SplitAtlas

Product scoring methodology - SplitAtlas

Principle: scores are computed from normalized structured product data. No hand-tuned per-product numbers and no sponsorship influence. Published pages show the methodology link, eligible category bands, excluded categories, and a data-completeness indicator.

Scale

SplitAtlas scores use an absolute 6.5-10.0 display scale. Every catalogued system is certified equipment, so visible scores run from solid to exceptional rather than from bad to good. Category calculations start on a 0-100 normalized scale, then the overall weighted result is mapped into the public 6.5-10.0 score.

Current methodology version: scoring-methodology.v2.

Changelog

  • 2026-08-27 - scoring-methodology.v2: Category scores changed from fixed min/max-style anchors to percentile ranks among scored published peers. For categories where capacity changes the fair market baseline, the comparison is made within standard capacity buckets; otherwise it is made across the scored catalogue. Scores moved because the comparison changed. The public 6.5-10.0 overall display scale and category weights did not change.

Payback calculator v2

Current payback methodology version: payback-v2. Executable version: payback@2.0.1; input and output schemas: payback-input@2.0.0 and payback-output@2.0.0.

Payback v2 computes operating-cost savings from residential energy prices. It does not look up a savings percentage for a fuel type. For heating fuels, the calculation normalizes both systems to one million Btu (MMBtu) of heat delivered to the room:

existing delivered $/MMBtu
  = price per fuel unit / (Btu per unit / 1,000,000) / efficiency

seasonal COP = HSPF2 x 1,000 / 3,412 Btu per kWh

mini-split delivered $/MMBtu
  = electricity $/kWh / (3,412 / 1,000,000) / seasonal COP

savings fraction
  = 1 - (mini-split delivered $/MMBtu / existing delivered $/MMBtu)

Gas and propane furnaces use an 80%-95% AFUE comparison range, based on the DOE FEMP residential-furnace comparison. Oil uses 80%-87%: DOE's home-heating efficiency ranges support the older endpoint and ENERGY STAR's furnace criteria publish 87% for qualifying oil furnaces. Electric resistance is compared at 100% point-of-use efficiency. These sources were retrieved September 10, 2026.

Window-AC comparisons use the same seasonal-output approach with SEER2 in Btu/Wh: electricity price divided by SEER2 / 1,000 yields delivered cooling $/MMBtu. Because an existing window unit's rating is not collected as an input, its 8-12 Btu/Wh range is an explicit SplitAtlas editorial assumption, not a published product fact.

Fuel prices and conversions

  • Electricity: EIA Electric Power Monthly table 5.6.A, May 2026 residential state and U.S. averages, published July 23, 2026 and retrieved September 10, 2026. This is the same period and the same values used by existing SplitAtlas cost-state pages.
  • Natural gas: EIA annual residential prices by state, 2025 dollars per thousand cubic feet, released August 31, 2026 and retrieved September 10, 2026. Connecticut, Maine, Maryland, North Carolina, Oklahoma, Rhode Island, and South Carolina are visibly dated 2024 because EIA marks their 2025 state values unavailable.
  • Propane and No. 2 heating oil: EIA Heating Oil and Propane Update, residential dollars per gallon excluding taxes for the week of March 30, 2026, retrieved September 10, 2026. EIA only surveys participating states. If the selected state has no participating-state figure for the existing fuel, the calculator requires the reader's price from their last delivery and does not calculate savings or payback from the U.S. average fallback.
  • Heat content: EIA energy units and calculators, retrieved September 10, 2026: 3,412 Btu/kWh; 100,000 Btu/therm; 1,036 Btu/cubic foot of natural gas, equal to 10.36 therms/Mcf; 91,452 Btu/gallon propane; and 138,500 Btu/gallon heating oil.

State is optional. Leaving it unset uses the dated U.S. averages and creates a visible warning. Selecting a state without an EIA propane or heating-oil figure requires the reader's delivery price when that is the existing fuel; state electricity remains the mini-split comparison price. The reader can override electricity, gas, propane, and oil prices with bill figures. An entered annual existing-system cost is the strongest baseline and replaces the climate-zone default; without it, annual dollar savings use the displayed climate-zone cost range. Those climate-zone annual-cost ranges remain explicit SplitAtlas editorial planning assumptions.

In climate zones 6-8, a selected catalog system uses its Region 5 HSPF2 first, or its 5°F COP when Region 5 HSPF2 is unavailable, as the conservative cold-season bound. When neither quantified field is available, the lower COP bound is 70% of the HSPF2-derived seasonal COP. That 70% figure is an explicit SplitAtlas editorial assumption shown on the result; it is not presented as measured model performance.

Negative savings are preserved. If expensive electricity and cheap gas make the mini split cost more per delivered MMBtu, the result says operating cost is higher and does not claim a payback. Incentives and rebates remain excluded.

