Was the band right?
Every quote published while the market was shut is a claim about a price nobody could observe yet. When the tape reopens, that claim is either right or it isn’t. This page scores all of them.
The record is the chain. The contract keeps only the latest quote in storage, but every quote it ever accepted survives as a QuotePosted log signed by an allow-listed key. So these numbers are not ours to adjust after the fact, and you can recompute them yourself from Robinhood Chain without asking us for anything.
Coverage at reopen
nominal 95% · one score per closureBy instrument
live vs backtested| Instrument | Live coverage | Scored | 95% CI | Backtest | Pending |
|---|---|---|---|---|---|
| HOOD Robinhood Markets Inc. | — | — | — | 93.8% | 30 |
| COIN Coinbase Global Inc. | — | — | — | 93.4% | 34 |
| NVDA NVIDIA Corporation | — | — | — | 95.0% | 14 |
| TSLA Tesla Inc. | — | — | — | 95.2% | 14 |
| AAPL Apple Inc. | — | — | — | 96.4% | 9 |
| MSTR MicroStrategy Inc. | — | — | — | 94.2% | 52 |
| SPY S&P 500 ETF Trust | — | — | — | 95.8% | 9 |
| TLT 20+ Year Treasury Bond ETF | — | — | — | 95.6% | 9 |
How this is scored
One score per closure. A closure is a run of quotes published while the tape was shut. It is scored using the last band published before the reopening print. That is deliberate: the fitted sigma is anchored on a complete close-to-open gap, so the band is only claiming to cover the whole gap once the whole gap has elapsed. Scoring the last quote is the only apples-to-apples comparison with the backtest.
Every band during the closure, too. Scored against the same print, all 0 of them come to —. Expect this to read lower, and it is not a bug: the band widens with elapsed staleness, so a quote published an hour into a 62-hour weekend carries an overnight-sized band against a weekend-sized move. It is reported because a consumer reading the feed on Saturday morning is handed that quote, not the Monday one.
Live prints are not scored against each other. A TRADED band is source dispersion plus execution noise around a price that already exists. It is not a forecast, so asking whether the next print landed inside it would be testing a claim the oracle never made.
The arithmetic matches the contract. Band edges are computed as (price * confidenceBps) / 10000 over integers, exactly as getBandedPrice does on-chain, so the interval scored here is the interval a consumer is actually handed. Doing this in floating point is not merely imprecise — it moves answers, because a price landing exactly on the edge falls the wrong way.
Intervals are Wilson score, 95%. With a handful of samples a bare percentage invites the reader to treat “3 of 3” and “3,790 of 3,992” as the same statement. They are not, and the interval says so.
Contract 0x65cf45524407a5e700188a8a8178d5d5c0c38d30 on Robinhood Chain (4663). Signer 0xA139E54E6c88420cb5546c64130d1Ba145Ff9C8A. Raw data at /api/coverage, method in src/lib/ledger.ts.