How uatgpt.com Keeps Evidence and Defects Tied to the Expectation

uatgpt.com keeps evidence and defects tied to the expectation by treating a drafted test, an observed result, and a release decision as three separate events, each anchored to the product actually checked. You record what should have happened next to what did happen, and you make a failed path reproducible by naming the path and build that were checked. This applies when you are preparing acceptance evidence and a release conversation; the site frames criteria and organizes notes, but it does not decide whether a product meets them.

The separation that makes evidence trustworthy

The site's own framing is explicit: "A drafted test, an observed result and a release decision are different events. Keep each one tied to the product you actually checked." That sentence is the mechanism behind everything else on the page.

• Drafted test — the criterion you wrote before running anything. It states the actor, starting condition, action, and visible outcome.
• Observed result — what you actually saw when you checked that criterion on a specific path and build.
• Release decision — the judgment you make after reviewing evidence, not a substitute for it.

Keeping these apart prevents the most common failure in acceptance work: letting a well-written test stand in for a result, or letting a release decision quietly rewrite what was observed.

Keeping observed results beside the expectation

The Evidence stage uses observation notes to hold the expected outcome and the observed outcome in the same place. The page's criterion frame defines what belongs in the expectation:

The frame also includes the starting state, so the full criterion reads as actor, starting condition, action, and visible outcome. The site notes that this frame "exposes missing parts; it does not decide whether a product meets them" — which is exactly why the observation note has to sit beside it rather than replace it.

The page makes the point concretely: "Intuitive" cannot be observed in the same way as a named message or changed state. If your expectation is a named message or a changed state, your observation note can record whether that message appeared or that state changed. If your expectation is a feeling, there is nothing to record against.

Making a failed path reproducible

The Defects stage uses reproduction steps and asks: which path and build were checked? A failure is only useful to a release conversation if someone else can retrace it. That means the defect record should carry:

• The path that was exercised
• The build that was checked
• The expectation it failed against (traceable back to the criterion)
• The observed result, kept as observed

The Coverage stage supports this with trace links that find gaps between criteria and tests, so a defect can point to the specific acceptance statement it contradicts rather than to a general impression.

Why the tie to the actual product matters

Every stage on the site carries the same constraint. The Brief stage turns a product request into acceptance conditions; the Paths stage writes scenarios with observable outcomes; Coverage maps criteria to tests; Evidence keeps results beside expectations; Defects makes failures reproducible; Release reviews evidence before a decision. Each stage is a local draft, and the page states plainly that drafts stay on the page.

The practical consequence: when you move from a drafted test to an observed result, you are recording something about a named path and build — not about the criterion in the abstract. A release decision built on that record can be defended; one built on a well-worded test cannot.

A worked example

Suppose a request says users should get clear feedback after submitting a form.

1. Brief — Draft the criterion: a signed-in user (person), on a completed form (starting state), submits it (action), and sees a named confirmation message (visible outcome).
2. Paths — Write the scenario with that observable outcome rather than "feels responsive."
3. Coverage — Link the criterion to the test so gaps are visible.
4. Evidence — Run it on a specific build and path; note beside the expectation whether the named message appeared.
5. Defects — If it did not, record the path, the build, and the observed result so the failure can be retraced.
6. Release — Review the evidence as recorded, and make the decision as a separate step.

At every point, the record stays attached to the product you actually checked. That is the whole method: the expectation is written so it can be observed, the observation is kept next to it, and the failure names where and on what build it happened.