How we match products

The core rule: verified, or not shown

SnowSift never publishes a spec it can't back. When our pipeline can't confirm a product, the product is held back — it simply doesn't appear on the site — and it is re-checked automatically on every run. When it can confirm the product but not one of its values, that value is held back and shown as a dash. A dash fills in only when the manufacturer's page is fetched again, two fresh reads agree on the value and the page's own spec table backs it; a plain re-read of the same page never does it. We'd rather show you fewer products, or a dash, than a wrong number.

Verification means independent channels agreeing

A single automated reading of a page — however sophisticated — can be wrong in ways it can't see itself. Two similar systems can even be wrong together. So a product is verified on SnowSift through agreement between independent channels with different failure modes:

1. Source capture. We fetch the manufacturer's own product page — the primary source for what a ski or board actually is — and store a snapshot of it. We never read specs from retailer pages.
2. Independent double extraction. Two separate automated systems each read the same snapshot and independently produce the full spec sheet (geometry, flex, profile, weights, MSRP). Neither sees the other's answer. The two sheets are compared by plain code, field by field, with fixed tolerances (a ski's waist width must match exactly; a weight may differ by a few percent).
3. The manufacturer's structured data. Most manufacturer pages carry machine-readable product data (schema.org markup — the same data Google reads). We parse it with plain code — no AI involved — and check the product's brand, model and price against it. If it names a different product, nothing is published.
4. The manufacturer's own spec table, read by code. Where the page prints a spec table or spec list, plain code reads the values straight off it, with no AI involved. It reads only clearly labelled values: a filter band such as “80–89mm”, an ambiguous size list or a value it can't tie to the shortest length gives no reading at all.
5. Verdict. A product is published only when at least one of the code-only channels backs it: the structured data confirms which product the page is, or the spec table confirms every key spec (a ski's waist and lengths, a boot's flex and last, and so on). Pages with no structured data are held to a stricter agreement bar between the two readings. The channels behind each product are listed under its specs.

More independent channels are coming: when live retailer data feeds arrive, their product attributes become another cross-check, and products verified today are automatically upgraded against them.

When the two readings disagree

Disagreements are settled by code, or not at all. Three rules decide, and when none applies the product or the value stays held back and is re-checked on the next run:

• Which product is it? If the two readings name the model differently, we accept the one the page's own structured product name carries — and only when that name and the page title agree, the page names one product, and both readings are for the same skier (men's, women's or junior). Otherwise nothing is published.
• Which value is right? If the two readings give different values for a spec, the value is shown only when the spec table, read by code, matches exactly one of them. If it matches neither, or the page has no table, the value is held back.
• MSRP shows only when both readings state the same price, or the page's structured data carries it in US dollars.

A season is part of what we publish, so it has to be printed on the manufacturer's page (or follow a fixed rule for how that manufacturer labels its seasons). We never infer one.

What a dash means

A dash (“—”) means we have no value we can stand behind. Usually the manufacturer doesn't publish that spec. It can also mean the two readings disagree on it and the manufacturer's spec table doesn't settle it, or a later check of the manufacturer's page contradicted the value we had. That later check can only remove a value; it never puts one back. A product can be published with some dashes once its key specs are confirmed and at least three specs agree. A dash is replaced only by a value confirmed the same way: the manufacturer's page has changed, both readings of the new page agree, and the spec table (or, for MSRP, the page's structured price) backs them. A product taken down by that later check comes back only when every key spec clears that bar. We don't estimate, interpolate, or copy numbers from third-party sites.

Size-varying specs are quoted at one length

On a ski or a snowboard, several specs are not one number at all — sidecut radius, weight, effective edge, tip and tail width, and on many models the waist too. A manufacturer publishing a spec table for six lengths is usually publishing six different turn radii. Which specs move, and by how much, varies by model: some skis hold one waist down the whole size run, others widen it by several millimetres.

That breaks the comparison at the heart of our verification: two independent readings of the same page can each be perfectly correct and still disagree, simply because one read the 166 cm row and the other read the 184 cm row. So both readers are told to use the same anchor — the shortest length the manufacturer lists — and that is the length we quote.

Being precise about how strong that is: the anchor is an instruction to both readers, not something we can verify deterministically after the fact, and the tolerances we compare them with are deliberately wide enough that two different-size readings still count as agreement. The rule also post-dates our first catalogue crawl, so the oldest rows were read before it existed and have not all been re-read since. It is how we ask for the number, not a guarantee we can prove about every row — and we would rather tell you that than imply a precision we don't have.

So we mark the affected rows in the spec panel and, where the manufacturer publishes a length list, name the anchor length in a footnote beneath them. Two consequences are worth stating plainly. If you ride a longer size, your turn radius will be longer and the ski will weigh more than the headline figure. And because each product is anchored to its own shortest length, two products are not directly comparable on those rows — a women's ski anchored at 146 cm will show a much shorter radius than the same ski anchored at 164 cm, and that gap is the anchor, not the design. Check each product's footnote before reading anything into the difference.

Why we don't quote an accuracy number

You'll notice this page claims no percentage. That's deliberate. An honest accuracy figure requires an independent measuring stick, and we won't manufacture one to have a marketing number. What we can honestly promise is structural: every published spec survived agreement between independent checks, anything that didn't is hidden, and every product shows which checks verified it. No number, no vibes — just a process you can inspect.

How this differs from a gear test

A gear test tells you how a ski feels: how it carves, floats or chatters, as judged by the people who rode it. That is real knowledge, and SnowSift does not try to replace it. We answer a narrower question: do independent readings of the manufacturer's own page agree on its specs? Waist, radius, weight and flex are numbers a manufacturer publishes, and we only show them once independent checks agree on them within tolerance.

We don't rate products, rank them or tell you which to buy, and a spec we can't confirm stays hidden. Read a gear test for how something rides, and use SnowSift for the manufacturer's published numbers once they have been cross-checked.

What happens when price comparison launches

For the 2026/27 season, SnowSift will add live cross-retailer price comparison from official retailer data feeds. Every price row will be matched to its product using the same philosophy: multi-signal agreement (brand + model + season + key dimensions), a strict confidence threshold, and asymmetric disclosure — a listing that can't be confidently matched is hidden, never shown with a “maybe”. Visible rows will carry a “✓ Verified match” badge; you'll never see a raw confidence percentage, because a number like that isn't information you can act on — hiding the row is.

Found an error?

Tell us: data@snowsift.com. Spec reports are re-checked against the manufacturer source and corrected. See also About SnowSift.