Receipts
Every pundit says they called it. These are the receipts.
Before the US midterms, Receipts wrote 13 published forecasts to Walrus Memory, each beside the market price at that moment. Sui certified every record by Nov 1, 20:37:10 UTC, 50 hours before the first polls closed. Now the results are in, and Receipts remembers who beat the market.
Who to listen to
A voice earns trust only by beating the market price at the time Receipts logged the call. Picking the favourite that wins earns nothing.
Edge is the market's Brier score minus the voice's, ×100, averaged over 13 scored calls. The bar is a 90% interval. Every interval still crosses the market line, so Receipts keeps weighting all 3 voices like the market until one of them clears it.
The receipts
One slip per race. Open a line to see where the call was published, the Walrus blob that remembers it, and the Sui transaction that certified it before the vote.
Will the Democratic Party control the House after the 2026 Midterm elections?
Will the Republicans win the Iowa Senate race in 2026?
Will the Republicans win the Kansas Senate race in 2026?
Will the Democrats win the Ohio Senate race in 2026?
Will the Democrats win the Maine Senate race in 2026?
Will the Democrats win the Michigan Senate race in 2026?
Will the Republicans win the South Carolina Senate race in 2026?
Will the Democrats win the Texas Senate race in 2026?
Ask the memory
Scores, results and trust profiles are written back to Walrus Memory, and every new profile supersedes the last one instead of overwriting it. Search it directly.
How a receipt works
- Log before the event. The call, the voice, its probability and the live market price go to Walrus Memory on Mainnet.
- Certify on Sui. Walrus certifies the blob in a Sui transaction. Its timestamp proves the call existed before the result. Tatum's Sui gateway confirmed all of them, and the app re-reads each blob object live through Tatum's Sui gRPC.
- Resolve. When the market resolves, Receipts writes the result to memory.
- Score against the market. Each call gets edge = market Brier − voice Brier. Being right about a 95% favourite earns almost nothing.
- Remember the trust. Each voice gets a new trust profile that supersedes the last. The ranking above is read from that memory.
What is real here
| Part | Status |
|---|---|
| 13 forecasts and 9 market snapshots on Walrus Memory Mainnet | Real, certified Nov 1, 20:36:29 UTC to Nov 1, 20:37:10 UTC |
| Forecasts themselves | Real, published by Silver Bulletin (Nate Silver), VoteHub, Decision Desk HQ, as reported by Newsweek on Oct 31 |
| Market prices at logging | Real Polymarket prices from Nov 1, 20:35:10 UTC, stored beside each call |
| Results | Polymarket resolutions as of Nov 4, 21:05:43 UTC |
| Who to trust | Measured, and honestly inconclusive: no voice is yet distinguishable from the market |
Limits: 13 scored calls from 3 voices is a small sample. The shrinkage constant k is estimated from the data rather than chosen, and right now the estimate is infinite, because the spread between voices is smaller than the noise. Receipts says so instead of inventing a ranking. The forecasts were published between Oct 29 and Oct 30 but the market price was read when Receipts logged them, so the market had up to 3 more days of news. That tilts every score against the voices.