Pinned Tweet
$NFA is live on Robinhood Chain.
Nine months ago this was a prototype that didn't work well enough to ship. We went quiet, tore it down, and rebuilt it. Today it's a working forecasting engine with a public record and a token to align the people who use it.
Read more:
One thing we didn't design for: the judge discounts the same shape of advocate argument across completely unrelated markets. Unsourced claim, confident tone, no decisive fact. Shows up in a Fed market and a ceasefire market the same way.
That emerged. We didn't build it in.
Most of what we've written about NFA, the engine, the debate structure, the independence of the judge, exists to serve one end: producing forecasts that are measurably better than the market, and proving it in public.
On 24 June a market opened on whether the US-Iran 60 day negotiation period would be extended.
It ran to 20 August and traded close to $1.8M.
🧵
Sports markets: solved. Quant models, base rates, xG.
Crypto price action: solved. Orderbook data, funding rates.
Will this ceasefire hold. Will the Fed dissent. Will this regulator blink.
Nothing. A third of the prediction market volume and traders are still just reading news and guessing.
Better tooling is what we are building. NFA allows you to forecast market results with market aligned accuracy.
nfa.club
NFA call highlight: Virtu Q2 earnings.
Market: 79% yes. NFA: 6%. It resolved No.
The threshold was $1.82 - the market was pricing the familiar question (does VIRT beat consensus?) while the actual question was "does a final figure exceed the company's own two-week-old preliminary number?", which almost never moves. The market answered the wrong question at 79%. The engine understood the resolution criteria and said 6%.
nfa.club/pages/accuracy
Prediction markets don't have a liquidity problem. They have a research throughput problem.
$1,000 across 20 positions is a correct way to trade them. Then you have to actually research 20 markets, each with its own actors, deadlines and resolution nuances. That's a week of work for $50 a position.
NFA solves this!
nfa.club
First batch of private beta testers is in.
They've got access to the full engine: pick a market, watch the advocates argue it out, read the judge's ruling, see the divergence against the market price, trade the outcome.
Next batch opens shortly. Get on the list: app.nfa.club
Weekend batch resolved: 14 of 17 calls correct, engine Brier 0.147 vs the market's 0.212 on the batch. Running beat-rate at 58%.
Wins and losses, updating live: nfa.club/pages/accuracy
For Private beta access, register at nfa.club
The NFA engine architecture, in one picture. Four layers around one core.
1. The engine researches, debates and judges a market into a probability.
2. The data layer anchors forecasts to real measurements.
3. The accuracy layer scores every forecast against the market on resolution.
4. The surfaces put it in reach.
Full technical details in the docs: nfa.gitbook.io/nfa-docs
Fresh batch of markets resolved overnight. The engine beat the market on 71% of that batch, moving the running beat-rate up to 60%.
The NFA engine is outperforming @polymarket by a 6% margin: Brier 0.210 vs 0.224.
Wins and losses, updating live: nfa.club/pages/accuracy
We're bringing the first group of beta testers in this week.
Small batches, approved wallets. If you trade @Polymarket seriously and want to put the engine through its paces, get on the list.
app.nfa.club
3/ This is the start of it, not the peak.
Public release, live trading, the MCP server for agents,
@Kalshi and beyond. The record keeps growing and stays public the whole way, including the parts we get wrong.
Register for early access at app.nfa.club
