Methodology

How a score is built

Each product is scored against a published rubric of weighted pillars and criteria. Specifications, declared capabilities and documented evidence states map to reference bands. Pillars and the overall score use weighted averages. The arithmetic is deterministic; the choice of criteria and weights is an editorial methodology decision. No product receives an individual discretionary adjustment.

What happens when a spec is missing

Every criterion declares a missing mode. Floor supplies 40 points when the recorded value is missing. Neutral supplies 50 points. Both are scoring assumptions, not measured product performance. A missing record does not by itself prove manufacturer non-disclosure. Exclude removes the criterion and renormalises the remaining weights within its pillar. The current rubric can therefore change a score when evidence is absent. Reviewed absence and genuine non-applicability must be read with the source notes; neither should be mistaken for a measured result.

Classes, not one leaderboard

Products are sorted into market classes (e.g., Premium European, Performance & Value) and judged within those classes. A boutique premium product should not be ranked against a volume residential product on scale, and vice versa. The class system prevents misleading cross-segment comparisons while still publishing all scores.

International publication

International safety standards and regional approval records have different scopes. The current inverter rubric rewards the number of evidenced national grid-code markets; it does not treat certification breadth as proof of better electrical performance. Local approval and availability must still be checked separately. Any future change to this scoring rule will be versioned and reviewed before publication.

Evidence and market context

Every audit-sheet row is labelled as sourced, reviewed absence, estimated, derived, or not found. Reviewed absence means the linked exact-product material was checked but did not publish the qualifying detail. It is evidence of non-disclosure in the reviewed material, not proof that a capability can never exist. Not found is reserved for rows where no direct source has been located. An estimate or missing source is never presented as a verified fact. Product scorecards also state when their market review was last checked. A product marked market review not yet completed has no implied availability, certification, warranty or installer-compatibility claim for your location: confirm those points before purchase.

Why no product scores 100

Each published rubric fixes its own reference bands. The composite mixes specifications, engineering proxies and warranty or support evidence; it is not a physical measurement of distance from perfection. Adding a new product does not recalibrate existing scores. Changes to reference bands or weights require an explicit rubric release.

Ageing technology: successor flags and the legacy tier

A newer-generation notice is informational, not proof of replacement. When sourced evidence establishes that an exact model is superseded or retired, we seal its last accepted assessment: rubric, inputs, pillar scores, audit sheet and subscores. It leaves current rankings and is not rescored under a later rubric. Its historical URL remains available. Current safety, recall and support notices remain separate from the frozen assessment. A regional withdrawal alone does not establish global retirement. Corrections require a recorded review rather than silently rewriting history.

What the score represents

Every score is tied to a named model or configuration. For panels, thin-film modules and inverters, the benchmark uses the highest published output only inside one physical, electrical and market-certification series; a lower model stays visibly in research until the maximum model's variant-sensitive evidence has been checked. Batteries instead show the supported capacity range and retain a named reference configuration, because a larger stack is not inherently a better battery. Plug-in ESS records score one exact model and publish grid-injection, backup and storage metrics separately. Regional ranges are kept distinct from the global family so an Australian maximum is never silently presented as a worldwide one.

External data sources

Scores draw on: manufacturer datasheets and warranty documents (primary sources, archived copies hosted under /sources/); IEC standards (61215, 61730, 61701, 62716, 62109, 62619, TS 62804); UL 9540A and UL 1741; AS/NZS 4777.2; independent test programmes: PVEL PV Module Reliability Scorecard, RETC, DNV battery scorecard, HTW Berlin Stromspeicher-Inspektion (RSP/SPI), ITP Renewables; BNEF Tier-1 lists (displayed, never scored); and peer-reviewed literature cited in individual rubric changelogs. What each of those third-party sources actually establishes, and which decision it answers, is set out in Three names on a spec sheet.

Methodology change log

Every rubric change is versioned and dated. Expand a rubric for its full history.

Cell balancing: active vs passive

Passive balancing dissipates excess charge as heat through resistors. Active balancing moves charge between cells or modules instead of simply dissipating it. The distinction can matter over long service life, but it is often poorly disclosed. A module optimiser, pack-level control or generic BMS claim is not treated as proof of active cell-level balancing. Where exact architecture is not published, the audit row records a reviewed absence and the rubric's non-disclosure band applies.

Warranty degradation guarantee: the clause most consumers never read

An end-point-only warranty (e.g., "≥70% capacity at year 10") permits a battery to lose 25% of its capacity in year two and then stabilise. The manufacturer satisfies the warranty but the customer suffers large early losses during the years of heaviest use. A stepped guarantee adds a midterm checkpoint. A linear annual-cap guarantee limits degradation per year and prevents front-loaded decline. Only Sonnen currently offers a linear degradation guarantee in this product set. Tesla offers stepped midterm + end-of-term. All others are end-point only.

Grid-forming vs grid-following

A grid-following battery needs an external AC voltage and frequency reference to operate. It cannot function without the grid or a running generator to synchronise to. A grid-forming battery creates that reference itself: it can black-start a home from a complete outage, run indefinitely from solar, and synchronise a generator or additional solar to its own microgrid. Sonnen, Tesla Powerwall 3, and Enphase IQ Battery (with IQ System Controller) are grid-forming in this product set. Most modular batteries (BYD, LG, Sungrow SBH, Huawei Luna, LG RESU) are grid-following. Their backup capability depends entirely on the paired hybrid inverter.

Gateway disclosure: not a penalty

Some batteries require a separate gateway or hub for full smart functionality (VPP dispatch, load management, blackout protection). Where this applies, the scorecard shows an amber gateway note. Gateway dependency is not penalised in the score. A gateway mounted at the switchboard can be the correct topology when the battery is installed in a garage, shed, or any location remote from the main electrical panel. The gateway note is a disclosure so the consumer knows a separate box exists and understands its role.

Blackout protection: a standalone sub-score

Outage resilience is decision-critical for many battery buyers, but the capabilities that define it sit in three different pillars of the main rubric, so no single pillar score answered the question. Battery scorecards therefore carry a separate blackout-protection sub-score (0–100, shown with the product's rank and the fleet median). It re-weights sourced specs already scored in the main rubric around that one question: backup coverage tier, switchover continuity, surge/motor-start strength, backup power depth, and off-grid solar recharge. It is a standalone benchmark: it never feeds the overall product score, and the same missing-data rules apply (non-disclosure floors at 40; not-applicable criteria are excluded and re-normalised). The sub-rubric is published as blackout_rubric_v1 with its own change log.

Semiconductor technology: why it matters

The current rubric gives semiconductor technologies different reference bands. This is an engineering proxy, not a measured guarantee that every implementation performs better. Peak, weighted and partial-load efficiency need their own exact-device evidence; a future outcome-based profile must avoid counting the same advantage twice.

Night standby consumption: the silent lifetime cost

An inverter drawing 10 W for 11 hours each night consumes approximately 402 kWh over ten years. Actual standby consumption and duty vary by device. The current rubric assigns missing standby values 50 points; this is an assumption, not a measurement or a guarantee that missing evidence has no effect on the total.

Reactive power range: voltage management on the real grid

Reactive-power range, voltage-response functions and operating limits need exact-device evidence. A power-factor range alone does not establish every grid-support function. Network settings and regional requirements remain separate from a globally published specification score.