"value":1283}," mid-session, cross-device), bootstrap vs ",{"type":54,"tag":84,"props":1285,"children":1287},{"className":1286},[],[1288],{"type":59,"value":1289},"\u002Fdecide",{"type":59,"value":1291}," disagreement, or a mid-run flag edit that rebucketed users. Read ",{"type":54,"tag":84,"props":1293,"children":1295},{"className":1294},[],[1296],{"type":59,"value":576},{"type":59,"value":179},{"type":54,"tag":84,"props":1299,"children":1301},{"className":1300},[],[1302],{"type":59,"value":919},{"type":59,"value":1304},":",{"type":54,"tag":213,"props":1306,"children":1307},{},[1308,1340,1387,1429],{"type":54,"tag":72,"props":1309,"children":1310},{},[1311,1316,1318,1324,1326,1331,1333,1338],{"type":54,"tag":76,"props":1312,"children":1313},{},[1314],{"type":59,"value":1315},"> 0.5%",{"type":59,"value":1317}," sustained — worth surfacing; with ",{"type":54,"tag":84,"props":1319,"children":1321},{"className":1320},[],[1322],{"type":59,"value":1323},"multiple_variant_handling = \"exclude\"",{"type":59,"value":1325}," (the default when ",{"type":54,"tag":84,"props":1327,"children":1329},{"className":1328},[],[1330],{"type":59,"value":494},{"type":59,"value":1332}," doesn't set it) these users are dropped, and on an ",{"type":54,"tag":76,"props":1334,"children":1335},{},[1336],{"type":59,"value":1337},"uneven",{"type":59,"value":1339}," split the drop is asymmetric, biasing results (then even > 0.1% matters).",{"type":54,"tag":72,"props":1341,"children":1342},{},[1343,1348,1350,1356,1357,1363,1365,1370,1372,1378,1380,1385],{"type":54,"tag":76,"props":1344,"children":1345},{},[1346],{"type":59,"value":1347},"Predictable mechanism check:",{"type":59,"value":1349}," a flag with ",{"type":54,"tag":84,"props":1351,"children":1353},{"className":1352},[],[1354],{"type":59,"value":1355},"bucketing_identifier: distinct_id",{"type":59,"value":570},{"type":54,"tag":84,"props":1358,"children":1360},{"className":1359},[],[1361],{"type":59,"value":1362},"ensure_experience_continuity: false",{"type":59,"value":1364}," on an experiment whose audience crosses an identity transition (new-user targeting, signup\u002Flogin flows) re-buckets every anonymous-to-identified user — ",{"type":54,"tag":84,"props":1366,"children":1368},{"className":1367},[],[1369],{"type":59,"value":89},{"type":59,"value":1371}," grows steadily from day one, and the excluded users are non-randomly the exact population under study. Read both fields off ",{"type":54,"tag":84,"props":1373,"children":1375},{"className":1374},[],[1376],{"type":59,"value":1377},"experiment-get",{"type":59,"value":1379},"'s ",{"type":54,"tag":84,"props":1381,"children":1383},{"className":1382},[],[1384],{"type":59,"value":525},{"type":59,"value":1386},"; when this shape matches, the finding is strong even with clean SRM.",{"type":54,"tag":72,"props":1388,"children":1389},{},[1390,1392,1397,1399,1404,1406,1412,1414,1419,1421,1427],{"type":59,"value":1391},"A sudden ",{"type":54,"tag":76,"props":1393,"children":1394},{},[1395],{"type":59,"value":1396},"step-change",{"type":59,"value":1398}," in the ",{"type":54,"tag":84,"props":1400,"children":1402},{"className":1401},[],[1403],{"type":59,"value":89},{"type":59,"value":1405}," timeseries dates a rebucketing event — cross-check ",{"type":54,"tag":84,"props":1407,"children":1409},{"className":1408},[],[1410],{"type":59,"value":1411},"feature-flags-activity-retrieve {id: \u003Cfeature_flag_id>}",{"type":59,"value":1413}," for a ",{"type":54,"tag":84,"props":1415,"children":1417},{"className":1416},[],[1418],{"type":59,"value":824},{"type":59,"value":1420}," diff at that date. A variant zeroed mid-run with ",{"type":54,"tag":84,"props":1422,"children":1424},{"className":1423},[],[1425],{"type":59,"value":1426},"parameters.excluded_variants",{"type":59,"value":1428}," set is a deliberate arm-drop (a product feature), but it still rebuckets that arm's users — frame it as a deliberate change with statistical side effects, not a mystery mutation.",{"type":54,"tag":72,"props":1430,"children":1431},{},[1432],{"type":59,"value":1433},"To dig into fragmentation: per-person variant counts —",{"type":54,"tag":997,"props":1435,"children":1437},{"className":999,"code":1436,"language":1001,"meta":1002,"style":1002},"SELECT person_id,\n count(DISTINCT properties.$feature_flag_response) AS variants_seen,\n count(DISTINCT distinct_id) AS distinct_ids\nFROM events\nWHERE event = '$feature_flag_called'\n AND properties.$feature_flag = '\u003Cflag-key>'\n AND properties.$feature_flag_response NOT IN ('$multiple', 'false', '')\n AND timestamp >= toDateTime('\u003Cstart_date>', 'UTC')\nGROUP BY person_id\nHAVING variants_seen > 1\nLIMIT 50\n",[1438],{"type":54,"tag":84,"props":1439,"children":1440},{"__ignoreMap":1002},[1441,1449,1457,1465,1472,1479,1486,1494,1501,1509,1517],{"type":54,"tag":1008,"props":1442,"children":1443},{"class":1010,"line":1011},[1444],{"type":54,"tag":1008,"props":1445,"children":1446},{},[1447],{"type":59,"value":1448},"SELECT person_id,\n",{"type":54,"tag":1008,"props":1450,"children":1451},{"class":1010,"line":1020},[1452],{"type":54,"tag":1008,"props":1453,"children":1454},{},[1455],{"type":59,"value":1456}," count(DISTINCT properties.$feature_flag_response) AS variants_seen,\n",{"type":54,"tag":1008,"props":1458,"children":1459},{"class":1010,"line":1029},[1460],{"type":54,"tag":1008,"props":1461,"children":1462},{},[1463],{"type":59,"value":1464}," count(DISTINCT distinct_id) AS distinct_ids\n",{"type":54,"tag":1008,"props":1466,"children":1467},{"class":1010,"line":1038},[1468],{"type":54,"tag":1008,"props":1469,"children":1470},{},[1471],{"type":59,"value":1053},{"type":54,"tag":1008,"props":1473,"children":1474},{"class":1010,"line":1047},[1475],{"type":54,"tag":1008,"props":1476,"children":1477},{},[1478],{"type":59,"value":1062},{"type":54,"tag":1008,"props":1480,"children":1481},{"class":1010,"line":1056},[1482],{"type":54,"tag":1008,"props":1483,"children":1484},{},[1485],{"type":59,"value":1071},{"type":54,"tag":1008,"props":1487,"children":1488},{"class":1010,"line":1065},[1489],{"type":54,"tag":1008,"props":1490,"children":1491},{},[1492],{"type":59,"value":1493}," AND properties.$feature_flag_response NOT IN ('$multiple', 'false', '')\n",{"type":54,"tag":1008,"props":1495,"children":1496},{"class":1010,"line":1074},[1497],{"type":54,"tag":1008,"props":1498,"children":1499},{},[1500],{"type":59,"value":1080},{"type":54,"tag":1008,"props":1502,"children":1503},{"class":1010,"line":1083},[1504],{"type":54,"tag":1008,"props":1505,"children":1506},{},[1507],{"type":59,"value":1508},"GROUP BY person_id\n",{"type":54,"tag":1008,"props":1510,"children":1511},{"class":1010,"line":1092},[1512],{"type":54,"tag":1008,"props":1513,"children":1514},{},[1515],{"type":59,"value":1516},"HAVING variants_seen > 1\n",{"type":54,"tag":1008,"props":1518,"children":1519},{"class":1010,"line":30},[1520],{"type":54,"tag":1008,"props":1521,"children":1522},{},[1523],{"type":59,"value":1524},"LIMIT 50\n",{"type":54,"tag":896,"props":1526,"children":1528},{"id":1527},"metric-machinery-broken-not-metric-movement",[1529],{"type":59,"value":1530},"Metric machinery broken (not metric movement)",{"type":54,"tag":62,"props":1532,"children":1533},{},[1534,1536,1541,1543,1548],{"type":59,"value":1535},"Variant win\u002Floss is the team's call — but a metric that ",{"type":54,"tag":76,"props":1537,"children":1538},{},[1539],{"type":59,"value":1540},"cannot produce an answer",{"type":59,"value":1542}," is a machinery fault, and the experiment burns calendar time measuring nothing. From ",{"type":54,"tag":84,"props":1544,"children":1546},{"className":1545},[],[1547],{"type":59,"value":919},{"type":59,"value":1549},", with healthy exposures:",{"type":54,"tag":213,"props":1551,"children":1552},{},[1553,1579],{"type":54,"tag":72,"props":1554,"children":1555},{},[1556,1558,1563,1565,1570,1572,1577],{"type":59,"value":1557},"A primary metric row with ",{"type":54,"tag":84,"props":1559,"children":1561},{"className":1560},[],[1562],{"type":59,"value":591},{"type":59,"value":1564}," (its query failed) or ",{"type":54,"tag":84,"props":1566,"children":1568},{"className":1567},[],[1569],{"type":59,"value":584},{"type":59,"value":1571}," in ",{"type":54,"tag":76,"props":1573,"children":1574},{},[1575],{"type":59,"value":1576},"all",{"type":59,"value":1578}," arms (e.g. baseline-mean-is-zero on a funnel whose conversion event never fires in control) — the headline result is unreadable.",{"type":54,"tag":72,"props":1580,"children":1581},{},[1582],{"type":59,"value":1583},"A metric whose definition contradicts the stated hypothesis — the description names a condition (\"tagged with X\", \"for product Y\") the metric's event\u002Fproperties don't filter on, so the measured signal is dominated by unrelated traffic. Confirm with one SQL count comparing filtered vs unfiltered volume before claiming this.",{"type":54,"tag":62,"props":1585,"children":1586},{},[1587,1589,1595],{"type":59,"value":1588},"Both are report-worthy: the team thinks they're collecting evidence and they aren't. A treatment-only conversion event legitimately reads ~zero in control — that's expected, not a fault (the control-arm ",{"type":54,"tag":84,"props":1590,"children":1592},{"className":1591},[],[1593],{"type":59,"value":1594},"not-enough-metric-data",{"type":59,"value":1596}," failure alone doesn't qualify).",{"type":54,"tag":896,"props":1598,"children":1600},{"id":1599},"exposure-stall-dormant-experiment",[1601],{"type":59,"value":1602},"Exposure stall \u002F dormant