vitna.ai/docs/concepts/reference-rate Machine view
Raw .md llms.txt
> Source: https://vitna.ai/docs/concepts/reference-rate
> Raw Markdown: https://vitna.ai/docs/concepts/reference-rate.md
> Corpus index: https://vitna.ai/llms.txt

# Reference rate

One number per product per day: the median list price among the providers
publishing one, by a rule with no judgement in it.

```bash
curl https://vitna.ai/api/v1/rates
```

## The rule

1. A series exists for a product that at least three distinct providers publish
   a confirmed price for on the day.
2. The rate is the median of those prices: the middle one, or the mean of the
   two middle ones, with the halfway case rounded up.
3. Arithmetic is exact, in whole millionths, the precision the record stores.
4. A price read once and not yet confirmed by a second read does not count.
5. A series is never restated. A day is published once, and a correction is a
   new row on a later day.

Rule 5 is the one that gives the series its value. A benchmark that can be
quietly recomputed after the fact is not a benchmark, it is an opinion with a
timestamp.

## What a rate is not

> [!WARNING]
> A reference rate is not a price anyone can buy at, not a volume-weighted
> average, and not a forecast. All three are reasonable assumptions about
> something called a reference rate, and none of them is true.

Published inside the API response itself, in an `isNot` array, because all
three are things a reader could reasonably assume and none is true:

- **not a price anyone can buy at.** It is a median of list prices. No seller is
  obliged to be at it.
- **not a volume-weighted average.** The record holds no volumes. Vitna sees
  published prices, not transactions.
- **not a forecast.** It is a statement about a day that has already happened.

## Why the method travels with the numbers

The rule is served in the same response as the rates rather than linked from
it. A benchmark whose method a reader has to go and find is one most readers
will use without reading, and the entire claim being made here is that the rule
is fixed and checkable.

## Recomputing

```bash
npm run witness:rates -- --verify
```

Recomputes every day ever published and reports disagreement. `--dry-run`
writes nothing.

Because inputs are append-only and the rule has no free parameters, an
independent implementation reading `/api/v1/skus/{slug}/history` should arrive
at the same series. That is the intended check.
Human Machine Shift+M