What changed from payback v1

Payback v1 assigned each existing-system type an editorial savings-fraction range, then nudged that range for SEER2/HSPF2. V2 removes those guessed savings fractions. It resolves dated state or national prices, applies published heat content and efficiency ranges, displays both delivered $/MMBtu figures, and derives the savings fraction from their ratio. V1 results therefore should not be compared as though they used local fuel prices; they did not.

Category bands on model pages

Model pages show scored categories as comparison bands, not as bare 0-100 category scores. A band is derived from the stored category percentile score. Because efficiency, cold-weather behavior, noise, and price norms vary by size, those categories normally compare a model against systems in the same standard capacity bucket: <=9k, 12k, 18k, 24k, 36k, or >36k. When a category/capacity bucket has fewer than 30 scored peers, the category falls back to the scored published catalogue and the page labels that fallback.

The band labels are bottom quartile, below average, above average, top quartile, and top 10%. The model page also shows the underlying specification figure used for the category, such as SEER2 for cooling efficiency or HSPF2 for heating efficiency.

Category weights

Cooling efficiency 15%. Heating efficiency 15%. Cold-weather performance 15%. Noise 10%. Warranty 10%. Installation accessibility 10%. Feature set 10%. Price-to-performance 10%. Documentation and parts availability 5%.

Normalization rules per category

Each category first calculates a raw value from sourced product facts, then normalizes that raw value to a percentile rank from 0-100. A score near 50 means the system is around the middle of the chosen peer group; a score near 90 means it is around the 90th percentile. Higher-is-better categories move upward as performance improves; lower-is-better categories, such as noise and price-to-performance, are inverted before percentile ranking.

  • Cooling efficiency: capacity-bucket percentile rank; SEER2 primary, EER2 secondary (70/30 blend).
  • Heating efficiency: capacity-bucket percentile rank; HSPF2 primary; COP at 47F secondary when published.
  • Cold-weather performance: capacity-bucket percentile rank; composite of capacity retention (rated heating at 5F divided by rated at 47F), minimum operating temperature, and published low-ambient rating existence. Missing all three means the category is not scored, never imputed.
  • Noise: capacity-bucket percentile rank; indoor minimum dB(A) (60%), outdoor dB(A) (40%); lower is better.
  • Warranty: catalogue-wide percentile rank; compressor years plus parts years, registered terms; dealer-only warranty restrictions are noted and penalized for DIY-class units.
  • Installation accessibility: catalogue-wide percentile rank; DIY status, precharged line set, included line-set length, 120V availability, and weight class. Professional-install systems are marked not applicable because DIY accessibility is not a valid comparison for them.
  • Feature set: catalogue-wide percentile rank; count of verified supported features, weighted (inverter, Wi-Fi, low-ambient kit, dehumidify, etc.).
  • Price-to-performance: capacity-bucket percentile rank; median current price record divided by capacity and efficiency blend, inverted. Stale prices (older than 60 days) are excluded. Professionally installed quote-only systems with no retail price are marked not applicable instead of being penalized.
  • Documentation and parts: catalogue-wide percentile rank; manuals on file, spec sheet on file, error-code coverage in the model database, parts-channel availability flag.

Applicability

Categories that do not apply to a unit type are excluded from the overall roll-up and disclosed on the model page. Their weight is redistributed across the scored categories. Missing required data is different: it is marked insufficient data and can keep a system from showing an overall score.

Quote-only professional systems are the main price example. If there is no retail equipment price to evaluate, price-to-performance is excluded and the model page says that installed quotes are the right comparison point.

Model page review snapshot

The buyer review snapshot is generated from the same normalized model record used by the score. It is not a separate hands-on review. Fit notes use the model's capacity, zone count, install path, voltage, cold-weather rating, and efficiency rating. Install difficulty weighs the refrigerant connection, DIY evidence, voltage, included line-set information, multi-zone layout, and electrical requirements.

Alternatives are same-zone models from other brands with similar capacity, a higher score, and a current exact retail destination. Availability notes use the most recent verified retailer price when one is current; older prices are labelled reference-only with the date last seen. Online listings are shown as context when a brand publishes a warranty or sales-channel rule that affects buying online.

Source dates come from the imported model record, page review metadata, and retrieval dates stored with source-backed facts. Warranty tiers show standard coverage and the best available registered or qualified coverage when those terms are published.

Confidence gating

  • Each category has required fields. A category with missing required fields is shown as insufficient data and excluded from the overall roll-up.
  • Overall score displays only if at least 5 applicable categories are scored and overall field completeness is at least 80%. A system must never look mediocre because data is missing; it looks unrated.
  • Score history is retained internally so methodology changes can be audited.

Governance

  • Weight or normalization changes require a new methodology version, a changelog entry on the public methodology page, and owner sign-off.
  • Sponsored status is invisible to scoring by construction; sponsorship fields are not included in scoring inputs.
  • Editorial "best for X" picks may consider scores but are labeled editorial judgment and never presented as computed output.