Scoring Model Validation

We tested all 30 score-affecting rule IDs with controlled positive and negative cases. Numeric boundary cases were retained where relevant.

Hypothesis: each documented rule should trigger under its positive condition and remain absent under its paired negative control, with expected coupling recorded separately.

1. The original validation evidence was retained unchanged.
2. A separate isolated study introduced six neutral mode baselines.
3. The twelve remaining mismatches received replacement pairs only where fixture isolation was defective.
4. Replacement cases ran twice in fresh Chromium contexts against the production artifact.
5. Target findings, dimensions, deductions, boundaries, and collateral findings were compared.

Severity was not inferred. The retained raw output does not expose a direct severity field, so severity remains explicitly unobservable.

The published result traces to the corrected v2.0 fixture set, two retained isolated executions, and the final 30-rule disposition. Dataset checksum: 20680a78a8d6b0c89b1ed8b8c4a1125f510e6064b32452fda6807df45773d2ea. Final disposition checksum: b8eb2eaa9915f4655deacb2ad4a17ad0d94cb2ed881a9cbbd1c3f5d09f433164.

The original 66-case study and the first isolated study remain retained as historical evidence. They are not replaced by this corrected result.

This supports the narrow claim that the documented v2.0 trigger behavior can be reproduced with retained controlled inputs. It does not show that a higher score improves rankings, traffic, conversions, or indexing.

The original failures remain part of the research record. They show why neutral baselines and explicit collateral classification are required.

• Fixtures are synthetic and text-based.
• Rule coupling can produce collateral findings.
• Severity is not observable from retained raw output.
• No live pages, SERPs, Search Console data, rankings, or outcomes were tested.
• The result is tied to Scoring Model v2.0 and must be rerun separately after a model change.

The retained manifest, fixtures, raw executions, and final disposition are stored with the repository research contract. The remediation command is:

node scripts/labs-scoring-remediation.mjs --run

Read the Scoring Model and Limitations before interpreting a result.