experiment",{"type":54,"tag":62,"props":1604,"children":1605},{},[1606,1608,1614,1615,1620,1622,1627,1629,1635,1637,1642,1644,1649,1651,1657],{"type":59,"value":1607},"A running experiment should accrue exposures continuously. Read the per-variant ",{"type":54,"tag":84,"props":1609,"children":1611},{"className":1610},[],[1612],{"type":59,"value":1613},"exposures.timeseries",{"type":59,"value":179},{"type":54,"tag":84,"props":1616,"children":1618},{"className":1617},[],[1619],{"type":59,"value":919},{"type":59,"value":1621}," (cumulative daily counts — a flat tail is the stall shape), or by SQL. ",{"type":54,"tag":76,"props":1623,"children":1624},{},[1625],{"type":59,"value":1626},"Query the experiment's actual exposure event",{"type":59,"value":1628},": default experiments use ",{"type":54,"tag":84,"props":1630,"children":1632},{"className":1631},[],[1633],{"type":59,"value":1634},"$feature_flag_called",{"type":59,"value":1636},", but if ",{"type":54,"tag":84,"props":1638,"children":1640},{"className":1639},[],[1641],{"type":59,"value":1109},{"type":59,"value":1643}," is set, query that event name instead (filtering on ",{"type":54,"tag":84,"props":1645,"children":1647},{"className":1646},[],[1648],{"type":59,"value":1117},{"type":59,"value":1650}," rather than ",{"type":54,"tag":84,"props":1652,"children":1654},{"className":1653},[],[1655],{"type":59,"value":1656},"$feature_flag",{"type":59,"value":1658},") — running the default query against a custom-exposure experiment returns zero rows and fakes a stall:",{"type":54,"tag":997,"props":1660,"children":1662},{"className":999,"code":1661,"language":1001,"meta":1002,"style":1002},"SELECT toDate(timestamp) AS day, count() AS exposures\nFROM events\nWHERE event = '$feature_flag_called' -- or exposure_criteria.exposure_event\n AND properties.$feature_flag = '\u003Cflag-key>'\n AND timestamp >= toDateTime('\u003Cstart_date>', 'UTC')\nGROUP BY day ORDER BY day\n",[1663],{"type":54,"tag":84,"props":1664,"children":1665},{"__ignoreMap":1002},[1666,1674,1681,1689,1696,1703],{"type":54,"tag":1008,"props":1667,"children":1668},{"class":1010,"line":1011},[1669],{"type":54,"tag":1008,"props":1670,"children":1671},{},[1672],{"type":59,"value":1673},"SELECT toDate(timestamp) AS day, count() AS exposures\n",{"type":54,"tag":1008,"props":1675,"children":1676},{"class":1010,"line":1020},[1677],{"type":54,"tag":1008,"props":1678,"children":1679},{},[1680],{"type":59,"value":1053},{"type":54,"tag":1008,"props":1682,"children":1683},{"class":1010,"line":1029},[1684],{"type":54,"tag":1008,"props":1685,"children":1686},{},[1687],{"type":59,"value":1688},"WHERE event = '$feature_flag_called' -- or exposure_criteria.exposure_event\n",{"type":54,"tag":1008,"props":1690,"children":1691},{"class":1010,"line":1038},[1692],{"type":54,"tag":1008,"props":1693,"children":1694},{},[1695],{"type":59,"value":1071},{"type":54,"tag":1008,"props":1697,"children":1698},{"class":1010,"line":1047},[1699],{"type":54,"tag":1008,"props":1700,"children":1701},{},[1702],{"type":59,"value":1080},{"type":54,"tag":1008,"props":1704,"children":1705},{"class":1010,"line":1056},[1706],{"type":54,"tag":1008,"props":1707,"children":1708},{},[1709],{"type":59,"value":1710},"GROUP BY day ORDER BY day\n",{"type":54,"tag":213,"props":1712,"children":1713},{},[1714,1781,1805],{"type":54,"tag":72,"props":1715,"children":1716},{},[1717,1722,1724,1729,1731,1737,1739,1745,1747,1752,1754,1759,1761,1766,1768,1773,1774,1779],{"type":54,"tag":76,"props":1718,"children":1719},{},[1720],{"type":59,"value":1721},"Zero ever, launched > 24h ago",{"type":59,"value":1723}," — broken wiring: the SDK method used doesn't record ",{"type":54,"tag":84,"props":1725,"children":1727},{"className":1726},[],[1728],{"type":59,"value":1634},{"type":59,"value":1730}," (bulk accessors like ",{"type":54,"tag":84,"props":1732,"children":1734},{"className":1733},[],[1735],{"type":59,"value":1736},"getAllFlags()",{"type":59,"value":1738}," don't), the flag is at 0% rollout or inactive, or a custom exposure event is missing its ",{"type":54,"tag":84,"props":1740,"children":1742},{"className":1741},[],[1743],{"type":59,"value":1744},"$feature\u002F\u003Cflag-key>",{"type":59,"value":1746}," property. Check ",{"type":54,"tag":84,"props":1748,"children":1750},{"className":1749},[],[1751],{"type":59,"value":1377},{"type":59,"value":1753},"'s flag state before filing a report — a ",{"type":54,"tag":76,"props":1755,"children":1756},{},[1757],{"type":59,"value":1758},"paused",{"type":59,"value":1760}," experiment (flag deactivated, status \"paused\") legitimately has no fresh exposures. And before diagnosing a custom-exposure experiment as dormant, confirm with both signals: the custom event by ",{"type":54,"tag":84,"props":1762,"children":1764},{"className":1763},[],[1765],{"type":59,"value":1744},{"type":59,"value":1767}," ",{"type":54,"tag":76,"props":1769,"children":1770},{},[1771],{"type":59,"value":1772},"and",{"type":59,"value":1767},{"type":54,"tag":84,"props":1775,"children":1777},{"className":1776},[],[1778],{"type":59,"value":1634},{"type":59,"value":1780}," for the flag — if the flag is being called but the custom event never fires, the break is in the custom event wiring, not the experiment.",{"type":54,"tag":72,"props":1782,"children":1783},{},[1784,1789,1791,1797,1798,1803],{"type":54,"tag":76,"props":1785,"children":1786},{},[1787],{"type":59,"value":1788},"Healthy baseline then a cliff to ~zero",{"type":59,"value":1790}," — the flag-reading call was removed from code, or an upstream deploy broke the path. Date the cliff; cross-check ",{"type":54,"tag":84,"props":1792,"children":1794},{"className":1793},[],[1795],{"type":59,"value":1796},"advanced-activity-logs-list",{"type":59,"value":570},{"type":54,"tag":84,"props":1799,"children":1801},{"className":1800},[],[1802],{"type":59,"value":1242},{"type":59,"value":1804}," around it.",{"type":54,"tag":72,"props":1806,"children":1807},{},[1808,1813],{"type":54,"tag":76,"props":1809,"children":1810},{},[1811],{"type":59,"value":1812},"Asymptotic plateau after weeks",{"type":59,"value":1814}," (e.g. +4 exposures over 100 days) — the eligible audience is exhausted; the experiment is done recruiting. Fold into the zombie check.",{"type":54,"tag":896,"props":1816,"children":1818},{"id":1817},"mid-run-flag-mutation",[1819],{"type":59,"value":1820},"Mid-run flag mutation",{"type":54,"tag":62,"props":1822,"children":1823},{},[1824,1829,1831,1836,1838,1843],{"type":54,"tag":84,"props":1825,"children":1827},{"className":1826},[],[1828],{"type":59,"value":1411},{"type":59,"value":1830}," returns the flag's edit history with diffs. Scan for changes ",{"type":54,"tag":76,"props":1832,"children":1833},{},[1834],{"type":59,"value":1835},"after",{"type":59,"value":1837}," the experiment's ",{"type":54,"tag":84,"props":1839,"children":1841},{"className":1840},[],[1842],{"type":59,"value":648},{"type":59,"value":1304},{"type":54,"tag":213,"props":1845,"children":1846},{},[1847,1867,1886,1891],{"type":54,"tag":72,"props":1848,"children":1849},{},[1850,1852,1858,1860,1865],{"type":59,"value":1851},"Variant ",{"type":54,"tag":84,"props":1853,"children":1855},{"className":1854},[],[1856],{"type":59,"value":1857},"rollout_percentage",{"type":59,"value":1859}," redistribution (e.g. 50\u002F50 → 70\u002F30) — rebuckets users, creates ",{"type":54,"tag":84,"props":1861,"children":1863},{"className":1862},[],[1864],{"type":59,"value":89},{"type":59,"value":1866},", biases everything after the edit. Report-worthy.",{"type":54,"tag":72,"props":1868,"children":1869},{},[1870,1872,1877,1879,1884],{"type":59,"value":1871},"Overall rollout ",{"type":54,"tag":76,"props":1873,"children":1874},{},[1875],{"type":59,"value":1876},"decrease",{"type":59,"value":1878}," — test users fall back to default UX; post-edit data is mixed. Worth surfacing. (Rollout ",{"type":54,"tag":76,"props":1880,"children":1881},{},[1882],{"type":59,"value":1883},"increase",{"type":59,"value":1885}," is the one safe mid-run change — skip.)",{"type":54,"tag":72,"props":1887,"children":1888},{},[1889],{"type":59,"value":1890},"Release-condition tightening, bucketing-key change, variant key rename — all rebucket.",{"type":54,"tag":72,"props":1892,"children":1893},{},[1894,1900],{"type":54,"tag":84,"props":1895,"children":1897},{"className":1896},[],[1898],{"type":59,"value":1899},"active",{"type":59,"value":1901}," flips date pause\u002Fresume windows — context for stalls, usually deliberate.",{"type":54,"tag":62,"props":1903,"children":1904},{},[1905,1907,1913],{"type":59,"value":1906},"Also ",{"type":54,"tag":84,"props":1908,"children":1910},{"className":1909},[],[1911],{"type":59,"value":1912},"advanced-activity-logs-list {scopes: [\"Experiment\"], item_ids: [\u003Cid>]}",{"type":59,"value":1914}," for experiment-level edits (exposure criteria swaps, metric changes near a decision point).",{"type":54,"tag":896,"props":1916,"children":1918},{"id":1917},"lifecycle-drift-zombie-decided-lingering-flags",[1919],{"type":59,"value":1920},"Lifecycle drift (zombie \u002F decided \u002F lingering flags)",{"type":54,"tag":62,"props":1922,"children":1923},{},[1924],{"type":59,"value":1925},"Cheap hygiene pass over the full list — P3 recommendations, not anomalies; bundle them into one finding rather than one per experiment:",{"type":54,"tag":213,"props":1927,"children":1928},{},[1929,1986,2010],{"type":54,"tag":72,"props":1930,"children":1931},{},[1932,1937,1939,1945,1947,1952,1954,1960,1962,1968,1970,1976,1978,1984],{"type":54,"tag":76,"props":1933,"children":1934},{},[1935],{"type":59,"value":1936},"Zombie:",{"type":59,"value":1938}," running