Mini Split Ratings: Evidence and Selection Method

Start with the equipment role

A mini split listing may sell a complete matched system, an indoor head, a condenser, several independent systems or an installation accessory. We classify this role before applying a capacity filter. Compatibility numbers printed on a mounting bracket cannot qualify it for a heat pump ranking. Multiple single-zone sets remain in a separate bundle collection rather than being treated as one multi-zone combination.

Three distinct buying questions

The 12,000 and 18,000 BTU lists require the exact nominal capacity in a single-zone listing title. The multi-zone shortlist requires a listing identifying several zones. Each ranked reference must be a complete system with a captured public average and positive review count for its exact ASIN. Unreviewed references remain searchable in the catalog. A nominal BTU label does not verify capacity at winter design temperature; indoor-head totals do not establish simultaneous condenser output.

Listing attributes and documentation

The listing-attribute component is 100 × [0.40 × (nominal BTU capacity documented ? 1 : 0) + 0.40 × (SEER2 stated ? min(SEER2 / 30, 1) : 0) + 0.20 × (supply voltage documented ? 1 : 0)], rounded to a whole number. Capacity does not earn more points simply for being larger. This component is available only for a complete system with documented nominal capacity. SEER and SEER2 are never interchanged, and nominal capacity is not cold-weather output. This is a listing-based index, not an HVAC performance or suitability test.

Documentation completeness = 100 × [0.5 × min(feature bullets / 8, 1) + 0.3 × min(listing images / 5, 1) + 0.2 × (brand stated ? 1 : 0)], rounded to a whole number. This fallback describes the stored record, never the heat pump performance.

We store the matching source phrase for every extracted capacity, SEER2 or voltage value. SEER is not silently converted into SEER2. A model with contradictory configuration evidence must be withheld from the affected shortlist until resolved. Our equipment index intentionally does not reward a larger nominal capacity merely for being larger: oversizing is not an improvement.

The public-review component

Where Amazon publicly displays customer reviews for a listing (average star rating and review count captured from the public product page), the raw editorial score is 100 × [0.45 × (Amazon average / 5) × min(review count / 50, 1) + 0.35 × min(log10(1 + review count) / log10(1 + 5000), 1) + 0.20 × documentation score / 100]. The confidence factor min(review count / 50, 1) damps averages based on very few reviews, and the logarithmic volume term rewards large public review volumes, so widely and highly reviewed listings rank ahead of thinly reviewed ones. Listings with no captured public review data use the documentation score alone. This combined index reflects the public Amazon reputation and the listing documentation — it is not a physical test and not our own product testing.

Only review-backed complete systems qualify for the three ranked lists, even though the general helper can calculate a documentation-only catalog score. The stored count is what the exact-ASIN Amazon page displayed. Amazon can group reviews from variants, so this does not prove each review concerns the selected configuration. We do not write customer testimonials or estimate missing counts.

Reading OUR RATING

The displayed editorial rating is the raw blended score divided by 10, shown to one decimal on a 0-10 scale. There is no catalog normalization, rank interpolation, forced winner, forced floor or artificial gap. Equal scores remain equal: the available data does not justify a distinction. Ranking uses the unmodified raw score descending, then listing slug for exact ties; slug is a stable identifier, not a quality signal. The same raw score always has the same display rating in every shortlist. This is not a physical test result or an Amazon customer rating.

The 45/35/20 blend combines public review reputation, review volume and the attribute/documentation component. It is not an Amazon star average. Two models sharing a rounded score remain tied; an identifier tie-break gives a repeatable list, not a hidden judgment about compressor quality. Sort controls offer alternative orders without rewriting the editorial selection.

What expires and what stays a snapshot

Price, availability and product photographs have a maximum 24-hour life from API retrieval. Client-side guards omit expired values rather than retaining historical prices. Review observations and comparator scores are research snapshots updated through a later evidence collection and rebuild. Generated editorial scenes are labelled and are never photographs of an identified commercial model, a field installation or a performance test.

Questions the rating cannot answer

It cannot calculate your building load, approve electrical supply, select refrigerant procedures, promise low-temperature heating output or confirm an installation warranty. Take the exact indoor and outdoor model numbers to a qualified HVAC contractor. For a correction, send the URL, ASIN and conflicting source to our editorial contact. We will compare the underlying record rather than adjusting a score to preserve a preferred winner.