Executable Stories Cloud
At 4:52 on a Friday, every test in the plan was green.
The rollback started at 2am.
You know this room. GitHub in one tab, a regression spreadsheet calledfinal_v3 in another, TestRail in a third, and a release manager asking one question nobody can answer from any of them: what did we actually test against this commit?
The villain is drift
The test plan was written in March. The code has changed four hundred times since. Two things hide that gap:
- Nobody rewrites the case repository after sprint three, so a green tick describes software that shipped hundreds of commits ago.
- A spreadsheet row keeps no build. Tick it on Tuesday and by Friday you cannot say what it was ticked against.
A release you can reopen
A release here records a decision. It names the commit it covered, the environment it ran against, and the person who called it safe. Open one from six months ago and you still see the exact wording of every test at the moment someone executed it, the evidence they attached, and any blocking check a human chose to override. Edit that case today. Last month's evidence does not move.
One loop, both kinds of testing
Your CI pushes automated runs in. Your team works the manual regression pack here, keyboard first, attaching evidence as they go. Both land against the same versioned catalogue, so the release page shows what passed, what nobody has run yet, and what is still blocking, without anyone reconciling two systems by hand.
AI evidence you can decide on
Model output arrives as evidence like any other result: the model, its input and output, the score with the scale it was measured on, latency, and a link to the trace. Set a threshold on a tag you govern and a score below it blocks the release rather than sitting in a report. Ship against a model the release never declared and the page says so.
Agents read the same graph over MCP and record results under their own identity, so the audit trail keeps what an agent claimed separate from what a person confirmed.
You do not pay for people who only look
Viewer seats are free and unlimited, on every plan. A product manager who opens a release page twice a week, a developer who reviews a run before approving a PR, an auditor who reads the trail — none of them costs a seat. You pay for the people who author cases and record results, and for nobody else.
Leaving is a documented procedure
One authenticated call returns every row your organization owns — catalogue, case versions, executions, releases, milestones, quality policies and the full audit trail — as JSON, or a table at a time as CSV. A second call reads it back into an empty organization, which is how we test that the first one is complete.
We would rather you could leave and choose not to, than stay because the exit costs more than the renewal.
Already on TestRail? The importer brings your cases, sections, milestones, and run history across, then gives you a reconciliation report for sign-off.
Next Friday at 4:52 you will know exactly what was tested against that commit, and be able to prove it a year later. The 2am call goes to someone else's product.