well past its useful life — exposures far above ",{"type":54,"tag":84,"props":1940,"children":1942},{"className":1941},[],[1943],{"type":59,"value":1944},"parameters.recommended_sample_size",{"type":59,"value":1946}," (often the cleaner test; ",{"type":54,"tag":84,"props":1948,"children":1950},{"className":1949},[],[1951],{"type":59,"value":850},{"type":59,"value":1953}," can be 0\u002Fabsent), or > 60 days with a plateaued exposure curve. The data is as good as it will get; recommend deciding. For high-stakes calls, ",{"type":54,"tag":84,"props":1955,"children":1957},{"className":1956},[],[1958],{"type":59,"value":1959},"experiment-timeseries-results",{"type":59,"value":1961}," (needs ",{"type":54,"tag":84,"props":1963,"children":1965},{"className":1964},[],[1966],{"type":59,"value":1967},"metric_uuid",{"type":59,"value":1969}," + ",{"type":54,"tag":84,"props":1971,"children":1973},{"className":1972},[],[1974],{"type":59,"value":1975},"fingerprint",{"type":59,"value":1977}," from the experiment's ",{"type":54,"tag":84,"props":1979,"children":1981},{"className":1980},[],[1982],{"type":59,"value":1983},"metrics",{"type":59,"value":1985}," array) shows whether the primary metric has been stable for weeks — a sustained flat answer strengthens \"decide now\".",{"type":54,"tag":72,"props":1987,"children":1988},{},[1989,1994,1995,2001,2003,2008],{"type":54,"tag":76,"props":1990,"children":1991},{},[1992],{"type":59,"value":1993},"Stopped but contaminating:",{"type":59,"value":1767},{"type":54,"tag":84,"props":1996,"children":1998},{"className":1997},[],[1999],{"type":59,"value":2000},"end_date",{"type":59,"value":2002}," set weeks ago, linked flag still ",{"type":54,"tag":84,"props":2004,"children":2006},{"className":2005},[],[2007],{"type":59,"value":1899},{"type":59,"value":2009}," with a multivariate split (no variant shipped to 100%). Users still see random variants of a concluded test; recommend ship-variant or flag cleanup.",{"type":54,"tag":72,"props":2011,"children":2012},{},[2013,2018],{"type":54,"tag":76,"props":2014,"children":2015},{},[2016],{"type":59,"value":2017},"Stale drafts:",{"type":59,"value":2019}," drafts untouched > 30 days — lowest priority, mention only in a bundle, never alone.",{"type":54,"tag":258,"props":2021,"children":2023},{"id":2022},"save-memory-as-you-go",[2024],{"type":59,"value":2025},"Save memory as you go",{"type":54,"tag":62,"props":2027,"children":2028},{},[2029,2031,2036,2037,2042,2043,2048,2049,2054,2055,2060,2061,2066],{"type":59,"value":2030},"Write a scratchpad entry whenever you observe something a future run should know. Encode the category in the key prefix — ",{"type":54,"tag":84,"props":2032,"children":2034},{"className":2033},[],[2035],{"type":59,"value":1213},{"type":59,"value":488},{"type":54,"tag":84,"props":2038,"children":2040},{"className":2039},[],[2041],{"type":59,"value":296},{"type":59,"value":488},{"type":54,"tag":84,"props":2044,"children":2046},{"className":2045},[],[2047],{"type":59,"value":303},{"type":59,"value":488},{"type":54,"tag":84,"props":2050,"children":2052},{"className":2051},[],[2053],{"type":59,"value":310},{"type":59,"value":488},{"type":54,"tag":84,"props":2056,"children":2058},{"className":2057},[],[2059],{"type":59,"value":318},{"type":59,"value":488},{"type":54,"tag":84,"props":2062,"children":2064},{"className":2063},[],[2065],{"type":59,"value":325},{"type":59,"value":1304},{"type":54,"tag":213,"props":2068,"children":2069},{},[2070,2118,2141,2157,2182,2206,2237],{"type":54,"tag":72,"props":2071,"children":2072},{},[2073,2075,2081,2082],{"type":59,"value":2074},"key ",{"type":54,"tag":84,"props":2076,"children":2078},{"className":2077},[],[2079],{"type":59,"value":2080},"pattern:experiments:running-inventory",{"type":59,"value":348},{"type":54,"tag":113,"props":2083,"children":2084},{},[2085,2087,2093,2095,2100,2102,2108,2110,2116],{"type":59,"value":2086},"\"Running: ",{"type":54,"tag":84,"props":2088,"children":2090},{"className":2089},[],[2091],{"type":59,"value":2092},"new-checkout",{"type":59,"value":2094}," (id 42, flag ",{"type":54,"tag":84,"props":2096,"children":2098},{"className":2097},[],[2099],{"type":59,"value":2092},{"type":59,"value":2101},", 50\u002F50, launched 2026-05-20, ~1.2k exposures\u002Fday, default exposure event); ",{"type":54,"tag":84,"props":2103,"children":2105},{"className":2104},[],[2106],{"type":59,"value":2107},"pricing-v2",{"type":59,"value":2109}," (id 57, 33\u002F33\u002F33, launched 2026-06-01, custom exposure event ",{"type":54,"tag":84,"props":2111,"children":2113},{"className":2112},[],[2114],{"type":59,"value":2115},"pricing_page_viewed",{"type":59,"value":2117},").\"",{"type":54,"tag":72,"props":2119,"children":2120},{},[2121,2122,2128,2129],{"type":59,"value":2074},{"type":54,"tag":84,"props":2123,"children":2125},{"className":2124},[],[2126],{"type":59,"value":2127},"pattern:experiments:new-checkout",{"type":59,"value":348},{"type":54,"tag":113,"props":2130,"children":2131},{},[2132,2134,2139],{"type":59,"value":2133},"\"Baseline ~1.2k exposures\u002Fday, observed split 50.3\u002F49.7 on 18k exposures at 2026-06-08, ",{"type":54,"tag":84,"props":2135,"children":2137},{"className":2136},[],[2138],{"type":59,"value":89},{"type":59,"value":2140}," 0.2%. Healthy; recheck ratio only if volume or flag version changes.\"",{"type":54,"tag":72,"props":2142,"children":2143},{},[2144,2145,2151,2152],{"type":59,"value":2074},{"type":54,"tag":84,"props":2146,"children":2148},{"className":2147},[],[2149],{"type":59,"value":2150},"noise:experiments:pricing-v2-forced-ios",{"type":59,"value":348},{"type":54,"tag":113,"props":2153,"children":2154},{},[2155],{"type":59,"value":2156},"\"Flag has a forced-variant release condition (iOS → test) — deliberate per config; per-variant ratio will never match the nominal split. Don't call SRM on the aggregate; compare within the random cohort only.\"",{"type":54,"tag":72,"props":2158,"children":2159},{},[2160,2161,2167,2168,2180],{"type":59,"value":2074},{"type":54,"tag":84,"props":2162,"children":2164},{"className":2163},[],[2165],{"type":59,"value":2166},"dedupe:experiments:42-srm",{"type":59,"value":348},{"type":54,"tag":113,"props":2169,"children":2170},{},[2171,2173,2178],{"type":59,"value":2172},"\"Filed SRM on ",{"type":54,"tag":84,"props":2174,"children":2176},{"className":2175},[],[2177],{"type":59,"value":2092},{"type":59,"value":2179}," (id 42) 2026-06-09: 56\u002F44 on 22k exposures, started at flag v7 edit 2026-06-05. If still skewed next run, skip; if team reset\u002Frelaunched, watch the fresh data instead.\"",{"type":59,"value":2181}," One stable key per issue — update it in place, don't mint a dated variant.",{"type":54,"tag":72,"props":2183,"children":2184},{},[2185,2186,2192,2193],{"type":59,"value":2074},{"type":54,"tag":84,"props":2187,"children":2189},{"className":2188},[],[2190],{"type":59,"value":2191},"addressed:experiments:31-zombie",{"type":59,"value":348},{"type":54,"tag":113,"props":2194,"children":2195},{},[2196,2198,2204],{"type":59,"value":2197},"\"Recommended ending ",{"type":54,"tag":84,"props":2199,"children":2201},{"className":2200},[],[2202],{"type":59,"value":2203},"old-onboarding",{"type":59,"value":2205}," (id 31, running 140 days) on 2026-05-15; team aware. Don't re-file unless it's still running in 30 days.\"",{"type":54,"tag":72,"props":2207,"children":2208},{},[2209,2210,2216,2217],{"type":59,"value":2074},{"type":54,"tag":84,"props":2211,"children":2213},{"className":2212},[],[2214],{"type":59,"value":2215},"report:experiments:new-checkout",{"type":59,"value":348},{"type":54,"tag":113,"props":2218,"children":2219},{},[2220,2222,2228,2230,2235],{"type":59,"value":2221},"\"Report ",{"type":54,"tag":84,"props":2223,"children":2225},{"className":2224},[],[2226],{"type":59,"value":2227},"019f0a96-…",{"type":59,"value":2229}," covers the ",{"type":54,"tag":84,"props":2231,"children":2233},{"className":2232},[],[2234],{"type":59,"value":2092},{"type":59,"value":2236}," (id 42) SRM. Edit it (append_note the fresh numbers) while the skew persists and the report is still live; if it was resolved and the experiment later re-skews, that's a fresh report.\"",{"type":54,"tag":72,"props":2238,"children":2239},{},[2240,2241,2247,2248],{"type":59,"value":2074},{"type":54,"tag":84,"props":2242,"children":2244},{"className":2243},[],[2245],{"type":59,"value":2246},"reviewer:experiments:new-checkout",{"type":59,"value":348},{"type":54,"tag":113,"props":2249,"children":2250},{},[2251,2253,2258,2260,2266],{"type":59,"value":2252},"\"",{"type":54,"tag":84,"props":2254,"children":2256},{"className":2255},[],[2257],{"type":59,"value":2092},{"type":59,"value":2259}," owned by ",{"type":54,"tag":84,"props":2261,"children":2263},{"className":2262},[],[2264],{"type":59,"value":2265},"alice",{"type":59,"value":2267}," (GitHub login) — route its reports there.