Quality
The Quality Team checks all work: the checks inside each Process, every outbound Message, and every money move, with verdicts, never Approvals.
Agents fill its three seats from the first day.
What it owns
- Checks: Evaluator, Guard.
- The eval suite behind every definition an Agent runs: instructions, Skills, Tools and Policies.
- Verdicts, each recorded with its Evidence.
How it runs
The Quality Team wakes with a Startup’s first Change and stays on at every Lifecycle state after. Three seats staff it: the Quality Lead, the Reviewer and the Tester.
Change verdict runs on every new Change. It runs the Change’s checks, replays known-good runs, and sends a synthetic paying stranger through signup, payment and first value. The Nightly quality run repeats all of it at 03:00 and opens an Issue per regression.
A Change to an Agent’s definition passes its eval suite offline before any traffic sees it. It must score no worse than the current Version. Online, it runs as an Experiment with guardrails, and Evals keeps each Worker’s track record.
Every outbound Message and money move at medium stakes or above gets a review before its delay window closes. Guards run first, then Evaluators. A failed review cancels or redirects the Action, with the reason recorded.
No Human sits in the critical path. When a check is wrong, the fix is a better check, shipped as a Change like any other.
Roles
| Role | Definition |
|---|---|
| Quality Lead | The Quality Lead verifies every Team’s checks and gives every Change a verdict, never an Approval. |
| Reviewer | The Reviewer checks every outbound Message and money move at medium stakes or above before its delay window closes. |
| Tester | The Tester runs the synthetic paying stranger and replays known-good runs against every Change before it merges. |
Processes
| Process | Accountable | Workflows |
|---|---|---|
| Verification | Quality Lead | Nightly quality run, Change verdict |
| Founder Run | Quality Lead | Run by its Roles |
Plays
None.
Metrics
| Metric | Definition |
|---|---|
| Check pass rate | Share of checks that pass on the latest run. |
| Escaped defects | Issues Customers report after a Change ships, per week. |
| Review coverage | Share of outbound Messages and money moves reviewed before their delay window closes. |
| Verdict time | Median minutes from a new Change to its verdict. |
| Escaped defects | Escaped defects counts the Issues a Customer reports each week that an existing check should have caught. Owner: Quality Lead. |
| Review turnaround | Review turnaround is the median hours from an Action entering review to its verdict. Owner: Reviewer. |
| Pre-merge catch rate | Pre-merge catch rate is the share of breaking Changes stopped before they merge. Owner: Tester. |