All categories · July 2026

An estimate and a datasheet figure used to score the same. They no longer do.

Every criterion on this benchmark carries an evidence state: sourced, derived, estimated, or not found. Until today that state was published but did not affect the score. A figure read off a datasheet and a figure we estimated from published context scored identically. On a benchmark whose position is that sourcing is the product, that was a hole, and this closes it.

The rule

An estimate keeps its band, then moves a quarter of the way toward the floor.

The rubric already had a floor for non-disclosure: a criterion a manufacturer does not publish at all scores 40. An estimate now sits between that floor and its own band, 25% of the way down. An estimated band of 100 scores 85. An estimated band of 60 scores 55. The discount shrinks as the band falls, because a low band is already saying most of what needs saying.

The estimated set is read from the same provenance tagging that labels the audit sheet on every scorecard. A criterion shown as "Estimated" in the evidence column is the same criterion the scorer discounts, so the page and the number cannot disagree.

Why an estimate is discounted and not discarded

The rule is meant to penalise opacity, not novelty.

A product new to market often has real published figures and no full datasheet yet. Discarding those values would score a recent product as though its manufacturer were withholding information, when the more likely explanation is that the documentation has not caught up. Equally, an estimate carries genuine uncertainty and should not read as though someone measured it. A quarter of the distance to the floor is our answer to both, and it is a judgement rather than a calculation.

What moved

37 of 122 scores changed. None by more than five points, and no product changed class.

review.solar

Products whose score moved, by category

Panels14

of 51

Inverters14

of 34, the most affected class

Batteries7

of 28

Plug-in ESS2

of 8

Chart data

Series	Value	Note
Panels	14	of 51
Inverters	14	of 34, the most affected class
Batteries	7	of 28
Plug-in ESS	2	of 8

Inverters moved most because inverters carry the most estimated criteria, at roughly 28% of scored rows against 16% for panels. That is a fact about how well the class is documented, and about our own evidence work on it, rather than a statement about inverters as products. It is also why the inverter catalogue is the current priority for re-sourcing.

What this does not change

No band was rewritten, and no product was rescored by hand.

The band tables are untouched. What changed is a single rule applied uniformly by the engine to every criterion in every category, and it is applied identically to every manufacturer including those stocked by the retail group disclosed on our disclosures page. A product's score can be restored to its pre-change value by turning the rule off, which exists so a historical figure can be audited rather than as an operating mode.

If a manufacturer believes a criterion is recorded as an estimate when a published document states it, that is a correction we want. The evidence correction route is the fastest way to move a criterion from estimated to sourced, and doing so raises the score automatically.

Sourcesrubric versions and dataset

• Rubric versions carrying this change, all dated 31 July 2026: panel 2.9.0, thin film 2.9.0 (inherits the panel rubric), battery 2.9.0, inverter 3.6.0, plug-in ESS 1.2.0. Each rubric's changelog entry states the rule. See the methodology page for current versions.
• Score movement measured by building the published dataset with and without the rule on 31 July 2026, across all 122 active products: 37 changed, maximum movement 5 points, no classification change, no product gained.
• Estimated-criteria share by category, from the published dataset: inverter 28%, battery 17%, panel 16%, plug-in ESS 1%, thin film 52% (one product).