\"",{"type":54,"tag":62,"props":2269,"children":2270},{},[2271],{"type":59,"value":2272},"By run #5 you should know every running experiment's expected split, exposure baseline, exposure-event type, and which quirks are deliberate — so a real contradiction stands out immediately and cheaply.",{"type":54,"tag":258,"props":2274,"children":2276},{"id":2275},"decide",[2277],{"type":59,"value":2278},"Decide",{"type":54,"tag":62,"props":2280,"children":2281},{},[2282,2284,2289],{"type":59,"value":2283},"For a candidate that clears the bar, the call is ",{"type":54,"tag":76,"props":2285,"children":2286},{},[2287],{"type":59,"value":2288},"edit an existing report, author a new one, remember, or skip",{"type":59,"value":2290}," — use judgment, these are the rails:",{"type":54,"tag":213,"props":2292,"children":2293},{},[2294,2354,2422,2536],{"type":54,"tag":72,"props":2295,"children":2296},{},[2297,2302,2304,2310,2312,2318,2319,2324,2326,2331,2333,2337,2339,2344,2346,2352],{"type":54,"tag":76,"props":2298,"children":2299},{},[2300],{"type":59,"value":2301},"Search the inbox first.",{"type":59,"value":2303}," The ",{"type":54,"tag":84,"props":2305,"children":2307},{"className":2306},[],[2308],{"type":59,"value":2309},"report:experiments:\u003Cslug>",{"type":59,"value":2311}," scratchpad pointer is the reliable path (it holds the ",{"type":54,"tag":84,"props":2313,"children":2315},{"className":2314},[],[2316],{"type":59,"value":2317},"report_id",{"type":59,"value":348},{"type":54,"tag":84,"props":2320,"children":2322},{"className":2321},[],[2323],{"type":59,"value":401},{"type":59,"value":2325}," it directly); with no pointer, ",{"type":54,"tag":84,"props":2327,"children":2329},{"className":2328},[],[2330],{"type":59,"value":372},{"type":59,"value":2332}," by the specific experiment name ",{"type":54,"tag":76,"props":2334,"children":2335},{},[2336],{"type":59,"value":1772},{"type":59,"value":2338}," flag key (",{"type":54,"tag":84,"props":2340,"children":2342},{"className":2341},[],[2343],{"type":59,"value":387},{"type":59,"value":2345},"), not a broad word like ",{"type":54,"tag":84,"props":2347,"children":2349},{"className":2348},[],[2350],{"type":59,"value":2351},"experiment",{"type":59,"value":2353}," (which matches hundreds of unrelated UX reports).",{"type":54,"tag":72,"props":2355,"children":2356},{},[2357,2362,2363,2368,2370,2375,2377,2383,2385,2391,2393,2399,2400,2406,2407,2413,2415,2420],{"type":54,"tag":76,"props":2358,"children":2359},{},[2360],{"type":59,"value":2361},"Edit",{"type":59,"value":282},{"type":54,"tag":84,"props":2364,"children":2366},{"className":2365},[],[2367],{"type":59,"value":150},{"type":59,"value":2369},") when a still-live report already covers the same experiment issue — an SRM still skewed, a stall that hasn't recovered, a ",{"type":54,"tag":84,"props":2371,"children":2373},{"className":2372},[],[2374],{"type":59,"value":89},{"type":59,"value":2376}," trend still climbing. ",{"type":54,"tag":84,"props":2378,"children":2380},{"className":2379},[],[2381],{"type":59,"value":2382},"append_note",{"type":59,"value":2384}," the fresh numbers, or rewrite the title\u002Fsummary on a report you authored. This is the default when a match exists. ",{"type":54,"tag":84,"props":2386,"children":2388},{"className":2387},[],[2389],{"type":59,"value":2390},"edit-report",{"type":59,"value":2392}," can't change status, so if the matched report is ",{"type":54,"tag":84,"props":2394,"children":2396},{"className":2395},[],[2397],{"type":59,"value":2398},"resolved",{"type":59,"value":144},{"type":54,"tag":84,"props":2401,"children":2403},{"className":2402},[],[2404],{"type":59,"value":2405},"suppressed",{"type":59,"value":144},{"type":54,"tag":84,"props":2408,"children":2410},{"className":2409},[],[2411],{"type":59,"value":2412},"failed",{"type":59,"value":2414},", don't append (it won't resurface) — author a fresh report for the relapse and repoint the ",{"type":54,"tag":84,"props":2416,"children":2418},{"className":2417},[],[2419],{"type":59,"value":318},{"type":59,"value":2421}," key.",{"type":54,"tag":72,"props":2423,"children":2424},{},[2425,2430,2431,2436,2438,2443,2445,2451,2453,2459,2461,2467,2469,2475,2477,2483,2484,2490,2492,2498,2500,2506,2507,2513,2515,2520,2522,2527,2529,2534],{"type":54,"tag":76,"props":2426,"children":2427},{},[2428],{"type":59,"value":2429},"Author",{"type":59,"value":282},{"type":54,"tag":84,"props":2432,"children":2434},{"className":2433},[],[2435],{"type":59,"value":142},{"type":59,"value":2437},") only when nothing live covers it. A good report names the experiment id and flag key, quantifies the contradiction (observed vs expected split with exposure counts, ",{"type":54,"tag":84,"props":2439,"children":2441},{"className":2440},[],[2442],{"type":59,"value":89},{"type":59,"value":2444}," percentage, days dormant), passes the sample-size gate, and dates the onset — ideally tied to a flag version or activity-log entry. Set ",{"type":54,"tag":84,"props":2446,"children":2448},{"className":2447},[],[2449],{"type":59,"value":2450},"priority",{"type":59,"value":2452}," (P0–P4) + ",{"type":54,"tag":84,"props":2454,"children":2456},{"className":2455},[],[2457],{"type":59,"value":2458},"priority_explanation",{"type":59,"value":2460}," — validity threats on a live decision (SRM, mid-run mutation, contamination) are P2, stalls P2–P3 by blast radius, lifecycle hygiene P3; it's the report's importance in the inbox, your call to make. Set ",{"type":54,"tag":84,"props":2462,"children":2464},{"className":2463},[],[2465],{"type":59,"value":2466},"suggested_reviewers",{"type":59,"value":2468}," via ",{"type":54,"tag":84,"props":2470,"children":2472},{"className":2471},[],[2473],{"type":59,"value":2474},"scout-members-list",{"type":59,"value":2476}," (objects — a ",{"type":54,"tag":84,"props":2478,"children":2480},{"className":2479},[],[2481],{"type":59,"value":2482},"{github_login}",{"type":59,"value":772},{"type":54,"tag":84,"props":2485,"children":2487},{"className":2486},[],[2488],{"type":59,"value":2489},"{user_uuid}",{"type":59,"value":2491},", not bare strings; cache under ",{"type":54,"tag":84,"props":2493,"children":2495},{"className":2494},[],[2496],{"type":59,"value":2497},"reviewer:experiments:\u003Cslug>",{"type":59,"value":2499},"); left empty the report reaches no one. A validity threat is an investigation a human confirms, not a one-line change → ",{"type":54,"tag":84,"props":2501,"children":2503},{"className":2502},[],[2504],{"type":59,"value":2505},"actionability=requires_human_input",{"type":59,"value":570},{"type":54,"tag":84,"props":2508,"children":2510},{"className":2509},[],[2511],{"type":59,"value":2512},"repository=NO_REPO",{"type":59,"value":2514}," (NO_REPO is what stops ",{"type":54,"tag":84,"props":2516,"children":2518},{"className":2517},[],[2519],{"type":59,"value":2450},{"type":59,"value":2521},"+reviewers from spawning a pointless repo-selection sandbox). After authoring, write the ",{"type":54,"tag":84,"props":2523,"children":2525},{"className":2524},[],[2526],{"type":59,"value":2309},{"type":59,"value":2528}," pointer with the ",{"type":54,"tag":84,"props":2530,"children":2532},{"className":2531},[],[2533],{"type":59,"value":2317},{"type":59,"value":2535}," so the next run edits instead of duplicating.",{"type":54,"tag":72,"props":2537,"children":2538},{},[2539,2544,2546,2551,2553,2558,2560,2565,2566,2571,2572,2577],{"type":54,"tag":76,"props":2540,"children":2541},{},[2542],{"type":59,"value":2543},"Remember",{"type":59,"value":2545}," if below the bar but worth carrying forward (a ratio drifting but inside the noise band, ",{"type":54,"tag":84,"props":2547,"children":2549},{"className":2548},[],[2550],{"type":59,"value":89},{"type":59,"value":2552}," creeping at 0.3%, a plateau that needs one more week); ",{"type":54,"tag":76,"props":2554,"children":2555},{},[2556],{"type":59,"value":2557},"skip",{"type":59,"value":2559}," with a one-line note if a ",{"type":54,"tag":84,"props":2561,"children":2563},{"className":2562},[],[2564],{"type":59,"value":296},{"type":59,"value":144},{"type":54,"tag":84,"props":2567,"children":2569},{"className":2568},[],[2570],{"type":59,"value":303},{"type":59,"value":144},{"type":54,"tag":84,"props":2573,"children":2575},{"className":2574},[],[2576],{"type":59,"value":310},{"type":59,"value":2578}," entry or an existing report already covers it.",{"type":54,"tag":62,"props":2580,"children":2581},{},[2582,2584,2590],{"type":59,"value":2583},"Sibling scouts share memory — the feature-flags scout owns non-experiment flag wiring, and the generalist (which ran an experiment-integrity lens before this specialist existed) may hold ",{"type":54,"tag":84,"props":2585,"children":2587},{"className":2586},[],[2588],{"type":59,"value":2589},"dedupe:general:experiment-*",{"type":59,"value":2591}," scratchpad entries; honor them like your own. When a prior run already covered a topic, default to edit-or-skip: the same fact twice in the inbox costs more than missing one finding for one tick.",{"type":54,"tag":258,"props":2593,"children":2595},{"id":2594},"close-out",[2596],{"type":59,"value":2597},"Close out",{"type":54,"tag":62,"props":2599,"children":2600},{},[2601,2603,2608],{"type":59,"value":2602},"Summarize the run in one paragraph: which experiments you checked, which reports you authored or edited, what you remembered, and what you ruled out. The harness saves it as the run summary; future runs read it via ",{"type":54,"tag":84,"props":2604,"children":2606},{"className":2605},[],[2607],{"type":59,"value":336},{"type":59,"value":2609},". Don't write a separate \"run metadata\" scratchpad entry. \"All running experiments healthy\" is a real, useful outcome.",{"type":54,"tag":161,"props":2611,"children":2613},{"id":2612},"disqualifiers-skip-these",[2614],{"type":59,"value":2615},"Disqualifiers (skip these)",{"type":54,"tag":213,"props":2617,"children":2618},{},[2619,2629,2639,2656,2673,2683,2706,2722,2740,2763],{"type":54,"tag":72,"props":2620,"children":2621},{},[2622,2627],{"type":54,"tag":76,"props":2623,"children":2624},{},[2625],{"type":59,"value":2626},"Launched \u003C 24h ago",{"type":59,"value":2628}," — exposure