An unknown check is different from a failed check. A page might have a broken canonical, or the reviewer might lack access to inspect its canonical at all. Collapsing both into zero hides the distinction between a defect and an unfinished review.
Use explicit states: pass, fail, unknown, and not applicable. Report how much of the applicable review has evidence before presenting a score. A high score based on a small visible fraction should not imply that the whole site is ready.
The arithmetic below is a deliberately simple original method for an internal readiness review. It is not a score endorsed by an answer engine, a ranking forecast, or a claim about how AI systems choose citations.
Define the states before reviewing pages
A pass means the available evidence satisfies the stated check. A fail means observed evidence does not satisfy it. Unknown means the reviewer lacks enough evidence to decide. Not applicable means the check legitimately does not apply to this page or product, with a written reason.
For example, an active purchase flow might be not applicable to an informational article. It should not be not applicable to a product page merely because the checkout is unfinished. In that case, the actual check and intended release state determine whether the result is fail or unknown.
Write each check as an observable question. “The page feels authoritative” is difficult to apply consistently. “The pricing claim matches the approved offer record” is more specific, provided the reviewer has both pieces of evidence.
Work through twelve checks
Suppose an internal review starts with twelve equally weighted checks. Two are legitimately not applicable, leaving ten applicable checks. Of those ten, six pass, two fail, and two remain unknown because the reviewer lacks required access.
| State | Count | Meaning |
|---|---|---|
| Pass | 6 | Evidence satisfies the check |
| Fail | 2 | Evidence identifies a defect |
| Unknown | 2 | Evidence is still missing |
| Not applicable | 2 | Excluded with a recorded reason |
Among the eight completed applicable checks, six pass: 6/8, or 75%. Review coverage is 8/10, or 80% of applicable checks. Report both numbers together. The 75% figure alone makes an incomplete review look more complete than it is.
If both unknown checks eventually fail, six of ten applicable checks will pass: 60%. If both eventually pass, eight of ten will pass: 80%. That 60–80% range is an arithmetic boundary for these unresolved checks, not a statistical confidence interval.
Do not hide critical issues inside the average
Some checks should block a specific release decision regardless of the average. An incorrect payment destination, for example, should not be canceled out by several well-written headings. Define those critical conditions separately from the summary score.
In the fictional review, suppose one unknown concerns whether the product can be delivered after payment. The release recommendation should say that fulfillment remains unverified and sales should not open until it is resolved. Reporting 75% observed pass rate does not authorize ignoring that missing evidence.
Keep severity and evidence state separate. A known minor typo is a fail of low consequence. An unknown commercial workflow can have a much larger consequence. The action list should reflect those differences rather than sorting only by the score contribution.
Make the denominator stable across reviews
If the rubric changes from twelve checks to twenty, an apparent score improvement may reflect the new mix. Preserve the rubric version and show a comparable subset when reporting a trend. Do not quietly remove difficult checks by relabeling them not applicable.
When a check becomes inapplicable because the product changed, record the reason and the release decision. If a site closes checkout, a payment test may no longer apply to the current public experience, but the team should still retain the commercial work needed before reopening it.
If weights are useful, define them before inspecting the result and show weighted coverage as well as the weighted pass calculation. Avoid adding elaborate weights merely to produce a more impressive score. A simple transparent rubric is often easier to act on.
Turn unknowns into an evidence request
For each unknown, name the missing evidence, the person or system that can provide it, and the decision it blocks. “Need access to the configured payment item” is more actionable than “checkout unclear.” Keep these requests separate from the list of confirmed defects.
Once evidence arrives, update the state and preserve the previous review record. A newly discovered failure is not necessarily a regression; it may be a problem that was previously unknown. Explain that distinction in the revision note.
Use the readiness-score introduction to frame the result. A useful score makes the evidence easier to understand and the next action easier to choose. It should never make uncertainty disappear.