Using the SCI as a benchmark
The Scopes Desk
The Scopes Calibration Index measures how well a venue's prices anticipated outcomes. That makes it the natural bar for anyone who claims to price events better than the market: a forecaster, a model, a signal vendor or an AI agent. This guide covers how to reference it in a mandate, a performance contract or a service level so the reference means one thing and settles the same way every time.
It restates the methodology; it adds no rule. Where the two differ, the methodology governs.
1Reference one cell
Every SCI statistic is computed per cell: one desk, one horizon, one venue and one period. A reference that leaves any of the four open can be read more than one way. A complete reference names all four, the methodology version, and the publication it reads from:
SCI-D-SPORTS-KAL is the permanent series code for desk Sports, venue Kalshi, horizon H1 (H1 is implied when the code names no horizon). Codes are never renamed or reused; the full scheme is in methodology section 2, and every code has its own page under Series.
2Choosing the cell
Desk and horizon. Horizons depend on how far ahead a desk's contracts are listed. A horizon a desk does not have cannot be referenced: there is no Sports H30, because games are not listed 30 days out.
| Desks | Horizons available |
|---|---|
| Rates, Polls | H90, H30, H7, H1 |
| Sports, Stocks | H3, H1, H6h |
| Energy | H3, H1 |
Pick the horizon at which the forecaster commits. A service that delivers probabilities the day before a game should be measured at H1, not H6h.
Venue. Name a venue (Kalshi or Polymarket) rather than the pooled value, even where a desk has only one venue today. Sports is scored on Kalshi only, so the pooled and Kalshi values are currently identical; when a second venue enters the desk, the pooled value changes meaning and the venue value does not.
Period. Each cell is published as monthly points and as a trailing value (the headline; 12 months once the record is old enough, from July 15, 2026 until then). Monthly values move a great deal. The example below is one cell from the November 2026 publication:
| Period | SCI value | 90% interval | Events | Base rate |
|---|---|---|---|---|
| July 2026 | 0.006 | -0.027 to 0.038 | 151 | 0.500 |
| August 2026 | 0.027 | -0.004 to 0.059 | 317 | 0.497 |
| September 2026 | 0.464 | 0.350 to 0.566 | 136 | 0.493 |
| October 2026 | 0.207 | 0.095 to 0.307 | 149 | 0.500 |
| Trailing | 0.139 | 0.106 to 0.173 | 753 | 0.497 |
The base rate barely moved, yet the monthly value ranged from near zero to above 0.4. A single month reflects which events happened to be on the schedule as much as how well the market priced them. The trailing value, built on several times as many events, carries a far narrower interval. For a reference that sets a payment, use the trailing value, or measure on the same events (section 4).
Direction. The SCI value is a Brier skill score. Higher is better; 0 means no better than knowing the base rate; it can be negative. A fall in the value means the prices were less informative, not more.
3Why not the composite
SCI-ALL is the equal-weight mean of every desk that clears the publication gate at H1. It is the right number for a headline and the wrong one for a contract:
- Its membership changes. Desks enter and leave as they clear or miss the gate, so the composite can move with no change in any desk's score. Each publication reports how much of a move came from membership and how much from scores (section 8).
- It averages unlike things. A sports desk with hundreds of events and an energy desk with a few dozen count equally, so the thinnest desk's noise passes straight into the composite.
- Nobody forecasts the composite. A forecaster works on a desk at a horizon, and should be measured there.
4Measuring against the market
The published value describes the market's prices on the events in the cell. A forecaster will not have covered exactly those events, so comparing a forecaster's score with a published value mixes skill with event selection. The stronger test scores both on the same events:
- The forecaster delivers a probability for each contract before the horizon timestamp (the anchor minus the horizon), and the delivery is timestamped.
- The market price for each contract is sampled under the methodology's rules (section 5): the last captured price at or before the horizon timestamp, with the same exclusions.
- Both are scored on the contracts both priced, with the same outcomes and the cell's own base rate, giving two Brier skill scores.
- The result is the difference, forecaster minus market, with a 90% interval from resampling events (all of an event's contracts travel together). This is the same paired test the SCI uses to compare two venues on shared events (section 2).
A difference whose interval lies above zero is evidence the forecaster added information the price did not have. An interval that spans zero is not evidence either way. Scheduling and base rate affect both sides equally, so most of the month-to-month swing in section 2 cancels out of the difference.
Where a same-events test is not possible, compare with the published trailing value of the matching cell, and treat the result as indicative.
5Terms a contract should fix
| Term | What to state | Methodology |
|---|---|---|
| The cell | Series code and period, for example SCI-D-SPORTS-KAL, trailing. | §2 |
| The publication | Which publication settles each period: by date, or the first publication on or after a stated date. Values publish on the fifth business day of each month, from midnight Eastern time. | §9 |
| Methodology version | The version in force at signing. A material change gets at least 30 days of consultation, and the prior version keeps publishing for 12 months; say which version governs after a change. | §14 |
| Corrections | Whether a payment follows the value as first published or as corrected. Errors are corrected within 5 business days, with the original shown struck through. | §14 |
| Gate fallback | What happens if the cell shows insufficient data (fewer than 30 observations or 10 events): fall back to the trailing value, carry the last published value, or no determination for that period. | §6 |
| LIVE only | For a forward-looking contract, that only LIVE values count. BACKFILL values were computed under rules fixed after their outcomes were known. | §10 |
| The hurdle | For a same-events test: the margin over the market, and whether the lower end of the 90% interval must clear it or only the point estimate. | §7 |
6An illustrative clause
For orientation only. It is not legal advice and not a recommended form; contracts should be drafted by counsel.
7What the SCI does not measure
- Profit. Calibration is not trading return. A forecaster can beat the market's skill score and still lose money after spreads and fees, or trade profitably while scoring below it. Pay on skill where skill is what is being bought (forecasts, models, agents). For trading mandates, the SCI shows where prices have been least informative; it does not settle returns.
- The Scopes Fair Value. The index scores venue prices only. The Scopes Fair Value is never an input to it and is never scored by it.
- A tradable level. The SCI is a reference statistic, not designed to be tracked by a financial product, and is not a regulated benchmark.
8Licensing
Citing a published SCI value with attribution to Scopes Reference Data, LLC is free. Referencing a value in a contract that determines a payment, a fee or a product's return is licensed use, as is computing a same-events comparison from the record. Terms are in Licensing and use; to discuss a mandate, use the inquiry form on Benchmarks. How the index is governed is set out in the Benchmark Governance Framework.
The index · Values · Methodology · Oversight Log
Published by Scopes Reference Data, LLC. Research and information only, not investment advice.