No-vig odds are a set of prices with the margin taken out, so their implied probabilities sum to exactly 100% rather than more.
They are also called fair odds, and the two terms mean the same thing. A market quoted at −110 on both sides implies 52.38% apiece, or 104.76% together; the no-vig version of that market is 50% and 50%, priced at +100 each. Nothing about the market's opinion changed — only the margin that was sitting on top of it was removed.
The number matters because a raw price is not a probability estimate you can use directly. It is a probability estimate with a business model attached. Two markets priced by different sources carry different margins, so their raw implied probabilities are not comparable until each has been de-vigged. Comparing a raw implied probability against your own model has the same problem: you are testing your estimate against an inflated one, and the more lopsided the market, the more inflated it is.
No-vig odds are not a prediction and not an edge. They are the market's own view, stated without the padding — which is exactly what makes them a reasonable benchmark to measure something else against. Whether that view is right is a separate question that no amount of arithmetic answers.
One caveat worth knowing: "the" no-vig price is slightly method-dependent. The standard approach divides each probability by the total, but three other methods distribute the removal differently and disagree by a couple of percentage points on lopsided markets. Anyone quoting a no-vig number without saying which method produced it is quoting the multiplicative one.
The CS2 market line we publish is no-vig by construction: each contributing source is de-vigged before the median is taken, so the served sides sum to exactly 100.0000%.