Coverage & scores

Coverage

Also: measurement coverage, N / M measured

How many leaf score components have a real (non-null) value for this run.

Coverage is a completeness meter, not a quality score. Each leaf component is either measured or left null.

In v1 the leaf set includes fetch health, crawler access, raw HTML quality, JS dependency, structured data, fact parity, fact recovery cost, entity coherence, provenance, freshness, salience, indexability, discovery, and search interpretation.

Search interpretation is always null in v1 (deferred). That alone can keep coverage below 14/14 even on a perfect run.

On the report Truth strip · Coverage

• Coverage is not “how good the page is.”
• Unmeasured is not FAIL — it is UNKNOWN.

Composite

Also: machineIngestion, machine ingestion score

Mean of measured leaf scores only. Null leaves are excluded from the average.

Composite answers: among the leaf measurements that actually ran, what is the average strength?

It never invents mid-scale placeholders for unmeasured leaves. A 94 with 12/14 measured means two leaves contributed nothing — not that they scored well.

Derived convenience scores such as AI readiness are excluded from the composite mean.

On the report Truth strip · Composite / 100

• Not an SEO rank predictor.
• Not an “AI visibility” product score.
• Not proof that unmeasured categories were observed.

Critical (UI) vs fail vs CRITICAL

Three different “critical” meanings that must not be collapsed.

In the report chrome, Critical counts findings with severity fail. The UI says “critical” because that is how operators triage.

In the engine, Finding.severity uses info | warn | fail. Evidence panes still show the engine severity string.

Separately, AtomicFact.criticality uses CRITICAL | IMPORTANT | SUPPORTING | INCIDENTAL — how important a fact is to recover, not whether a check failed.

On the report Outcome line · Critical count · Evidence severity

• UI “Critical” is not the same field as fact criticality CRITICAL.

Unknown count

How many gaps are still open: unmeasured leaves, skipped leaves, and render unavailability when applicable.

The Unknown number in the truth strip mixes coverage gaps (unmeasured / skipped) and, when the browser step did not run, a render-unavailable marker.

It is a gap inventory. It is not a severity. Gaps stay gaps until a later contract measures them.

On the report Truth strip · Unknown

AI readiness

Derived blend of crawler access, structured data, fact parity, and JS dependency — only when all four are measured.

AI readiness is a convenience readout for operators who want a single “can machines use this?” blend of four related leaves.

It is excluded from Composite so it cannot double-count those leaves in the mean.

If any of the four inputs is null, AI readiness stays null.

On the report Measurement components · AI readiness bar

• Not a claim about ChatGPT/Perplexity citation likelihood.

Search interpretation

Also: v2

Deferred leaf. Always unmeasured in contract 1.0.0.

Reserved for live Search Console / SERP / answer-surface interpretation. The leaf exists so coverage stays honest about what v1 does not observe.

Seeing it listed under unmeasured is expected, not a site defect.

On the report Coverage unmeasured · Measurement components · label v2

Fetch health

Whether the default fetch produced usable content, with the fetch verdict as the bar label.

When the HTTP classification says usable: PASS-like verdicts map toward 100; WARN toward 70. Non-usable maps toward 25.

If classification is missing, the leaf stays null.

On the report Measurement components · Fetch health

Fact recovery cost

How expensive it is for a machine to recover important facts from the page representation.

Each atomic fact carries recoveryCost on a 0–10 effort scale (higher = harder). The scorecard converts the mean of CRITICAL facts (or all facts if none are CRITICAL) into a 0–100 leaf where higher is better: roughly 100 − meanCost × 12.

In the Fact ledger, the recovery column shows the per-fact effort ticks. In Measurement components, the leaf is the inverted goodness score.

On the report Fact ledger · recovery · Measurement components

• Do not read the leaf score as “cost dollars.” Higher leaf score means cheaper recovery.

Leaf components

The atomic measured scores that can feed Composite. Null stays null.

Leaves are independently measured. Derived scores may hang off them, but Composite only averages measured leaves.

Skipped modes (no crawler matrix, no browser) force related leaves to null rather than inventing values.

On the report Measurement components