precomputation lags ~15 min and day-one volume is unrepresentative; zero or skewed exposures right after launch are not findings yet.",{"type":54,"tag":72,"props":2630,"children":2631},{},[2632,2637],{"type":54,"tag":76,"props":2633,"children":2634},{},[2635],{"type":59,"value":2636},"Ratio claims below the sample-size gate",{"type":59,"value":2638}," — no SRM call under ~1,000 bucketed exposures, and never inside the 3σ band. Low-volume splits wobble; that's variance.",{"type":54,"tag":72,"props":2640,"children":2641},{},[2642,2647,2649,2654],{"type":54,"tag":76,"props":2643,"children":2644},{},[2645],{"type":59,"value":2646},"Metric movement",{"type":59,"value":2648}," — a variant winning, losing, or wobbling is the team's decision surface, not a scout finding. Only flag metric ",{"type":54,"tag":113,"props":2650,"children":2651},{},[2652],{"type":59,"value":2653},"machinery",{"type":59,"value":2655}," (validity), with one exception: a long-stable answer on a zombie feeds the \"decide now\" recommendation.",{"type":54,"tag":72,"props":2657,"children":2658},{},[2659,2664,2666,2671],{"type":54,"tag":76,"props":2660,"children":2661},{},[2662],{"type":59,"value":2663},"Paused experiments with no fresh exposures",{"type":59,"value":2665}," — that's what pause means. Check flag ",{"type":54,"tag":84,"props":2667,"children":2669},{"className":2668},[],[2670],{"type":59,"value":1899},{"type":59,"value":2672}," before calling a stall.",{"type":54,"tag":72,"props":2674,"children":2675},{},[2676,2681],{"type":54,"tag":76,"props":2677,"children":2678},{},[2679],{"type":59,"value":2680},"Rollout increases mid-run",{"type":59,"value":2682}," — the safe change; new users enter cleanly.",{"type":54,"tag":72,"props":2684,"children":2685},{},[2686,2691,2692,2697,2699,2704],{"type":54,"tag":76,"props":2687,"children":2688},{},[2689],{"type":59,"value":2690},"Forced-variant release conditions",{"type":59,"value":282},{"type":54,"tag":84,"props":2693,"children":2695},{"className":2694},[],[2696],{"type":59,"value":533},{"type":59,"value":2698}," set) — deliberate non-random assignment; aggregate ratios won't match the nominal split by design. Note it once in ",{"type":54,"tag":84,"props":2700,"children":2702},{"className":2701},[],[2703],{"type":59,"value":296},{"type":59,"value":2705}," memory.",{"type":54,"tag":72,"props":2707,"children":2708},{},[2709,2714,2716,2721],{"type":54,"tag":76,"props":2710,"children":2711},{},[2712],{"type":59,"value":2713},"Declared A\u002FA, placebo, or engine-validation experiments",{"type":59,"value":2715}," (name\u002Fdescription says A\u002FA, placebo, validation, identical variants) — long runtimes and null results are the point; skip lifecycle \"decide now\" nudges. SRM checks still fully apply — a skewed A\u002FA is exactly the kind of machinery fault these exist to catch. Note the intent once in ",{"type":54,"tag":84,"props":2717,"children":2719},{"className":2718},[],[2720],{"type":59,"value":296},{"type":59,"value":2705},{"type":54,"tag":72,"props":2723,"children":2724},{},[2725,2730,2732,2738],{"type":54,"tag":76,"props":2726,"children":2727},{},[2728],{"type":59,"value":2729},"Holdout-enrolled experiments",{"type":59,"value":2731}," — the holdout slice shifts effective ratios; read ",{"type":54,"tag":84,"props":2733,"children":2735},{"className":2734},[],[2736],{"type":59,"value":2737},"holdout_id",{"type":59,"value":2739}," before judging a split.",{"type":54,"tag":72,"props":2741,"children":2742},{},[2743,2748,2749,2754,2756,2761],{"type":54,"tag":76,"props":2744,"children":2745},{},[2746],{"type":59,"value":2747},"Bucketing failures",{"type":59,"value":282},{"type":54,"tag":84,"props":2750,"children":2752},{"className":2751},[],[2753],{"type":59,"value":1125},{"type":59,"value":2755}," = false\u002Fempty) counted as variants — exclude from ratios; only their ",{"type":54,"tag":113,"props":2757,"children":2758},{},[2759],{"type":59,"value":2760},"share",{"type":59,"value":2762}," trending up is interesting.",{"type":54,"tag":72,"props":2764,"children":2765},{},[2766,2771,2773,2778],{"type":54,"tag":76,"props":2767,"children":2768},{},[2769],{"type":59,"value":2770},"Experiments already concluded with a conclusion set",{"type":59,"value":2772}," — the team decided; lingering ",{"type":54,"tag":113,"props":2774,"children":2775},{},[2776],{"type":59,"value":2777},"flag",{"type":59,"value":2779}," state is the only thing left worth checking.",{"type":54,"tag":62,"props":2781,"children":2782},{},[2783],{"type":59,"value":2784},"When in doubt, write a memory entry instead of filing a report.",{"type":54,"tag":161,"props":2786,"children":2788},{"id":2787},"mcp-tools",[2789],{"type":59,"value":2790},"MCP tools",{"type":54,"tag":62,"props":2792,"children":2793},{},[2794],{"type":59,"value":2795},"Direct calls (read-only):",{"type":54,"tag":213,"props":2797,"children":2798},{},[2799,2825,2889,3016,3027,3050,3067,3084,3137],{"type":54,"tag":72,"props":2800,"children":2801},{},[2802,2808,2810,2815,2817,2823],{"type":54,"tag":84,"props":2803,"children":2805},{"className":2804},[],[2806],{"type":59,"value":2807},"experiment-list",{"type":59,"value":2809}," — cheap candidate discovery: id, name, status (draft \u002F running \u002F paused \u002F stopped), dates, ",{"type":54,"tag":84,"props":2811,"children":2813},{"className":2812},[],[2814],{"type":59,"value":444},{"type":59,"value":2816},". Filter by ",{"type":54,"tag":84,"props":2818,"children":2820},{"className":2819},[],[2821],{"type":59,"value":2822},"status",{"type":59,"value":2824},"; start here.",{"type":54,"tag":72,"props":2826,"children":2827},{},[2828,2833,2834,2839,2841,2846,2848,2853,2855,2860,2861,2866,2868,2873,2874,2879,2881,2887],{"type":54,"tag":84,"props":2829,"children":2831},{"className":2830},[],[2832],{"type":59,"value":919},{"type":59,"value":348},{"type":54,"tag":76,"props":2835,"children":2836},{},[2837],{"type":59,"value":2838},"the flagship detector",{"type":59,"value":2840},": exposure block (",{"type":54,"tag":84,"props":2842,"children":2844},{"className":2843},[],[2845],{"type":59,"value":552},{"type":59,"value":2847},", daily ",{"type":54,"tag":84,"props":2849,"children":2851},{"className":2850},[],[2852],{"type":59,"value":560},{"type":59,"value":2854},", native ",{"type":54,"tag":84,"props":2856,"children":2858},{"className":2857},[],[2859],{"type":59,"value":568},{"type":59,"value":488},{"type":54,"tag":84,"props":2862,"children":2864},{"className":2863},[],[2865],{"type":59,"value":576},{"type":59,"value":2867},") plus per-metric ",{"type":54,"tag":84,"props":2869,"children":2871},{"className":2870},[],[2872],{"type":59,"value":584},{"type":59,"value":144},{"type":54,"tag":84,"props":2875,"children":2877},{"className":2876},[],[2878],{"type":59,"value":591},{"type":59,"value":2880},". Heavy response with many metrics — read the exposure + validation fields, skip the per-metric stats. New-engine experiments only; pass ",{"type":54,"tag":84,"props":2882,"children":2884},{"className":2883},[],[2885],{"type":59,"value":2886},"refresh: false",{"type":59,"value":2888},".",{"type":54,"tag":72,"props":2890,"children":2891},{},[2892,2897,2899,2904,2906,2911,2912,2918,2919,2924,2925,2930,2932,2938,2939,2944,2945,2951,2953,2958,2959,2964,2966,2971,2973,2979,2980,2986,2987,2993,2994,2999,3001,3006,3008,3014],{"type":54,"tag":84,"props":2893,"children":2895},{"className":2894},[],[2896],{"type":59,"value":1377},{"type":59,"value":2898}," — full config for a candidate: ",{"type":54,"tag":84,"props":2900,"children":2902},{"className":2901},[],[2903],{"type":59,"value":478},{"type":59,"value":2905}," (configured split), ",{"type":54,"tag":84,"props":2907,"children":2909},{"className":2908},[],[2910],{"type":59,"value":486},{"type":59,"value":488},{"type":54,"tag":84,"props":2913,"children":2915},{"className":2914},[],[2916],{"type":59,"value":2917},"recommended_sample_size",{"type":59,"value":488},{"type":54,"tag":84,"props":2920,"children":2922},{"className":2921},[],[2923],{"type":59,"value":1426},{"type":59,"value":488},{"type":54,"tag":84,"props":2926,"children":2928},{"className":2927},[],[2929],{"type":59,"value":494},{"type":59,"value":2931}," (custom ",{"type":54,"tag":84,"props":2933,"children":2935},{"className":2934},[],[2936],{"type":59,"value":2937},"exposure_event",{"type":59,"value":488},{"type":54,"tag":84,"props":2940,"children":2942},{"className":2941},[],[2943],{"type":59,"value":502},{"type":59,"value":488},{"type":54,"tag":84,"props":2946,"children":2948},{"className":2947},[],[2949],{"type":59,"value":2950},"filterTestAccounts",{"type":59,"value":2952},"), ",{"type":54,"tag":84,"props":2954,"children":2956},{"className":2955},[],[2957],{"type":59,"value":517},{"type":59,"value":488},{"type":54,"tag":84,"props":2960,"children":2962},{"className":2961},[],[2963],{"type":59,"value":2737},{"type":59,"value":2965},", linked ",{"type":54,"tag":84,"props":2967,"children":2969},{"className":2968},[],[2970],{"type":59,"value":525},{"type":59,"value":2972}," (active, ",{"type":54,"tag":84,"props":2974,"children":2976},{"className":2975},[],[2977],{"type":59,"value":2978},"version",{"type":59,"value":488},{"type":54,"tag":84,"props":2981,"children":2983},{"className":2982},[],[2984],{"type":59,"value":2985},"bucketing_identifier",{"type":59,"value":488},{"type":54,"tag":84,"props":2988,"children":2990},{"className":2989},[],[2991],{"type":59,"value":2992},"ensure_experience_continuity",{"type":59,"value":488},{"type":54,"tag":84,"props":2995,"children":2997},{"className":2996},[],[2998],{"type":59,"value":533},{"type":59,"value":3000}," overrides), ",{"type":54,"tag":84,"props":3002,"children":3004},{"className":3003},[],[3005],{"type":59,"value":1983},{"type":59,"value":3007}," (each