Observed Health
Uses successfully tested checks only and stays hidden below the evidence threshold.
Evidence-first methodology
See how evidence becomes a recommendation.
Every finding keeps the public observation, fixed check, sampled scope, confidence, limitation, next action, and verification path connected.
The finding model
One evidence chain keeps the decision traceable.
Follow the main decision from observation to verification. Scope, confidence, and limitation stay attached without competing with the merchant’s next action.
Observation
The public value retained from a sampled page.
Fixed check
The same documented rule is applied to the same kind of evidence.
Finding
The supported issue that follows from the observed evidence and fixed check.
Recommendation
The lowest-risk supported next action.
Verification
How to run the same check after a change.
Three separate measures
Performance, completeness, and evidence strength answer different questions.
Evidence Coverage
Untested eligible checks stay in the denominator, so incomplete scanning cannot look complete.
Finding Confidence
Describes evidence strength. It is separate from severity and expected business impact.
Open the registered formulas
- Observed Health
Σ(weight × score) / Σ(tested weight)- Evidence Coverage
100 × tested eligible weight / eligible weight- Finding Confidence
100 × Σ(weight × evidence coefficient) / Σ(weight)
Traceable by design
A recommendation leads back to retained proof.
Evidence keeps a source, location, observation, timestamp, rule or formula version, assumptions, exclusions, and limitations. Failed, skipped, unknown, and unavailable states stay visible.
- Page
- Template
- Test
- Evidence
- Finding
- Next step
Technical detail when you need it
States, confidence, transfer math, and verification.
Observed, scenario, verified, and unavailable
- Observed
- Direct retained evidence from the bounded sample.
- Scenario
- User-entered inputs with visible formulas, never relabeled as measured.
- Verified improvement
- A later before-and-after result that passes a defined acceptance test and guardrails. This product currently claims none.
- Unavailable
- Evidence the scan could not inspect. It is not a zero, warning, or pass.
Score suppression and confidence classes
Observed Health disappears when Evidence Coverage is below the configured threshold. Finding Confidence uses the registered evidence classes and weights. Direct observation is stronger than a weak fingerprint, but neither is a guarantee of business impact.
Image transfer example and unit rules
The supplied calculator case compares 6.80 MB with a 3.25 MB validated candidate: a 3.55 MB measured difference. At an entered 8 Mbps, transfer math yields a 3.55-second ceiling. At 120,000 entered views and 15% cache misses, the monthly origin scenario is 63.9 GB.
Measured difference3.55 MB
MB and GB are decimal, and exact integer bytes are stored. None of this predicts LCP, PageSpeed, conversion, orders, or revenue.
Verification workflow
- Preserve the baseline evidence, versions, scope, and limitations.
- Make the narrowest reversible change in a duplicate or development theme.
- Repeat the same rule against the same URLs and evidence boundary.
- Report verified, unchanged, regressed, or inconclusive and check guardrails.
What public evidence cannot establish
Public scanning cannot prove private configuration, app value, indexing, analytics outcomes, customer behavior, revenue, conversion, orders, search demand, or real-user performance.
Claim boundary
A change is not an outcome.
No finding promises PageSpeed, LCP, conversion, orders, revenue, indexing, ranking, or answer-engine visibility. Those outcomes need their own measured evidence.