How the grading works

This site measures nothing. Four independent sources publish figures about Flare providers; we read them, join them on each provider’s on-chain identity address, and apply the weighting below. Every figure on a dossier links back to the site that produced it, and where two sources disagree we show both rather than picking a winner.

The weighting is the only thing we author. That is the part to argue with, and the reason it is published here in full rather than described in general terms.

The uptime and availability curve runs through fixed points: 90% → 0, 95% → 40, 99% → 80, 99.9% → 95, 100% → 100.

The composite is the weighted mean of the axes that are known, and the letter is cut at A ≥ 85, B ≥ 70, C ≥ 55, D ≥ 40, F below that.

1. Absence is never a fault. A provider that published no self-declaration, runs no validators, or is too new for a Cerberus reading scores nothing on that input — the axis drops out of the denominator instead of scoring zero. Missing data must never read as a bad result.
2. We publish measurements unadjusted. No caveat, correction or normalisation is applied to anyone’s number, including our own where one would be convenient. If a figure looks wrong, the argument belongs with the source that published it — and the dossier links you straight there.
3. No grade when too little is known. A grade needs at least 70% of the total weight and an accuracy reading. A provider registered but not submitting this epoch gets a dossier and no letter: grading it against the working field would claim more than the data supports.

Cerberus fingerprints every registered provider and publishes the probability that it is running Flare’s stock example implementation rather than an independent production stack. It is a compliance signal about how a provider is built. It is not a reading on whether its prices are any good.

Two things follow, and both are printed on every dossier. The detector scores agreement with the example — and the example itself tracks the consensus price, so a provider that is accurate for entirely independent reasons will agree with it often. And the published number moves with how much evidence the detector has: a provider whose evidence channels have not all filled in is scored on the ready ones, reweighted, which can read materially differently once the rest populate. That is why the axis carries the smallest weight of the five, and why the dossier shows the per-channel evidence rather than the headline alone.

Providers may sign a registry listing declaring how they are built. Where a signed declaration exists we print it beside the independent measurement, both linked. Most of the field has published no declaration at all — listing is optional, and “no claim” is not an accusation and is never rendered as one. Where a signed claim and a measurement point different ways, we print both facts and no adjective: we assert neither and have not adjusted either.

Flare Systems Explorer Flare Foundation

Band hit rates, availability, reward rates and the FIP.10 minimal-conditions record. The explorer has no per-provider page; this is the API query the figures were read from.

Flare Registry BurstLabs

Validator uptime and stake, signature-verified listings, self-declarations and eligibility standing.

Cerberus On-Chain cerberusonchain.xyz

The independent measurement of how closely a provider tracks the stock example implementation. Cerberus per-provider pages require an account. This is their open census endpoint — the same record this figure was read from, for every provider at once.

FlareMetrics flaremetrics.io

The variation in a provider’s reward rate, computed over FlareMetrics’ own history — available immediately, where a figure derived from this site’s snapshots needs many epochs before it means anything.

These are independent projects run by other people. We read their public endpoints once per refresh and identify ourselves in the request. None of them has endorsed this site or these grades, and any error in the joining or the weighting is ours, not theirs.

Seraphim Lab operates Seraphim FTSO, one of the providers graded here. We are publishing a scorecard that includes our competitors and ourselves. That is a real conflict and pretending otherwise would be worse than declaring it, so:

• The rubric and its weights were written and committed before any provider — including ours — was scored under them.
• Our own dossier is rendered by the same template as every other, with no additional context and no section other providers do not get.
• The provider register does not rank by grade and our row is never pinned; it sorts wherever the numbers put it.
• Where our own figures are unflattering they are printed as they stand. The independence axis is the live example: our published reading is mid-band, and it is on our page at full size.

If a measurement looks wrong, raise it with the source that published it — they own it and can fix it at the root; the dossier links to each one directly. If we have joined, weighted or described something wrongly, that is ours: write to hello@seraphimlab.ca and we will correct it and say what changed. Providers may also sign a listing at the Flare Registry at any time, which is the fastest way to put your own account of how you are built next to the measurement.

• Flare only. The registry also covers Songbird; those providers are not graded here yet.
• Ranked axes are ranked within the providers active in the current reward epoch, so a score can move when the field changes and not because the provider did.
• Grades are computed and stored per reward epoch under the rubric version in force at the time. Changing the rules bumps the version rather than silently re-grading the past.

Convinced by the data? Delegating takes two minutes. Delegate to Seraphim →