with ",{"type":54,"tag":84,"props":3009,"children":3011},{"className":3010},[],[3012],{"type":59,"value":3013},"uuid",{"type":59,"value":3015}," + fingerprint). Large response — candidates only.",{"type":54,"tag":72,"props":3017,"children":3018},{},[3019,3025],{"type":54,"tag":84,"props":3020,"children":3022},{"className":3021},[],[3023],{"type":59,"value":3024},"experiment-stats",{"type":59,"value":3026}," — project-wide velocity aggregate (launched \u002F completed last 30d, active count). Cheap context for the hygiene pass.",{"type":54,"tag":72,"props":3028,"children":3029},{},[3030,3035,3037,3042,3043,3048],{"type":54,"tag":84,"props":3031,"children":3033},{"className":3032},[],[3034],{"type":59,"value":1959},{"type":59,"value":3036}," — day-by-day per-variant results for one metric (",{"type":54,"tag":84,"props":3038,"children":3040},{"className":3039},[],[3041],{"type":59,"value":1967},{"type":59,"value":1969},{"type":54,"tag":84,"props":3044,"children":3046},{"className":3045},[],[3047],{"type":59,"value":1975},{"type":59,"value":3049}," from the metrics array). Use sparingly, for the zombie \"decide now\" check.",{"type":54,"tag":72,"props":3051,"children":3052},{},[3053,3059,3060,3065],{"type":54,"tag":84,"props":3054,"children":3056},{"className":3055},[],[3057],{"type":59,"value":3058},"feature-flag-get-definition",{"type":59,"value":144},{"type":54,"tag":84,"props":3061,"children":3063},{"className":3062},[],[3064],{"type":59,"value":1242},{"type":59,"value":3066}," — flag state and edit-history diffs; the latter is how you date mid-run mutations.",{"type":54,"tag":72,"props":3068,"children":3069},{},[3070,3075,3076,3082],{"type":54,"tag":84,"props":3071,"children":3073},{"className":3072},[],[3074],{"type":59,"value":1796},{"type":59,"value":282},{"type":54,"tag":84,"props":3077,"children":3079},{"className":3078},[],[3080],{"type":59,"value":3081},"scopes: [\"Experiment\"]",{"type":59,"value":3083},") — experiment-level edit timeline.",{"type":54,"tag":72,"props":3085,"children":3086},{},[3087,3092,3094,3100,3102,3107,3109,3114,3116,3121,3123,3128,3130,3135],{"type":54,"tag":84,"props":3088,"children":3090},{"className":3089},[],[3091],{"type":59,"value":626},{"type":59,"value":3093}," against ",{"type":54,"tag":84,"props":3095,"children":3097},{"className":3096},[],[3098],{"type":59,"value":3099},"events",{"type":59,"value":3101}," — exposure analysis. Properties: ",{"type":54,"tag":84,"props":3103,"children":3105},{"className":3104},[],[3106],{"type":59,"value":1656},{"type":59,"value":3108}," (flag key) + ",{"type":54,"tag":84,"props":3110,"children":3112},{"className":3111},[],[3113],{"type":59,"value":1125},{"type":59,"value":3115}," (variant, incl. ",{"type":54,"tag":84,"props":3117,"children":3119},{"className":3118},[],[3120],{"type":59,"value":89},{"type":59,"value":3122},") on ",{"type":54,"tag":84,"props":3124,"children":3126},{"className":3125},[],[3127],{"type":59,"value":1634},{"type":59,"value":3129},"; ",{"type":54,"tag":84,"props":3131,"children":3133},{"className":3132},[],[3134],{"type":59,"value":1744},{"type":59,"value":3136}," on custom exposure events.",{"type":54,"tag":72,"props":3138,"children":3139},{},[3140,3146],{"type":54,"tag":84,"props":3141,"children":3143},{"className":3142},[],[3144],{"type":59,"value":3145},"read-data-schema",{"type":59,"value":3147}," — confirm a custom exposure event and its properties exist before aggregating over them.",{"type":54,"tag":62,"props":3149,"children":3150},{},[3151],{"type":59,"value":3152},"Inbox & reviewer routing:",{"type":54,"tag":213,"props":3154,"children":3155},{},[3156,3179,3197],{"type":54,"tag":72,"props":3157,"children":3158},{},[3159,3164,3165,3170,3172,3177],{"type":54,"tag":84,"props":3160,"children":3162},{"className":3161},[],[3163],{"type":59,"value":372},{"type":59,"value":144},{"type":54,"tag":84,"props":3166,"children":3168},{"className":3167},[],[3169],{"type":59,"value":401},{"type":59,"value":3171}," — the reports already in the inbox; check before authoring so you edit instead of duplicating (",{"type":54,"tag":84,"props":3173,"children":3175},{"className":3174},[],[3176],{"type":59,"value":387},{"type":59,"value":3178},").",{"type":54,"tag":72,"props":3180,"children":3181},{},[3182,3188,3190,3195],{"type":54,"tag":84,"props":3183,"children":3185},{"className":3184},[],[3186],{"type":59,"value":3187},"inbox-report-artefacts-list",{"type":59,"value":3189}," — a comparable report's artefact log, where the routed ",{"type":54,"tag":84,"props":3191,"children":3193},{"className":3192},[],[3194],{"type":59,"value":2466},{"type":59,"value":3196}," live (the report record doesn't expose them) — reviewer precedent.",{"type":54,"tag":72,"props":3198,"children":3199},{},[3200,3205,3207,3213,3215,3220,3222,3227,3229,3234,3236,3241],{"type":54,"tag":84,"props":3201,"children":3203},{"className":3202},[],[3204],{"type":59,"value":2474},{"type":59,"value":3206}," — this project's members with their resolved ",{"type":54,"tag":84,"props":3208,"children":3210},{"className":3209},[],[3211],{"type":59,"value":3212},"github_login",{"type":59,"value":3214},", to route ",{"type":54,"tag":84,"props":3216,"children":3218},{"className":3217},[],[3219],{"type":59,"value":2466},{"type":59,"value":3221}," to an experiment's owner (wrap as a ",{"type":54,"tag":84,"props":3223,"children":3225},{"className":3224},[],[3226],{"type":59,"value":2482},{"type":59,"value":3228}," object, or pass the member's ",{"type":54,"tag":84,"props":3230,"children":3232},{"className":3231},[],[3233],{"type":59,"value":2489},{"type":59,"value":3235}," and let the server resolve; null ",{"type":54,"tag":84,"props":3237,"children":3239},{"className":3238},[],[3240],{"type":59,"value":3212},{"type":59,"value":3242}," → try the next owner). The in-run roster; the org-scoped resolver tools aren't available in a scout run.",{"type":54,"tag":62,"props":3244,"children":3245},{},[3246],{"type":59,"value":3247},"Harness-level:",{"type":54,"tag":213,"props":3249,"children":3250},{},[3251,3280,3296],{"type":54,"tag":72,"props":3252,"children":3253},{},[3254,3259,3260,3265,3266,3271,3272,3278],{"type":54,"tag":84,"props":3255,"children":3257},{"className":3256},[],[3258],{"type":59,"value":185},{"type":59,"value":144},{"type":54,"tag":84,"props":3261,"children":3263},{"className":3262},[],[3264],{"type":59,"value":280},{"type":59,"value":144},{"type":54,"tag":84,"props":3267,"children":3269},{"className":3268},[],[3270],{"type":59,"value":336},{"type":59,"value":144},{"type":54,"tag":84,"props":3273,"children":3275},{"className":3274},[],[3276],{"type":59,"value":3277},"scout-runs-retrieve",{"type":59,"value":3279}," — orientation + dedupe.",{"type":54,"tag":72,"props":3281,"children":3282},{},[3283,3288,3289,3294],{"type":54,"tag":84,"props":3284,"children":3286},{"className":3285},[],[3287],{"type":59,"value":142},{"type":59,"value":144},{"type":54,"tag":84,"props":3290,"children":3292},{"className":3291},[],[3293],{"type":59,"value":150},{"type":59,"value":3295}," — author a report \u002F edit an existing one (the report-channel contract is in the harness prompt).",{"type":54,"tag":72,"props":3297,"children":3298},{},[3299,3305,3306,3312],{"type":54,"tag":84,"props":3300,"children":3302},{"className":3301},[],[3303],{"type":59,"value":3304},"scout-scratchpad-remember",{"type":59,"value":144},{"type":54,"tag":84,"props":3307,"children":3309},{"className":3308},[],[3310],{"type":59,"value":3311},"scout-scratchpad-forget",{"type":59,"value":3313}," — remember \u002F prune stale memory keys.",{"type":54,"tag":161,"props":3315,"children":3317},{"id":3316},"when-to-stop",[3318],{"type":59,"value":3319},"When to stop",{"type":54,"tag":213,"props":3321,"children":3322},{},[3323,3336,3348,3372],{"type":54,"tag":72,"props":3324,"children":3325},{},[3326,3328,3334],{"type":59,"value":3327},"No experiments in use → ",{"type":54,"tag":84,"props":3329,"children":3331},{"className":3330},[],[3332],{"type":59,"value":3333},"not-in-use:",{"type":59,"value":3335}," entry, close out empty.",{"type":54,"tag":72,"props":3337,"children":3338},{},[3339,3341,3346],{"type":59,"value":3340},"All running experiments match their config (ratio in band, fresh exposures, no post-launch flag edits) → close out empty; refresh ",{"type":54,"tag":84,"props":3342,"children":3344},{"className":3343},[],[3345],{"type":59,"value":1213},{"type":59,"value":3347}," baselines if stale.",{"type":54,"tag":72,"props":3349,"children":3350},{},[3351,3353,3358,3359,3364,3365,3370],{"type":59,"value":3352},"Candidates all gated by ",{"type":54,"tag":84,"props":3354,"children":3356},{"className":3355},[],[3357],{"type":59,"value":296},{"type":59,"value":144},{"type":54,"tag":84,"props":3360,"children":3362},{"className":3361},[],[3363],{"type":59,"value":303},{"type":59,"value":144},{"type":54,"tag":84,"props":3366,"children":3368},{"className":3367},[],[3369],{"type":59,"value":310},{"type":59,"value":3371}," entries, or an existing inbox report → edit-or-skip and close out.",{"type":54,"tag":72,"props":3373,"children":3374},{},[3375],{"type":59,"value":3376},"You've filed (or edited) reports for what's solid → close out. One sharp validity report beats a laundry list of P3 hygiene nits.",{"type":54,"tag":62,"props":3378,"children":3379},{},[3380],{"type":59,"value":3381},"\"Looked but found nothing meaningful\" is a real outcome.",{"type":54,"tag":3383,"props":3384,"children":3385},"style",{},[3386],{"type":59,"value":3387},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":3389,"total":3505},[3390,3405,3423,3440,3455,3469,3487],{"slug":3391,"name":3391,"fn":3392,"description":3393,"org":3394,"tags":3395,"stars":26,"repoUrl":27,"updatedAt":3404},"analyzing-experiment-session-replays","analyze session replays for PostHog experiments","Analyze session replay patterns across experiment variants to understand user behavior differences. Use when the user wants to see how users interact with different experiment variants, identify usability issues, compare behavior patterns between control and test groups, or get qualitative insights to complement quantitative experiment results.