The formula, in the open.
Every number on BatteryRank is derived from sourced data and documented assumptions. This page shows exactly what goes into a score, where the data comes from, and what we deliberately don’t model — because honest limits build more trust than false precision.
Scoring catalog v1.3 · updated Jul 6, 2026
Three buyer questions, three engines.
Which one should I buy?
The BatteryRank Score and use-case rankings — a composite of seven sourced sub-scores, re-weighted per use case (camping, CPAP, home backup, van life…).
What size do I need?
Device-first runtime math. Pick your CPAP, fridge, or phones and we return honest hours — real conversion losses already subtracted, not the label.
Is this price a real deal?
The fair-price engine judges today's price against the model's own 12-month history, then calls a verdict — never against an invented MSRP.
Seven sub-scores behind one number.
The headline 0–100 score blends seven single-facet sub-scores, each a normalized function of sourced specs. They’re combined by a versioned weighting we tune deliberately — no single factor is allowed to dominate.
- Capacity & expandabilityusable watt-hours and how far the pack can grow
- Output power & versatilitycontinuous and surge watts, plus the port mix
- Trust & reliabilityratings weighted by review volume, and documented recalls
- Charging speed & solarhow fast it refills from the wall and the sun
- Durability & longevitycell chemistry, rated cycles, and warranty
- Portabilityreal weight and bulk, judged against its size class
- Quiet / off-grid livabilitynoise under load and off-grid niceties
The score renormalizes over whatever data a model has — so a model is never penalized for a spec nobody in its class publishes.
The score is price-independent. There is no “value” axis in the composite. Price moves daily; a station’s quality doesn’t. All price intelligence lives in the fair-price engine below — so the score you read today means the same thing next week.
Four pillars on every model page.
The seven sub-scores collapse into four readable pillars for the radar on each model page — the same math, grouped the way a buyer thinks.
Power
Capacity + output power
Trust
Reliability + durability
Charging
Charge speed + solar
Livability
Portability + quiet running
A price judged against its own history — not a fake MSRP.
We expand a model’s tracked prices to daily coverage over the trailing 12 months, then ask where today’s effective price (after coupons) falls in that model’s own distribution. The percentile maps to one of four verdicts.
Discount depth
How far below MSRP, judged against the model’s own typical discount — so “40% off” isn’t a deal if it’s always 40% off.
Value at this price
Even at today’s price, is its cost-per-usable-watt-hour good versus peers in its size class right now?
Buy-now vs wait
A disclosed rule combining the verdict with known upcoming events (e.g. Prime Day lows). Rule-based, not a black box.
With fewer than ~30 days of tracked history, the verdict shows “not enough price history yet” — never a fabricated percentile. Out-of-stock MSRP snap-backs are excluded so a price nobody could pay never makes the next real price look like a deal.
The conventions behind every number.
Peer groups, not one big pile
Relative scores are computed inside six size cohorts (from ≤350 Wh carry units to >4,200 Wh whole-home). A 128 Wh power bank is never ranked against a 12 kWh home unit.
Two honest normalizations
Class-relative facts use a percentile within the cohort; facts with an absolute “good” (noise, warranty years, star rating) use fixed, documented thresholds.
Missing data renormalizes — never zero
A blank spec is dropped from its sub-score’s weight pool, not scored as zero. Lose more than half the inputs and the sub-score shows N/A rather than a misleading number.
Claimed vs. verified, always distinct
Manufacturer figures wear a dashed outline; independently confirmed ones earn a solid pine shield. Estimates are tagged “est.” and shown as ranges, never point figures.
Where every number comes from.
No press releases, no sponsored data, no brand partnerships. Each figure traces to one of these, carries its source, and is dated.
OEM spec sheets & manuals
claimedClaimed specs — capacity, rated output, cycle life, ports, warranty — read from the manufacturer's own datasheet or manual, one field per concept across brands.
FCC / UL / DOE & lab results
verifiedWhere an independent result confirms a claim — certifications, teardown or lab measurements — the figure is promoted to verified and wears a pine shield.
12-month price history
pricesRetailer price observations backfilled from Keepa, then extended daily by our own scraper. Each price is timestamped to its source date; the table is append-only.
Ratings & documented reputation
trustAmazon star ratings weighted by review-count confidence, plus sourced, countable facts — documented recalls, complaint volume. Never coded from vibes or unsourced sentiment.
What we deliberately don’t do.
- No editorial listicles — every ranking is computed from the published formula, and nobody pays to sit at the top.
- No invented hands-on impressions or review sentiment. We own no hardware and never pretend to.
- No fabricated numbers. If a spec can't be traced to an authoritative source, it stays claimed, marked N/A, or is hidden — never guessed.
- No number without a source and a date. Prices always carry an “as of.”
Scoring catalog v1.3 · last updated Jul 6, 2026. Think a number is wrong? We want to know — with a source.
Browse the rankings