Methodology

Where the numbers come from, how the rubric works, and what we won't do.

Where specs come from

Every numeric field on this site — pricing, legroom, headroom, cargo volume, towing capacity, seating — comes from a manufacturer specification page, a recognized automotive-data source (e.g. Car and Driver, Edmunds, Kelley Blue Book), or a press kit issued by the automaker. Each model carries a sources array recording the label, URL, and exact field names that source verifies. You can read the source list at the bottom of every model page.

Verified vs unverified

Specs flow into the site through an editorial spreadsheet. A field is considered verified only when (a) a value is recorded against the model and (b) at least one source row marks that exact field as verified. Until both are true the value renders as Needs verification (or a dash in tables) with a visible badge. We do not publish numbers we have not sourced — verification gaps are shown, not hidden.

Persona scores are derived, not editorial

Each model carries six persona scores on a 0–10 scale: tall adults, car seats, towing, luxury feel, value, and EV interest. These are not opinions. Each score is a deterministic function of the model's verified specs, anchored to the segment's own minimum and maximum:

Tall adults — Row 3 legroom percentile within the segment, with a +1/−1 adjustment when Row 3 headroom is at or above 38 inches or below 36 inches.
Car seats — Row 2 legroom percentile, plus a +1 adjustment when seating capacity reaches eight.
Towing — Maximum tow rating percentile within the segment.
Luxury feel — Starting MSRP percentile, with a −1 adjustment for body-on-frame chassis and a floor of 5 (every model in this segment is a luxury vehicle by definition).
Value — Sixty percent inverse-MSRP percentile (lower price scores higher), forty percent utility percentile (cargo volume and seating).
EV interest — Powertrain type: 10 for fully electric, 6 for plug-in hybrid, 3 for hybrid, 1 for gas.

When the segment's specs change, the rubric automatically re-anchors. The audit trail for every score is committed alongside the generated data file as a per-dimension comment: which inputs were used, which percentile they produced, and what adjustment (if any) applied.

Recommendations are deterministic

The recommendation quiz maps each of your six answers to a fixed weight across the six persona dimensions, then computes a weighted dot product against each model's persona scores and returns the top three. There is no language-model call anywhere in the recommendation path. Two users who answer the quiz identically will see identical results. The reasons shown alongside each match are templated from the answers you gave — no free-text generation.

A model with any unfilled persona dimension is excluded from the quiz entirely; the page renders a calibration state if fewer than three models survive. Partial data never produces partial rankings.

Editorial copy is structural, not comparative

The prose on each model page — one-line positioning, best-for buckets, watch-outs — is written by hand and frames the trade-offs the specs imply. We do not write “the most refined” or “best-in-class” or comparative claims about specific named models without a sourced numeric basis. The recommendation in the editorial copy is the framing; the recommendation about which model to buy is the quiz output for your answers.

Third-row fit estimator

The third-row simulator on the homepage estimates how a chosen adult height fits a chosen model's third row. Because manufacturer-reported legroom and headroom follow SAE J1100 — they're measured against a standardized seated manikin, not as raw cabin distances — the estimator anchors its math to the same reference point and scales up or down by stature.

The reference manikin is approximately a 5′9″ (69-inch) adult male, the 50th-percentile reference used by J1100. We assume this reference manikin needs about 31″ of third-row legroom and 36″ of headroom to sit comfortably — values calibrated against the editorial thresholds that the buyer guides use.

Knee clearance = verified row3LegroomIn − (31 + 0.45 × (your height − 69)). Each additional inch of stature requires roughly 0.45″ more legroom (lower-leg plus thigh proportional growth).
Head clearance = verified row3HeadroomIn − (36 + 0.52 × (your height − 69)). The 0.52 ratio is the standard Dreyfuss sitting-height proportion (top of head to seat surface, seated upright).

The two stature-scaling ratios come from Henry Dreyfuss Associates, The Measure of Man and Woman: Human Factors in Design (revised, 2002), and from SAE J1100 itself. They produce reasonable estimates for typical proportions but do not account for individual variation in leg-to-torso ratio, posture, seat recline, cushion firmness, or seat-track position.

The qualitative buckets (“tight”, “adequate”, “comfortable”) are editorial thresholds tuned against the same numeric anchors the buyer guides use. They are not a substitute for sitting in the vehicle. The simulator labels every result with both the verified spec and the required value for the input height, so a reader can audit the math directly.

Slouch tolerance. Head clearance allows up to 1″ of negative geometric clearance before the verdict flips to “doesn't fit.” The 1″ buffer models the slack the strict upright-posture math doesn't credit: natural slouching, seat-cushion compression under weight, slight head tilt. Within that band, an adult can typically adjust posture to fit even if uncomfortably; beyond it (≥ 1″ of overlap), the seat geometry genuinely doesn't accommodate that stature. Knee clearance has no equivalent tolerance — femur length doesn't compress the way posture does.

What we won't do

Publish a number we have not sourced and verified.
Generate persona scores from intuition rather than the spec-anchored rubric above.
Use a language model in the recommendation path.
Write comparative editorial claims between named models without a sourced basis.
Hide verification gaps; null data renders as null.

Corrections

If you spot a spec that disagrees with a manufacturer page, or a source link that has rotted, the data file lives in version control and the editorial workflow regenerates the runtime overlay from a single spreadsheet. Corrections are deliberately easy to make and audit.