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3396,3397,3400,3401],{"name":21,"slug":22,"type":15},{"name":3398,"slug":3399,"type":15},"Design","design",{"name":9,"slug":8,"type":15},{"name":3402,"slug":3403,"type":15},"User Research","user-research","2026-04-06T18:44:38.291781",{"slug":3406,"name":3406,"fn":3407,"description":3408,"org":3409,"tags":3410,"stars":26,"repoUrl":27,"updatedAt":3422},"assessing-heatmaps","analyze page heatmaps and suggest improvements","Assesses what a page's heatmap is telling you and recommends concrete changes. Pulls click \u002F rageclick \u002F scroll-depth data for a URL, names the hot elements by cross-referencing autocapture events on the same page, and can create a saved heatmap the user opens in PostHog, then summarizes the behavior and proposes improvements.\nTRIGGER when: user asks what a heatmap shows, why people aren't clicking something, where users rage-click, how far they scroll, what to change on a page based on heatmap\u002Fclick data, or to 'analyze\u002Fassess\u002Freview the heatmap' for a URL.\nDO NOT TRIGGER when: the user only wants to create a saved heatmap screenshot with no analysis (use heatmaps-saved-create directly), or is asking about session replay in general (use investigating-replay).",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3411,3412,3415,3416,3419],{"name":21,"slug":22,"type":15},{"name":3413,"slug":3414,"type":15},"Frontend","frontend",{"name":9,"slug":8,"type":15},{"name":3417,"slug":3418,"type":15},"Product Management","product-management",{"name":3420,"slug":3421,"type":15},"UX Design","ux-design","2026-06-05T07:40:43.37798",{"slug":3424,"name":3424,"fn":3425,"description":3426,"org":3427,"tags":3428,"stars":26,"repoUrl":27,"updatedAt":3439},"auditing-experiments-flags","audit PostHog experiments and feature flags","Audit PostHog experiments and feature flags for configuration issues, staleness, and best-practice violations. Read when the user asks to audit, health-check, or review experiments or feature flags, check flag hygiene, or verify experiment setup.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3429,3432,3435,3436],{"name":3430,"slug":3431,"type":15},"Audit","audit",{"name":3433,"slug":3434,"type":15},"Feature Flags","feature-flags",{"name":9,"slug":8,"type":15},{"name":3437,"slug":3438,"type":15},"QA","qa","2026-04-06T18:44:30.657553",{"slug":3441,"name":3441,"fn":3442,"description":3443,"org":3444,"tags":3445,"stars":26,"repoUrl":27,"updatedAt":3454},"authoring-scouts","author and edit PostHog Signals scouts","How to author, edit, and adapt PostHog Signals scouts — the scheduled agents that scan a project and write reports into the Signals inbox. Use when a user wants to customize a canonical scout for their own setup (narrow its scope, retune its thresholds, add disqualifiers), tweak a scout's schedule or dry-run posture, or write a brand-new scout from scratch for a specific use case (a custom event, a product surface no canonical scout covers), or steer a scout without editing it at all by leaving it a note. Covers the scout SKILL.md anatomy, the report contract, the dedupe + scratchpad-memory conventions, the scout-notes steering channel, the per-team skills-store path vs the canonical in-repo path, and the write-and-inspect test loop (with dry-run as an optional safety net). Trigger on \"write\u002Fedit\u002Fcustomize a signals scout\", \"new scout for X\", \"tune my scout schedule\", \"make a scout that watches \u003Cevent>\", \"leave a note for \u002F give feedback to a scout\", \"tell the scouts about X\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3446,3449,3452,3453],{"name":3447,"slug":3448,"type":15},"Agents","agents",{"name":3450,"slug":3451,"type":15},"Automation","automation",{"name":13,"slug":14,"type":15},{"name":9,"slug":8,"type":15},"2026-07-28T05:33:45.509154",{"slug":3456,"name":3456,"fn":3457,"description":3458,"org":3459,"tags":3460,"stars":26,"repoUrl":27,"updatedAt":3468},"building-a-dashboard","build and update PostHog dashboards","Build a new dashboard, or update an existing one, from a set of insights — the same job the in-app assistant does with its upsert-dashboard tool, but over MCP. Use when a user asks to create a dashboard, put several metrics\u002Fcharts together on one page, assemble a dashboard for a topic (product analytics, retention, revenue, activation, etc.), or add\u002Fremove\u002Freplace insights on a dashboard they already have. Covers deciding create vs update, reusing existing insights vs creating new ones, and using PostHog's vetted dashboard templates as reference for what a strong dashboard on a topic looks like.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3461,3462,3465],{"name":21,"slug":22,"type":15},{"name":3463,"slug":3464,"type":15},"Dashboards","dashboards",{"name":3466,"slug":3467,"type":15},"MCP","mcp","2026-07-21T06:07:38.060598",{"slug":3470,"name":3470,"fn":3471,"description":3472,"org":3473,"tags":3474,"stars":26,"repoUrl":27,"updatedAt":3486},"checking-deploy-timing","correlate PostHog deployments with GitHub commits","Determine when a PostHog code change reached a given environment by reading the hidden GIT deploy annotations in the project and correlating them with the merge commit on GitHub. Use when PostHog staff ask \"when was X deployed\", \"is my change live in the US\u002FEU yet\", \"has my PR shipped\", \"did the fix roll out to prod-us\", or otherwise want to know whether\u002Fwhen a commit, PR, or feature went out to a region. Do not answer deploy-timing questions from event\u002Fdata volume alone — that only shows when data changed, not when code shipped.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3475,3478,3481,3484,3485],{"name":3476,"slug":3477,"type":15},"Deployment","deployment",{"name":3479,"slug":3480,"type":15},"Git","git",{"name":3482,"slug":3483,"type":15},"GitHub","github",{"name":13,"slug":14,"type":15},{"name":9,"slug":8,"type":15},"2026-06-28T07:46:59.53536",{"slug":3488,"name":3488,"fn":3489,"description":3490,"org":3491,"tags":3492,"stars":26,"repoUrl":27,"updatedAt":3504},"choosing-trend-or-slope-view","visualize trends and growth over time","Clarify how to visualize change over a time range before building a trend. Use whenever the user asks how much something changed, grew, dropped, improved, or regressed between two points or periods — \"how much did X change from A to B\", \"before vs after\", \"start vs end\", \"week over week\", \"compare this month to last\", \"change over time\" — or mentions a \"slope chart\" \u002F \"slopegraph\". Two readings of \"change\" need different charts: the whole trend (a line, every interval) versus just the two endpoints (a slope, start vs end). Ask which they want, then render it. Not for choosing a saved insight ChartDisplayType in the insight editor.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3493,3494,3497,3500,3501],{"name":21,"slug":22,"type":15},{"name":3495,"slug":3496,"type":15},"Charts","charts",{"name":3498,"slug":3499,"type":15},"Data Visualization","data-visualization",{"name":9,"slug":8,"type":15},{"name":3502,"slug":3503,"type":15},"Reporting","reporting","2026-06-18T08:18:57.960157",56,{"items":3507,"total":3664},[3508,3523,3533,3546,3559,3574,3588,3601,3613,3628,3638,3654],{"slug":3509,"name":3509,"fn":3510,"description":3511,"org":3512,"tags":3513,"stars":3520,"repoUrl":3521,"updatedAt":3522},"analyzing-expensive-users","analyze expensive users in AI observability","Analyze the most expensive users in AI observability and explain why they cost so much. Use when the user asks about top spenders, expensive users, per-user LLM cost, user-level cost drivers, or patterns behind high AI observability spend.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3514,3515,3518,3519],{"name":21,"slug":22,"type":15},{"name":3516,"slug":3517,"type":15},"Cost Optimization","cost-optimization",{"name":13,"slug":14,"type":15},{"name":9,"slug":8,"type":15},35568,"https:\u002F\u002Fgithub.com\u002FPostHog\u002Fposthog","2026-07-28T05:34:11.117757",{"slug":3524,"name":3524,"fn":3525,"description":3526,"org":3527,"tags":3528,"stars":3520,"repoUrl":3521,"updatedAt":3532},"auditing-endpoints","audit PostHog project endpoints","Audit every endpoint in a PostHog project for staleness, failed materialisations, and unused materialised versions. Use when the user asks \"what endpoints can I clean up?\", \"are any of my endpoints broken?\", \"which materialised versions are still being called?\", or wants a one-shot cleanup pass over the Endpoints product. Produces a prioritised report grouped by issue type, with recommended actions but does not modify anything without explicit confirmation.