Skip to content

Methodology

A score is a claim. It should come with its derivation.

Heuristic tools produce numbers that move for reasons you cannot inspect. Obcend publishes the ruleset, the weights, and the arithmetic, so any score can be recomputed by hand from the findings list.

Determinism

Why we discard uncertain observations

Every rule executes three times across independent crawl workers. If the three runs disagree, the observation is suppressed and logged as unstable rather than reported as a finding.

This costs coverage. A handful of genuinely intermittent conditions never make it into a report. We accept that trade because a report where one finding in twenty is noise is a report nobody trusts — and an untrusted report changes nothing.

Versioning

Rulesets are pinned, not rolling

A run records the ruleset version it executed. Comparisons across runs are only offered when the versions match, or when the changelog delta is shown alongside the comparison.

Rules are added and retired openly. If your score moves and your site did not change, the changelog will say why.

Scoring

The formula, in full

Composite integrity is a weighted mean of category scores. Each category score is the proportion of rule evaluations that passed, weighted by severity.

composite integrity
category_score = 1 - ( Σ severity_weight(f) for f in failures
                    / Σ severity_weight(r) for r in evaluated )

composite = Σ ( category_weight × category_score ) × 100

severity_weight: critical 8 · high 5 · medium 3 · low 1
  • CategorySecurity
    Weight0.30
    BasisHeader strictness, transport, exposure
  • CategoryAccessibility
    Weight0.25
    BasisWCAG 2.2 A/AA mechanical failures
  • CategoryCrawl integrity
    Weight0.20
    BasisStatus anomalies, redirect depth
  • CategoryIndexation
    Weight0.15
    BasisCanonical and directive agreement
  • CategoryDelivery
    Weight0.10
    BasisBlocking resources, cache correctness

Limits

What Obcend does not claim

Overstating capability is the fastest way to lose an audience that reads carefully.

  • Obcend does not predict rankings, revenue, or conversion outcomes. It reports observable conditions.
  • Obcend does not assess design quality, editorial judgement, or brand fit. Those are not mechanically decidable.
  • Mechanical accessibility checks cover a subset of WCAG 2.2. They do not substitute for manual and assistive-technology testing, and reports say so.
  • A passing score is the absence of detected failures, not a guarantee of correctness.

See a deterministic audit of your own site

We run a scoped crawl, produce evidence-backed findings, and walk your team through the report. No obligation, no pipeline pressure.