Who We Are
Not a black box.
A glass box.
Every trade emerges from a deterministic pipeline — real-time event feeds, closed-form signal extraction, Kelly-constrained sizing. The system either has a measurable edge or it does not trade. Every decision is logged. Every claim is mathematically defensible.
Structural edge, not prediction
Markets are information systems. When new information arrives, prices adjust — but not instantly, and not uniformly. We build the infrastructure to identify, measure, and act on those adjustment windows before the market closes the gap.
Deterministic execution, no exceptions
The signal pipeline is pure math — zero learned parameters in the execution path. The risk layer enforces Kelly criterion, hard position limits, and daily loss circuit breakers. No discretion. No overrides.
Infrastructure over intuition
Edge in liquid markets is fragile. We don't rely on discretion, pattern recognition, or market intuition. The system either identifies a statistically measurable opportunity or it does nothing. Speed and precision of execution are the moat.
Auditable at every layer
Every trade is persisted to an immutable audit trail with full signal context: market conditions at trigger, model output, position size, and latency from signal to fill. No decision is opaque.
Our Approach
Event-driven systems at the
speed of information
Our pipeline is engineered around a single thesis: real-world events reprice markets with measurable latency. We close that gap with real-time signal ingestion, deterministic evaluation, and fully automated execution across seven concurrent async agents.
See System BenchmarksSystems
Sub-microsecond
execution path
Hot-path benchmarks on Apple M-series (n=100,000 calls each). At 500 ticks/sec across two symbols, the feature + pricing budget is under 1 ms/sec total. Measured usage: <0.5%.
Risk-neutral contract probability
Rolling window with exact tick expiry
Expected value with full fee model
7 Concurrent Async Agents
CryptoFeed, Feature, WebSocket, Scanner, Risk, Execution, Resolution — all coordinated through typed asyncio queues. No shared mutable state between agents.
133+ Passing Tests
11 test modules covering units, integrations, and property-based tests via Hypothesis. AAA pattern throughout. Coverage on core math and all agent state machines.
Immutable Data Pipeline
All models are frozen dataclasses. Every signal, tick, and trade opportunity is immutable from creation to resolution. No hidden side effects in the execution path.
Risk Management
Every threshold
has a derivation
No magic numbers. Every control below is defined with a mathematical rationale tied to fee structure, position sizing theory, and observed market dynamics.
| Control | Value | Rationale |
|---|---|---|
| Position sizing cap | 0.25× theoretical max | Absorbs model probability estimation error without ruin. Conservative multiplier applied to theoretical maximum — standard for systems with unverified edge at current fill count. |
| Minimum edge | 4% | Covers Kalshi taker fee (peaks at 1.75% at P=0.5) plus bid-ask spread. Below this floor, expected value is negative after fees. |
| Max concurrent positions | 5 | At 10% max per position, this caps deployed capital at 50%. Preserves margin buffer against adverse correlated moves. |
| Max single exposure | 10% bankroll | Per-position concentration limit. Prevents any single contract from causing catastrophic drawdown regardless of Kelly output. |
| Daily loss circuit breaker | 20% bankroll | Hard halt on correlated loss scenarios. Also proactively blocks new positions if pending worst-case loss would breach this threshold. |
| Consecutive-loss halt | 3 losses → 24h pause | Catches streak-based edge decay not yet large enough to trigger the percentage-based circuit breaker. Distinct gate, distinct scenario. |
| Max signal age | 2 seconds | Core latency arb invariant. Signals older than 2 seconds are stale — the Kalshi market has likely already repriced to fair value. |
| Max hours to expiry | 4 hours | Far-dated contracts can stay mispriced indefinitely. Latency arb requires near-expiry convergence pressure to realize the edge. |
| Spread floor | 4% | Kalshi offers no maker rebates. Markets tighter than 4% spread leave insufficient edge after fees to justify a position. |
| Burst cooldown | 30s between fills | Prevents a single momentum signal from filling all 5 position slots in rapid succession before the market has time to reprice. |
| Symbol concentration | 2 per symbol | Caps correlated BTC/ETH exposure. Prevents a single asset's adverse move from triggering 5× the intended concentration. |
Careers
Build autonomous
systems that trade
We need people who can design real-time event pipelines, build deterministic signal architectures, and engineer execution systems that operate without human intervention in adversarial markets. If you find yourself reading the source code more than this page, reach out.
Get in TouchGet in Touch
General inquiries,
partnerships, or press
Kinzie is a proprietary trading research firm. We trade our own capital and do not accept outside investment.
Kinzie is a proprietary trading research vehicle. Nothing on this site constitutes an offer or solicitation to invest, a recommendation to buy or sell any security, or investment advice of any kind. Kinzie is not registered as an investment adviser, broker-dealer, or commodity trading adviser. All trading activity described is paper-mode simulation. Past performance, whether real or simulated, is not indicative of future results. All information is provided for informational and research purposes only.