How many of your test steps can actually be verified?
Paste manual test cases. The checker reads each step and tells you which ones state a checkable expected result, which ones only describe an action, and which ones say “verify” without saying what. Nothing you paste leaves your browser.
Analysis is a text heuristic that runs entirely in this page. No upload, no account, nothing stored.
- 0 verifiable
- 0 vague check
- 0 action only
A passing run is not a verified run.
An automated run can only confirm what the test says should be true. A step like “Click Save” with no expected result passes whenever the click happens, whatever the application did next. When Spec2RunAI runs a test case, each step is marked Verified (expected result checked and met), Not verified (ran and passed, nothing confirmed it), or Failed, and the Run Evidence Report leads with the share that was verified. This checker gives you that number before anything runs, from the test cases as written.
Verifiable
The step, or an expected-result line after it, names something specific that can be checked: a message, a status, a value, a screen, a count. “Status changes to Bound” is verifiable. These become Verified or Failed when run.
Vague check
The step uses a checking verb — verify, confirm, check — but names nothing specific. “Verify the page works” cannot fail in any particular way. Rewrite it to say what should be true, and it moves to the first column.
Action only
The step describes what to do and nothing about what should happen. These run and pass whenever the action completes. Add the outcome the tester would look for, in their own words; no special syntax is needed.
What the heuristic cannot see
This page reads text, not your application. It will miss expected results written in unusual ways and may credit a specific-looking phrase that is not checkable on your screen. Spec2RunAI resolves each step against the live page; this is the fast approximation.
About the checker.
Does my test data leave the browser?
No. The page contains the entire checker as JavaScript; nothing is sent to a server. The only thing recorded is an anonymous analytics event with the counts (steps, verifiable, vague, action-only), never the text.
What formats does it accept?
Plain lines, numbered lines, lines beginning “Step:”, “Expected:”, “ER:” or “Result:”, and tab-separated columns pasted from Excel or a test management export where the second column is the expected result. Headings such as “Test case”, “Preconditions”, or “Title” are skipped.
Is this how Spec2RunAI decides what is verified?
Not exactly. Spec2RunAI reads the step with an AI model and checks the expected result against the live screen at run time; this page uses a text heuristic that approximates the first half of that. Treat the number as an estimate of the ceiling: steps with no stated outcome cannot be verified by any tool.
What is a good coverage number?
There is no universal threshold. In our own work with enterprise UAT suites, roughly a quarter of steps arrive with no expected result, so a suite that starts near 75% is typical. What matters is which steps are unverified: a missing outcome on “Bind the policy” costs more than one on “Open the menu.”
Send us ten of these. We will run them and send the Run Evidence Report.
Verification coverage per step, failures by cause, and the AI provenance appendix — from your test cases, as written.