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3529,3530,3531],{"name":21,"slug":22,"type":15},{"name":3430,"slug":3431,"type":15},{"name":9,"slug":8,"type":15},"2026-06-08T08:08:33.693989",{"slug":3534,"name":3534,"fn":3535,"description":3536,"org":3537,"tags":3538,"stars":3520,"repoUrl":3521,"updatedAt":3545},"auditing-warehouse-source-health","audit PostHog data warehouse source health","Audit the health of a PostHog project's data warehouse sources and syncs — find every broken or degraded source connection, sync schema, and webhook channel. Use when the user asks \"why are my imports failing?\", \"what's broken with my sources?\", \"why is my warehouse data stale?\", or wants a one-shot triage of source\u002Fsync health before deciding where to dig in. Produces a prioritized report grouped by severity, with recommended next steps. For materialized-view health use `auditing-warehouse-view-health`; for a single failing sync use `diagnosing-failed-warehouse-syncs`.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3539,3540,3543,3544],{"name":3430,"slug":3431,"type":15},{"name":3541,"slug":3542,"type":15},"Data Warehouse","data-warehouse",{"name":13,"slug":14,"type":15},{"name":9,"slug":8,"type":15},"2026-06-18T08:22:57.67984",{"slug":3547,"name":3547,"fn":3548,"description":3549,"org":3550,"tags":3551,"stars":3520,"repoUrl":3521,"updatedAt":3558},"auditing-warehouse-view-health","audit PostHog materialized view health","Audit the health of a PostHog project's materialized views (saved queries) — find every failed materialization and flag unused or stale materialized views that cost storage and compute. Use when the user asks \"which of my views are broken?\", \"why is this materialized view failing?\", \"are any of my views wasting compute?\", or wants a one-shot triage of view health. For source\u002Fsync health use `auditing-warehouse-source-health`.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3552,3553,3554,3557],{"name":3430,"slug":3431,"type":15},{"name":3541,"slug":3542,"type":15},{"name":3555,"slug":3556,"type":15},"Performance","performance",{"name":9,"slug":8,"type":15},"2026-06-18T08:25:10.936787",{"slug":3560,"name":3560,"fn":3561,"description":3562,"org":3563,"tags":3564,"stars":3520,"repoUrl":3521,"updatedAt":3573},"authoring-error-tracking-alerts","author PostHog error tracking alerts","Author error tracking alerts that fire when an issue is created, reopened, or starts spiking. Use when the user asks to set up error notifications, route exceptions to Slack\u002Fwebhook\u002FLinear, or evaluate which error events are worth alerting on. Covers trigger-event selection, integration choice, dedup against existing alerts, and shipping with the canonical message body shape.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3565,3568,3571,3572],{"name":3566,"slug":3567,"type":15},"Alerting","alerting",{"name":3569,"slug":3570,"type":15},"Debugging","debugging",{"name":13,"slug":14,"type":15},{"name":9,"slug":8,"type":15},"2026-06-18T08:24:40.318583",{"slug":3575,"name":3575,"fn":3576,"description":3577,"org":3578,"tags":3579,"stars":3520,"repoUrl":3521,"updatedAt":3587},"authoring-log-alerts","author log alerts in PostHog","Author useful, low-noise log alerts on services in a PostHog project. Use when the user asks to set up alerts for their logs, suggest alerts they should add, or evaluate whether a service is worth monitoring. Covers service triage, baseline characterisation, threshold drafting, back-testing via simulate, and shipping with a notification destination.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3580,3581,3582,3583,3586],{"name":21,"slug":22,"type":15},{"name":18,"slug":19,"type":15},{"name":13,"slug":14,"type":15},{"name":3584,"slug":3585,"type":15},"Operations","operations",{"name":9,"slug":8,"type":15},"2026-07-18T05:10:54.430898",{"slug":3589,"name":3589,"fn":3590,"description":3591,"org":3592,"tags":3593,"stars":3520,"repoUrl":3521,"updatedAt":3600},"building-workflows","build and edit PostHog workflows","Build, edit, test, enable, and monitor PostHog workflows over MCP. Author the action\u002Fedge graph so it runs and opens cleanly in the visual editor, then change drafts surgically with patch operations. Use when asked to build, set up, automate, change, fix, or debug a workflow, campaign, broadcast, drip sequence, or event-triggered automation in the workflows product.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3594,3595,3596,3597],{"name":3450,"slug":3451,"type":15},{"name":3466,"slug":3467,"type":15},{"name":9,"slug":8,"type":15},{"name":3598,"slug":3599,"type":15},"Workflow Automation","workflow-automation","2026-07-28T05:34:12.167015",{"slug":3602,"name":3602,"fn":3603,"description":3604,"org":3605,"tags":3606,"stars":3520,"repoUrl":3521,"updatedAt":3612},"check-posthog-loading","inspect PostHog SDK loading across URLs","Inspect how the PostHog JavaScript SDK is loaded across a list of URLs. Use to confirm consistent installation across pages, find pages missing the snippet, detect mismatched API keys or hosts between pages, and verify the load method (head snippet vs deferred vs array.js).\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3607,3608,3609,3610,3611],{"name":21,"slug":22,"type":15},{"name":3569,"slug":3570,"type":15},{"name":3413,"slug":3414,"type":15},{"name":13,"slug":14,"type":15},{"name":9,"slug":8,"type":15},"2026-05-07T05:56:19.828048",{"slug":3614,"name":3614,"fn":3615,"description":3616,"org":3617,"tags":3618,"stars":3520,"repoUrl":3521,"updatedAt":3627},"consuming-endpoints-from-client-code","integrate PostHog endpoints into client applications","Wire a PostHog endpoint into a client app or SDK. Covers fetching the OpenAPI spec, generating a typed client with openapi-generator or @hey-api\u002Fopenapi-ts, sending the right auth header, shaping the variables payload (HogQL code_name vs insight breakdown property), handling rate-limit and materialised-endpoint error responses. Use when the user says \"how do I call my endpoint\", \"generate a client for this\", or \"what auth header do I use\".\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3619,3622,3623,3624],{"name":3620,"slug":3621,"type":15},"API Development","api-development",{"name":3413,"slug":3414,"type":15},{"name":9,"slug":8,"type":15},{"name":3625,"slug":3626,"type":15},"SDK","sdk","2026-06-08T08:08:34.929454",{"slug":3629,"name":3629,"fn":3630,"description":3631,"org":3632,"tags":3633,"stars":3520,"repoUrl":3521,"updatedAt":3637},"copying-endpoints-across-projects","copy PostHog endpoints across projects","Copy a PostHog endpoint (a saved HogQL\u002Finsight query exposed as an API route) to another project in the same organization, or duplicate it under a new name in the same project. Use when the user wants to duplicate an endpoint, promote an endpoint from staging to production, replicate an endpoint's query\u002Fvariables\u002Ffreshness config in another workspace, or clone an endpoint to iterate on it. Unlike feature flags and experiments, endpoints have NO native cross-project copy tool — this skill covers the read-then-recreate flow (endpoint-get then endpoint-create), the active-project switching it requires, name-collision checks, and the safe defaults (land unmaterialised in the target, verify with endpoint-run). Does not cover editing endpoint versions (see managing-endpoint-versions) or authoring a brand-new endpoint from scratch (see creating-an-endpoint).\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3634,3635,3636],{"name":3620,"slug":3621,"type":15},{"name":3584,"slug":3585,"type":15},{"name":9,"slug":8,"type":15},"2026-07-15T05:29:58.442727",{"slug":3639,"name":3639,"fn":3640,"description":3641,"org":3642,"tags":3643,"stars":3520,"repoUrl":3521,"updatedAt":3653},"creating-ai-subscription","schedule recurring AI-generated PostHog reports","Create a recurring AI-generated PostHog report — schedule a free-text prompt to run on a cron, with the LLM-synthesized markdown delivered to email or Slack on each tick. Use when the user wants a recurring AI summary of X on any cadence (daily, weekly, monthly, yearly) rather than a one-off report. (To attach an AI summary to an existing insight\u002Fdashboard subscription instead of a free-text prompt, see `managing-subscriptions` and its `summary_enabled` option.)\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3644,3645,3648,3649,3650],{"name":3450,"slug":3451,"type":15},{"name":3646,"slug":3647,"type":15},"Email","email",{"name":9,"slug":8,"type":15},{"name":3502,"slug":3503,"type":15},{"name":3651,"slug":3652,"type":15},"Slack","slack","2026-06-09T07:32:27.935712",{"slug":3655,"name":3655,"fn":3656,"description":3657,"org":3658,"tags":3659,"stars":3520,"repoUrl":3521,"updatedAt":3663},"creating-an-endpoint","create PostHog API endpoints","Create a PostHog endpoint with the right shape on the first try — covers query kind choice, name conventions, what to expose as variables (HogQL code_name vs insight breakdown), data_freshness_seconds, and whether to materialise on day one. Use when the user says \"create an endpoint\", \"expose this query as an API\", \"turn this insight into an endpoint\", or asks for help structuring a new endpoint. Steers away from common mistakes: materialising a query with cohort breakdowns or compare mode, inline-only variables on a materialised endpoint, unbounded date ranges, ambiguous names.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[3660,3661,3662],{"name":21,"slug":22,"type":15},{"name":3620,"slug":3621,"type":15},{"name":9,"slug":8,"type":15},"2026-06-08T08:08:29.624498",231]