Know who a release breaks before you ship it.
A green test suite tells you the code does what the tests expect. Release tells you whether the people paying you can still finish what they came to do.
| Cohort | Accounts | MRR | Result |
|---|---|---|---|
| Trial | 1,204 | $0 | Queued |
| Standard | 476 | $87,870 | Queued |
| Power users | 86 | $22,400 | Queued |
| Enterprise | 14 | $61,880 | Queued |
| Unusual combinations | 41 | $12,140 | Queued |
“142 tests passed” is not the same as “nobody is worse off”.
What you can already see
- Unit tests
- 1,204 passing
- Integration tests
- 188 passing
- Staging smoke test
- Green
- Coverage
- 76%
What nothing tells you
- Which real workflows this build actually breaks.
- Who runs them, and how often.
- How much revenue is standing behind those people.
Tests encode what you thought of. Customers do the things you didn't. Release closes that gap by replaying what they actually do, against the build you are about to ship.
Replay reality against the build.
- 01
Learn
Product analytics become cohorts: how each kind of customer actually moves through the product.
- 02
Synthesise
Representative workflows are generated per cohort, including the unusual feature combinations.
- 03
Replay
Every workflow runs against the staging build before the release is promoted.
- 04
Report
Failures come back as cohorts, workflows and the MRR attached to them.
A regression is an engineering fact until you price it.
- ARR at stake
- $221K
- Customers
- 312 accounts
- Margin
- —
- Contracts
- 1
- Claims
- 0
- Engineering
- ~2 days
- Risk
- Medium
Release answers the rows it can see. The blank ones are not missing data — they are outside what this system reads.
Replaying v4.3 against real cohorts finds a regression in the export workflow. Fourteen per cent of enterprise accounts — 312 of 2,180 — run it, which is the sort of number that gets argued about in a release meeting and shipped anyway.
Release knows who those accounts are. It does not know that one of them holds an availability control naming that workflow, or what their renewals are worth. Preflight asks Contract and Margin the same question and the argument ends.
312 accounts
$221K at renewal
Sources this product is designed to read
PostHog, Stripe can be connected today. The rest are sources the model is built for and cannot be connected yet — the app never offers a provider whose consent flow it cannot complete.
Behaviour
- PostHog — can be connected today
- Amplitude
- Mixpanel
- Segment
Delivery
- GitHub Actions
- Vercel
- Preview deploys
- Staging
Revenue
- Stripe — can be connected today
- Chargebee
- Paddle
Release is in beta.
Built and running: journeys replay against staging in a real browser, a retry that rescues a failure is reported as flaky rather than green, and a journey nobody has measured raises severity instead of lowering it. What it does not have yet is analytics weighting, so reach reads not measured rather than a number nobody collected.