Methodology

Mars Score is a transparent, category-weighted composite score from 0 to 100. Every contributing spec, its weight, and its normalization range is published here. Nothing hidden, nothing weighted by hand.

Data sources

Amazon Product Advertising API — primary metadata, pricing, imagery, variant data. Refreshed every 6 hours for prices, every 24 hours for specs.
BestBuy API — US availability cross-reference, merged by UPC where possible.
Editorial review — top 200 products in each category receive manual editorial review. Specs the parser is not confident about are flagged for human verification.
User suggestions— moderated queue. Approved suggestions are credited in the product's edit history.

The Mars Score formula

For any product in category C:

For every spec group in C, we collect every required spec value the product has.
Each value is normalized to a 0–1 range using one of three methods declared in the category config — minmax (linear), minmax_inverse(for “lower is better” specs like weight), or log_minmax (for specs with wide dynamic range like camera megapixels).
Group score = arithmetic mean of normalized values × 100.
Raw Mars Score = weighted average of group scores, using the per-category weights below.
Display rescale. The raw 0–100 score is mapped to a 35–95 presentation range so flagship products land where buyers expect them (88–94) without inflating budget models. The mapping is a pure function:
```
display = 35 + (raw / 100) ^ 0.85 × 60
```
Anchor points: raw 0 → display 35, raw 50 → 67, raw 70 → 79, raw 90 → 89, raw 100 → 95. Relative ranking between any two products is preserved exactly — rescaling is monotonic. The transformation lives in rescaleScore in src/lib/score/mars.ts.

Confidence (the percentage of required specs that were actually present) is tracked separately and surfaced on the product page when below 80%. Products below 40% confidence have their score hidden entirely — we don't guess at numbers we can't back up with data.

Example: Smartphones weights

For smartphones, the Mars Score is a weighted average of these seven groups:

Group	Weight	Required specs
Design	8%	none
Display	18%	Screen size, Refresh rate
Performance	25%	RAM, AnTuTu, CPU total clock
Cameras	18%	Main camera
Battery	20%	Battery capacity
Connectivity	7%	none
Operating System	4%	none

Normalization ranges

Each required spec has a published min/max range. A smartphones battery of 3,000 mAh normalizes to 0; 8,000 mAh normalizes to 1; everything in between is linear.

Spec	Method	Min	Max
antutu_score	minmax	200,000	3,500,000
ram_gb	minmax	4	24
battery_capacity_mah	minmax	3,000	8,000
screen_size_in	minmax	4.7	7.5
refresh_rate_hz	minmax	60	240
main_cam_mp	log_minmax	8	250
weight_g	minmax_inverse	130	250
promised_os_years	minmax	1	7
cpu_total_ghz	minmax	5	30

How this differs from other comparison sites

Most product-comparison sites — including versus.com, RTINGS, GSMArena — compute a single composite score, but the formula behind it is either undocumented, locked inside thousands of lines of internal code, or moved between versions without retroactive recomputation. Buyers see “76 / 100” and have no way to verify why.

vsMars publishes the entire pipeline on this page:

The exact per-category weights (table above) — auditable, versioned.
The exact normalization ranges (table above) — every spec's 0-1 bounds visible.
The exact display rescale formula (above) — a one-line function.
An open-source audit script (scripts/audit-mars-scores.ts) that lets you reproduce every Mars Score on every product, locally, in 10 seconds.
A confidence threshold (≥40%) below which the score is hidden, not guessed.

If you find a Mars Score you disagree with, the disagreement is concrete and resolvable: either you have a different spec value (in which case tell us, with a citation), or you weight categories differently (in which case the methodology is fully exposed and you can compute your own ranking). No “trust the algorithm” black box.

What we don't do

No hand-tuning per product.Mars Score is computed from the formula above. We never bump a product's score because we like it or because a brand asked.
No paid placement in rankings. Affiliate revenue does not influence Mars Score, sort order, or which products appear in category top-10 lists.
No black-box AI scoring. Every input, weight, and step is public. The full implementation is in src/lib/score/mars.ts in our codebase.

Versioning

Mars Score weights are versioned per category. The current smartphones weights are version 1. When we change weights — typically once or twice a year as a category matures — we publish a changelog and recompute every score retroactively, so historical comparisons stay consistent.