Methodology
How we score mattresses
Our physical-testing rubric, version 2.1, has 9 weighted categories totalling 100%. A numeric measured score requires that protocol's recorded measurements. The rubric below is a future protocol, separate from current research verdicts.
Three kinds of evidence — kept separate
| Evidence type | What it is | What we publish |
|---|---|---|
| Physical measurements | Future recorded, dated readings; none currently on record. | Only future supported results could include category and overall scores with raw measurements. |
| Verified specifications | Manufacturer-reported height, construction, firmness, trial, warranty and prices where a source is available; a verification date is shown only when recorded. | The facts themselves, with the source and date where we have them. |
| Editorial research judgement | Our reading of specifications, materials and policies for a given sleeper. | Written verdicts, fit notes and who-should-skip guidance — not a number. |
No mattress has completed the current instrumented physical protocol with recorded measurements. Current mattress records are research evaluations, not lab-tested results. Research evaluations compare published manufacturer specifications, construction, materials and buying terms. They are not personally tested results. Third-party claims require attribution; historical experience is not evidence of completion of the current physical protocol. A historical account does not establish current measured performance. Existing list positions are not proof of measured superiority.
The future physical-testing rubric: proposed weights
| Category | Weight | Proposed input |
|---|---|---|
| Cooling | 18% | °F surface temp delta after 60 min |
| Motion Transfer | 14% | Peak g-force at 12" from a 10 lb dropped weight |
| Pressure Relief | 14% | Peak shoulder pressure (mmHg) for a 165 lb side sleeper |
| Edge Support | 12% | Vertical sinkage (inches) under 230 lb seated load at perimeter |
| Response Time | 10% | Seconds for surface to fully rebound from a 25 lb release |
| Sex / Combo Sleep | 8% | Composite of bounce, friction, edge, and quiet |
| Off-Gassing | 8% | Days until trained panel detects no VOC odor |
| Durability Simulation | 8% | Pass/fail on 30,000-cycle rolling load test |
| Company / Trust | 8% | Composite of trial period, warranty, certifications, and CS responsiveness |
Proposed score bands — not current results
- 9.0 – 10
- A proposed rubric interval, not a measured product result, dataset percentile or clinical recommendation. Comparative claims require recorded evidence before use.
- 8.0 – 8.9
- A proposed rubric interval, not a measured product result, dataset percentile or clinical recommendation. Comparative claims require recorded evidence before use.
- 7.0 – 7.9
- A proposed rubric interval, not a measured product result, dataset percentile or clinical recommendation. Comparative claims require recorded evidence before use.
- 6.0 – 6.9
- A proposed rubric interval, not a measured product result, dataset percentile or clinical recommendation. Comparative claims require recorded evidence before use.
- Below 6.0
- A proposed rubric interval, not a measured product result, dataset percentile or clinical recommendation. Comparative claims require recorded evidence before use.
Rules we hold ourselves to
- Evidence policy requires a completed, recorded physical protocol before presenting a score as measured performance. Research verdicts should use words or "numeric rating unavailable", not an estimated lab score; legacy editorial numbers are not physical measurements.
- Manufacturer firmness numbers are labelled as the manufacturer's, never as our measurement.
- Category scores are position-agnostic; where a mattress suits one sleeper and not another, that goes in the verdict, not into a fudged number.
- Editorial policy requires conclusions to follow evidence rather than affiliate commissions. No undocumented history of brand requests or separate editorial teams is asserted.
- Documented construction changes should prompt review of affected research. No fixed retest cadence is established; dates should reflect real content changes, not automatic refreshes.