Preserve the source artefact
Record the canonical URL, publisher, source class, retrieval time, product scope and capture state.
The evidence trace connects each scoped conclusion back to the exact claims, assessments and source artefacts that support it. This is a two-product pilot, not complete catalogue coverage.
Two exact productsValidated claim-level traces cover two of 160 catalogue records.
Partial legacy projectionMissing snapshots, hashes and calculation links stay explicit.
No purchase adviceBoth pilot outcomes support ranked comparison only.
A citation identifies a source. A decision trace records what it supports, under which conditions, and whether changing it affects the result.
Record the canonical URL, publisher, source class, retrieval time, product scope and capture state.
Keep the source wording, raw value, normalised value, conditions, exact variant and checked time together.
Record support, contradiction, qualification, derivation, syndication and genuinely independent corroboration.
Evaluate validity, applicability, independence, incentives, freshness, conflict and missing conditions separately.
Show which claims and assessments produced each material score contribution or eligibility gate.
Distinguish ranked comparison, purchase advice, insufficient evidence and blocked outcomes.
Checked 28 Jul 2026, 16:00 BST. Counts come from the validated trace records, not the page render time.
The exact GB model has enough traceable published information for its specification ranking. Controlled cooling and noise evidence, current availability and second-person review are absent.
The exact GB model has enough traceable published information for the category ranking. Delivered energy, sustained output, thermal behaviour and independent review remain unresolved.
It does not establish complete evidence coverage, controlled performance, sensitivity, second-person review or purchase suitability. One pilot trace does not retain a criterion contribution ledger. Across the pilot, 5 gaps still block purchase advice.
Authority is contextual. No source receives one universal trust score that settles every product question.
| Evidence class | Strongest normal use | Must not be treated as |
|---|---|---|
| Manufacturer specification or manual | Identity, dimensions, declared modes, warranty terms and published specifications. | Not independent performance evidence. |
| Retailer observation | Exact SKU price, stock, delivery scope and checked availability. | Not evidence of product quality. |
| Regulatory or certification record | Compliance inside the formal scope of the record. | Not a general performance endorsement. |
| Controlled measurement | The property measured under the recorded equipment, mode and conditions. | Not automatically valid for every real-world use. |
| Independent editorial test | Measured and observed context from a named external method. | Not an Agent Review Data test. |
| Owner report or discussion | Failure-mode discovery and qualitative ownership context. | Not a representative estimate of prevalence. |
| Commercial review pool | A source-specific native rating and issue-discovery pool. | Not internet-wide consensus. |
“Missing evidence should weaken the conclusion, not disappear from the page.”
The current catalogue remains useful for ranked specification comparison. Claim-level traces will expand in bounded cohorts as source artefacts, assessments and calculation links are validated.