How we verify

How we collect data covers where the numbers come from. This page covers the harder half: what we do when a source is unclear, and what each label on this site actually means. Read it before you cite us — the labels are load-bearing.

The one-sentence version. We would rather publish a gap than a guess, so every figure here is either read directly from the manufacturer's own page, marked as derived, or left blank — and where we don't trust our own reading, we say so on the page in a banner rather than quietly dropping the row.

Right now, across 108 models

State	Models	What you see on the page
Sourced — figure read from the manufacturer's page	108	Source link and the date we read it, on every model page
Data doubt — we don't trust our own reading	13 (12%)	Warning banner at the top of the page, and excluded from every average
Class-switchable — verified multi-mode bike	62 (57%)	Neutral explanatory note. No banner — a fact about the bike, not a fault in our data
Derived — capacity computed, not published	33	Marked as derived, with the volts and amp-hours it came from
Blank — no capacity we could attribute	14	An em dash. Never a zero, never a figure borrowed from a sibling model

Flagged figures get a banner, not a deletion

When our consistency checks fail — watt-hours against volts × amp-hours, e-bike class against top speed, peak power against nominal — we don't pick the more plausible number and move on. The model stays on the site with a banner at the top of its page naming which figure is in doubt and why, and it is excluded from every average and comparison we compute.

13 of 108 models are in that state today, and the reason is printed on each one. That is deliberately a visible cost to us: a flag is an admission. If the number of them ever starts climbing quietly, we want it obvious on the pages themselves rather than buried in a log.

Being flagged does not mean the manufacturer did anything wrong. It most often means one product page is describing several variants at once.

Capacity is variant-attributed, or it is left blank

This is the rule that costs us the most rows, and the one we are least willing to bend. A single product page often lists two batteries — say 15 Ah and 20 Ah — because the same page sells two variants of the bike. Taking the larger inflates the spec. Taking the smaller understates it. Taking either invents a fact the page does not state.

So: a capacity is recorded only if something on the page ties that capacity to this specific variant. A model name containing the amp-hours counts. A spec table row scoped to the variant counts. Two numbers floating on a shared page do not. When nothing ties them, the field stays empty and the page shows a dash.

Worked example: Himiway Himiway A7 Commuter lists two capacities on one page with nothing attributing either to this model, so its capacity is blank. We could have written a number there and nobody would have checked.

14 of 108 models currently have no capacity for this reason. A site showing a number for all 108 is not better sourced than us.

Derived values are marked as derived

Watt-hours are defined as volts × amp-hours. When a manufacturer publishes 48 V and 15 Ah but never publishes 720 Wh, we compute it — arithmetic on their own figures, not an estimate. It is still not a number they printed, so we say so.

33 of 108 capacities on this site are derived. Each shows the volts and amp-hours it came from, so you can redo the multiplication yourself. See Himiway Himiway D7 Pro Mountain.

We derive watt-hours from volts and amp-hours. We derive nothing else — not range from capacity, not capacity from range, not top speed from class. Those relationships are real but loose, and a figure computed through a loose relationship is a model output, not a spec.

Certification records what the manufacturer states — and "not stated" is not "no"

We track two separate standards: UL 2271 (the battery pack) and UL 2849 (the whole electrical system). They are not interchangeable, and we never infer one from the other — even though in practice a bike certified to 2849 almost always has a 2271 pack.

Each model gets one of two values, and the difference between them matters:

Certified — the manufacturer's own page states this certification, and we store the sentence that says so.
Not stated — the manufacturer's page does not mention it.

"Not stated" is a fact about the page, not about the bike. A bike whose page says nothing about UL 2271 may well be certified: the manufacturer may not advertise it, or may state it in a PDF we don't parse. We have no way to tell from the page, so we report exactly what we can see and nothing more. Do not read our "not stated" as "uncertified", and please don't cite it that way.

Today 67 of 108 models state UL 2271 and 72 state UL 2849; 41 say nothing about UL 2271 either way. The honest headline from that data is "41 manufacturers don't tell you" — which is itself the finding.

The full table is on the certification page.

Laws are read by a person, always

11 states are written up, and each page cites the statute and carries the date a human last read it. We run a monitor that watches those official sources for edits, but a detected change never edits a page automatically — it opens a worksheet for a person to read the statute and decide. A script that interprets a statute is a liability, not an automation.

We cover a deliberately small number of states well rather than 50 badly. Fifty stale law pages would be worse than none, because a rider would trust one.

What would make us wrong

Worth saying plainly, because a methodology page that claims no weaknesses is hiding them:

We read manufacturer pages. If a manufacturer publishes a wrong number, we will publish it too — accurately attributed, and still wrong.
We read HTML. Specs in a PDF, in an image, or behind a configurator are invisible to us, and surface here as "not stated" or as a blank.
A page is a snapshot. Every figure carries the date we read it; a bike revised since then is stale until the next run.
Claimed range is a marketing number produced under conditions nobody publishes. We report it as a claim, and it is the entire reason we are building an owner-reported alternative.

Citing us

You're welcome to. Two asks, both of which follow from everything above. Cite the figure with its label — "not stated", "derived", "flagged" — because the label is part of the fact. And link the model page rather than quoting a bare number, because the page carries the source and the date and a bare number doesn't.

If a figure here is wrong, we would rather fix it than defend it. Every page carries the manufacturer's own source link — send us the discrepancy and we'll re-read the page and note the correction.

One example of why the source link matters. Aventon Aventure M shows 250 W where its siblings show 750 W. That looks like a scraping error and isn't: it is a mid-drive rated 250 W sustained, and the manufacturer's page says so in those words. The figure that looks wrong is sometimes the one that's right, which is why we publish where it came from instead of asking you to trust us.