Y-Cap_

The Self-Improving Edge Factory

THE QUESTION

What's the fair value
of something?

Every price is a bet on the truth, and the gap between the quote and fair value is the only thing anyone profits from. Prediction markets are the cleanest place to play it: known truth, live mispricing, priced by retail and too small for the giants. The gap is our lane.

WHAT WE'RE BUILDING

An engine that prices
every settling market.

Not a fund. We estimate the fair value of any binary outcome better than the market and trade the gap. We start where feedback is fastest, 5-minute crypto, then carry the same machine up into every binary that settles: 15m and 1h crypto, sports, politics.

Trading P&L is the proof. The factory is the asset. The endgame: own the engine that prices truth across every prediction market, and the capacity that compounds with it.

THE REFRAME

This isn't a hedge fund.
It's a self-improving
edge factory.

We don't predict prices. We estimate fair value better than the market and trade the gap, then build a machine that discovers, tests, and retires that edge autonomously, forever. The trading P&L is the exhaust. The factory is the asset.

HOW WE PRICE

Millions of agents, debating fair value.

N
Independent agents price
the same outcome,
different signal, model, data
→ p*
They debate to a consensus
fair value, sharper than any
single read, variance falls
Δ
Market price − fair value
= the edge.
Sharper p*, cleaner trades

The marginal agent isn't more coverage, it's another independent read on the truth.

IT COMPOUNDS

Every trade makes the next one smarter.

Compounding
Data moat
every fill and decoded
bot deepens the corpus
Knowledge moat
355 post-mortems
sharpen the next signal
Meta-learning
the system improves
how it improves
THE GROUNDWORK

30,000 predictions. Six months of data.

0k
Predictions placed on
live money, fills and outcomes
WS ticks
WebSocket price ticks,
captured for 6 months
CLOB 1s
Order-book ticks second by
second, 6 months of depth
OI · 6mo
Open interest history
captured continuously

Started on 5-minute crypto on purpose: the fastest feedback on fills, colocation and execution.

PROOF · ALREADY RUNNING

433 ideas in. 7 survived.

Hypotheses proposed
0
Killed by 9 gates
0
Promising → shadow
0
Live with real money
0

The machine kills 82% of its own ideas autonomously, before a dollar is risked.

THE TREADMILL · THE LOOP

Every edge decays. Find fast. Gate hard. Retire faster.

Edge Proposer
AUTO
Backtest
9 GATES
Shadow Build
APPROVAL
Shadow Validate
7+ DAYS
Real Proposal
APPROVAL
Deploy Live
AUTO
Monitor
CONTINUOUS
Retire & Learn
WRITE-BACK
Automated Human Gate Continuous Writes to memory
THE PROBLEM · EXECUTION

We get filled
when we're wrong.

Adverse selection: the worst fills land exactly when the price is about to move against you. Some bots are beating us, and we need to know how.

THE PROBLEM · LATENCY

A millisecond race,
not a nanosecond one.

Polymarket matches off-chain in the cloud, so the fill is a race to the matching engine over a public-internet hop. The lever is colocation, not silicon.

CAPACITY

Capacity is a sum, not a ceiling.

Qmax = 2 ρ0 e per market
Qtotal = Σ 2 ρ0 e across thousands of books

Same machine, same formula, summed across 15m and 1h crypto, sports and politics. The floor, not the ceiling.

WHERE WE STAND

Three axes of edge. We own two.

Jane Street
Axis · Speed
FPGA · ~100 ns
Nanosecond latency wars. A cloud CLOB tops out at the millisecond, won by colocation, not silicon.
Renaissance
Axis · Signal
Decades of data
Mining decades of history. Prediction markets are too new to have decades to mine.
Y-Cap
Axis · Discovery + Adaptation
Find fast. Gate hard. Retire faster.
Find the signals fast, run real algorithms once they pass the statistical gates, then retire them faster if the edge dies.

Why now: prediction-market open interest is compounding 6× a year. We grow with it.

THE ASK

What I need to pull this off.

Capital + 2-3 quant engineers
A small, elite team and the trading capital to scale what already works
Rigorous colocation experiments
Systematic placement tests across AWS and GCP to win the fill
Latency below 100ms
Cut the hop to the matching engine under 100ms and beat every bot to the book
Faster sports data APIs
Lower-latency sports feeds to capture mispricing the moment it opens

The machine is built and running. This round buys the team, the colocation and the speed to push the same engine into every bigger book.

THE DISCOUNT ON A BILLION-DOLLAR FUTURE
raising
$0
@
valuation
$0
0%
Trading Capital
0%
Quant Engineers
0%
Infrastructure
A cheap call on a billion-dollar future, priced before the capital arrives.
Y-Cap · The Self-Improving Edge Factory
MACHINE
EXPERIMENTS 433
KILLED·BY·GATES 355 (82%)
SHADOW 70
LIVE 7
AGENT·TYPES 8
STAT·GATES 9
OI·YoY 6× ↑
VOL·RUN-RATE $300B/yr
MARGINAL·COST/AGENT ~$0
EXPERIMENTS 433
KILLED·BY·GATES 355 (82%)
SHADOW 70
LIVE 7
AGENT·TYPES 8
STAT·GATES 9
OI·YoY 6× ↑
VOL·RUN-RATE $300B/yr
MARGINAL·COST/AGENT ~$0
← → NAVIGATE · F FULLSCREEN