Website Audit Engine

1) The problem this lesson solves

A single “site health” score is not an audit. Without target URL, scope, tool version, raw finding and execution context, two runs are not reproducibly comparable and remediation cannot be proven.

2) Lesson boundary

The owned artifact is Website Audit Evidence Engine Contract v1. This lesson does not own CRM handoff or niche research, and it does not crawl or mutate a live client site.

3) Fact: what Lighthouse is

Lighthouse is an automated page-auditing tool with categories such as Performance, Accessibility, Best Practices and SEO. A failed audit is an improvement indicator, not a complete verdict on website quality.

4) Fact: do not use one score as the performance oracle

Chrome recommends the Performance panel for deeper performance debugging. Treat a Lighthouse performance score as summary evidence, not as the only diagnostic truth.

5) Inference: reproducibility depends on context

Chrome documents that setup, device conditions, extensions and throttling can influence results. Each run therefore needs explicit scope, execution context and tool/version provenance.

6) Design: finding-level evidence

MAKNOW gives each finding an identity, raw evidence, severity, confidence, automated/manual provenance, owner, remediation and rerun. Aggregate scores summarize; they do not replace findings.

7) Lock scope before interpreting scores

target_url, canonical_page_id and crawl_scope define what the finding applies to. A fix deployed to one template cannot verify the entire site.

8) Preserve tool provenance

Store tool_name, tool_version and check_id. If a new tool version removes a check, absence is not equivalent to remediation.

9) Keep raw result separate from severity

Historical raw evidence stays stable. Severity and confidence can change after policy or manual review without rewriting the original observation.

10) Automated and manual evidence are different

An automated accessibility or SEO signal does not fabricate human review. automated_or_manual and reviewer evidence preserve accountability.

11) A composite score cannot hide a critical finding

If the aggregate score rises from 72 to 95 while a critical finding remains unresolved, the audit is not verified. Finding-level blockers are non-compensable.

12) Core Evidence Contract

The artifact has 22 dimensions, 18 critical. The compact table below shows a traceable baseline; JSON/CSV retains the full field set.

13) Verification means rerunning the evidence-producing check

After remediation, preserve a new rerun_ref and compatible comparison context. A higher score alone is not verification.

14) False positive and dispute states

A manual dispute requires reviewer identity and evidence. Without both, a finding cannot be promoted to FALSE_POSITIVE; it remains DISPUTED.

15) Lab and transfer

The lab has eight fixtures and 48 assertions. In transfer, a diagnostic changes, only one page template is fixed, and a reviewer disputes a high-severity finding. Recompute from changed evidence instead of copying the old score.