Project examples

Prediction market research

Six research projects show how a measurable outcome can be matched to source data, tested against a simple comparison, and tracked beside a listed contract. They are project examples, not client engagements or realized trading returns.

Discuss a research project

Research areas

Current project examples

Results below are dated research snapshots. Source data, contracts, and prices change.

Daily count · tested model

TSA passenger screenings

Daily national screening counts are forecast from published TSA history and checked against listed passenger-count thresholds.

2026 holdout
76,723 mean absolute error versus 147,083 for the comparison model.
Contract check
255 finalized thresholds matched the recorded TSA counts.
View the TSA project →
Season total · research

Mammoth snowfall

Main Lodge monthly snowfall history bounds the December 24 result at each listed threshold; some past outcomes remain unresolved from monthly data.

Known outcomes
194 of 333 historical threshold outcomes are determined; 139 remain uncertain.
Status
On the provable subset, the model scored worse than a simple baseline.
View the Mammoth project →
Monthly count · tested model

SpaceX launches

Launch records and monthly pace support a forecast of the number of completed launches in a calendar month.

Nine-month holdout
0.437 launches mean error versus 0.563 for the baseline.
Source check
Official settlement timing still requires separate review.
Market series ↗
Year-end price · research

New Jersey gas price

AAA's state regular-gas average is tracked against a year-end price threshold.

Holdout
The current model did not beat an unchanged-price baseline.
Status
Probability intervals need improvement before trade decisions.
Market series ↗
Weekly views · research

Netflix Top 10 views

Netflix's published weekly Global Shows (English) chart supplies the outcome and view-count history.

Holdout
The view-count estimate beat a previous-week baseline.
Status
Threshold probabilities did not beat a simple historical baseline.
Market series ↗
Hourly demand · source-aligned check

ERCOT daily peak load

Hourly Texas grid load is used to estimate the day's highest reported TOTAL and compare it with daily MW thresholds.

2026 archive holdout
2,391 MW mean error versus 3,863 MW for a seven-day peak baseline.
First-report check
5,268 MW mean error across 27 October dates.
View the ERCOT project →

Model estimates are not verified trading returns. A client project can include historical tests, dated quote capture, probability calibration, cost review, documentation, and a handoff process. Scope depends on data availability and the contract's exact settlement rule.

Project scope

From a question to a reviewable model

1

Define

Confirm the outcome, settlement source, and available contract.

2

Build

Collect historical data and preserve source versions and timestamps.

3

Test

Compare a model with a simple baseline on later outcomes.

4

Transfer

Document updates, decision controls, and the client's operating process.

Discuss a project