Algorithms & Optimization
The math that turns a working hypothesis into a durable edge.
Pricing algorithms, matching engines, scoring models, and optimization systems built from first principles. We have validated 5,697 statistical patterns against a decade of tick data. We know what walk-forward validation looks like and how lookahead bias hides in naive backtests.
5,697
Statistically validated patterns discovered
< 5ms
Signal generation p99 latency
Zero
Lookahead bias. Walk-forward validated.
METRICS
By the numbers
< 5ms
Signal generation latency
10yr+
Historical backtest window
100%
Algorithm IP ownership
3 wks
Avg time to live trading
CAPABILITIES
What we build
01
Pricing and bidding algorithms
Dynamic pricing that updates on the signals that actually move demand: inventory level, time-to-event, competitor price feed, and session-level conversion probability. Auction and bid optimization systems with budget pacing and click fraud filtering built in.
02
Scoring and ranking models
Lead scoring, credit risk scoring, content ranking, and search relevance. We build composite scores that combine static firmographic features with real-time behavioral signals and calibrate the output so the score means the same thing across different audience segments.
03
Matching engines
Supply-demand matching with custom criteria weighting and constraint enforcement. We have built marketplace matching systems that simultaneously optimize fill rate, counterparty preference scores, and geographic constraints across thousands of concurrent requests.
04
Statistical validation infrastructure
Walk-forward backtesting with realistic friction, causal inference with proper control group construction, and hypothesis testing with pre-specified alpha levels. We design the statistical test before we look at the data so we are not p-hacking our way to a result.
ASSUMPTIONS
Backtest assumptions, written down
A backtest is a hypothesis under specific frictions. We document every assumption before any signal earns the green light. When the assumptions move, the validation report flags it.
| Assumption | Value used | Why it matters |
|---|---|---|
| Slippage model | 0.5 bps on liquid, 5 bps on small-cap | Naive zero-slippage backtests inflate Sharpe by 30 to 50% on intraday signals. |
| Fee structure | IBKR tiered + SEC + FINRA + clearing | Round-trip fees flip break-even strategies into losers under realistic costs. |
| Borrow cost | Hard-to-borrow rate per security | Short-side P&L without borrow costs is fiction. We use the actual loan rate. |
| Survivorship | Point-in-time universe | Survivorship bias adds 1 to 2% annualized return on equity backtests. We use the universe as it existed. |
| Walk-forward window | Refit monthly, validate 30 days | Static train-test splits hide regime decay. Walk-forward exposes it. |
| Capacity | Max 5% of average daily volume | Strategies that work at $1M break at $50M. Capacity limit is in the backtest. |
APPLICATIONS
Where this applies
- 01Dynamic pricing for a B2B SaaS product. We built a seat-and-usage pricing engine that adjusts list price based on company size, deal stage, and historical conversion at each price point. Average deal size increased 18% without a change in win rate.
- 02Risk scoring for alternative lending. Multi-factor underwriting model trained on 24 months of repayment history. The model ranked the top quartile of applicants with a default rate 4x lower than the portfolio mean.
- 03Fraud detection scoring. Real-time scoring pipeline on payment events with sub-10ms p99 latency. Built in Rust. Integrated with Stripe webhooks to decline high-risk transactions before settlement.
- 04Signal discovery and validation. Systematic search across 5,697 pattern candidates on 10 years of minute-bar equity data. Walk-forward validation with slippage and fee models to filter candidates with genuine out-of-sample edge.
TECHNOLOGY
Tech stack
PROCESS
How we deliver
Every engagement follows the same three phases. No surprises, no scope creep.
Signal Research + Backtest Design
We research signal candidates against 10+ years of tick history. Walk-forward backtests with realistic slippage, fees, and position sizing validate edge before live.
Paper Trade + Risk Model Calibration
Algorithms run in paper mode with live data feeds. Position limits, drawdown stops, and Kelly sizing are calibrated until live metrics match backtest expectations.
Live Deployment + SRE Handoff
Algorithm deployed to your execution environment with alerts, kill switches, and a P&L dashboard. Full IP transfer and SRE runbook included.
GET STARTED
Ready to build?
Most projects ship in 2 to 4 weeks. Fixed price. Full IP transfer.