01Data
Ingest & normalize
Exchange ticks, order books, fundamentals and alternative data are cleaned, aligned and stored point-in-time — every value stamped with the moment it was knowable.
- Point-in-time
- Survivorship-free
- Corporate actions
Quantitative Technology Company
Unending, unborn, all things abide at the ORIGIN.
Touch the origin.
The origin is the source, the cause and the result.
Manifesto
Every number is measured from an origin.
We begin each idea at zero — no edge assumed, nothing believed until the data insists.
The lab forms hypotheses, tests them on clean data, and publishes what it rejects as openly as what survives.
Start at the origin. Prove it with data.
Six principles
Each teaching, held up to one principle — then proved.
- 01śūnyatāEmptinessStart from H₀: α = 0. Assume no edge.Sutra · Principle · Proof
- 02paṭiccasamuppādaDependent originationEvery alpha must name the conditions that produce it.Sutra · Principle · Proof
- 03aniccaImpermanenceEverything is re-tested, continuously.Sutra · Principle · Proof
- 04anattāNon-selfRules over emotion.Sutra · Principle · Proof
- 05majjhimā paṭipadāThe Middle WayModerate size. Moderate models.Sutra · Principle · Proof
- 06satiMindfulnessContinuous, real-time risk monitoring.Sutra · Principle · Proof
Beneath the six · risk
(0, 0)
Every curve starts at the origin.
This one is pure chance: a new seed on every visit, printed so you can reproduce it. Some draws will look like skill. That is the point — and why every idea here starts at zero.
The origin curveseed ·····
Demo: simulated prices for illustration only, not market data.
How it works
A disciplined process, from first tick to final fill.
Illustrative data
ingest · normalize4 feeds → as-of t 02Research
Hypothesis to evidence
Ideas become factors, factors become backtests. Walk-forward validation with realistic costs separates a real edge from an overfit one — before any capital is at risk.
- Walk-forward
- Costs & slippage
- Out-of-sample
research · validatewalk-forward · OOS 03Signal
Alpha to portfolio
Validated alphas are blended and sized under explicit constraints — exposure, turnover, liquidity — into target weights you can trace back to their origin.
- Alpha blend
- Risk model
- Constraints
signal · constructnet 0.00 · gross 2.0× 04Execution
Intent to fills
Target weights become child orders. Schedulers and smart routing work the book to minimise market impact, and every fill is measured against arrival price.
- VWAP / POV
- Smart routing
- TCA
execution · routefills vs arrival 05Risk
Measure, limit, learn
Pre-trade checks, live limits and latency monitoring run continuously. What production teaches us flows straight back into research — closing the loop.
- Pre-trade checks
- Live limits
- Feedback loop
risk · monitorp99 2.4 ms
What we build
One stack from raw data to a decision you can trace.
- 01
Alpha research
A reproducible research environment: point-in-time data, factor libraries and notebooks that promote straight to production.
- 02
Signal infrastructure
Low-latency Go services that turn research into live, versioned signals — monitored, auditable and traceable to their origin.
- 03
GPU-scale backtesting
Thousands of parameter sweeps on GPUs with realistic costs, slippage and risk limits — results in minutes, not days.
- 04
Learning hub
Field notes on market microstructure, portfolio construction and systematic execution, written by practitioners.
By the numbers
Measured from a single point of origin.
Illustrative figures
- market events stored point-in-time, per day
- 12.4B
- venues and data feeds normalised into one schema
- 42
- median signal-to-order latency
- 0.38ms
- backtest runs per hour on GPUs
- 12,800
Under the hood
Built like trading infrastructure, because it is.
Go
Low-latency services
Signal and order services written in Go on the standard library — small, predictable, measured at p99.
Python · CUDA
Research on GPUs
Vectorised research and backtesting in Python on GPUs: thousands of parameter sets per sweep, reproducible by seed.
Next.js
A fast, accessible web layer
Server-rendered by default, served through an edge proxy, with every interaction reachable by keyboard.
Research notes
Field notes from the desk, written by practitioners.
Data
Point-in-time, or it didn't happen
Why every value in a research dataset needs the timestamp of when it became knowable — and how look-ahead bias hides in plain sight.
Read note5 min
Research
Walk-forward validation, honestly
A single in-sample / out-of-sample split is one draw from a noisy process. Rolling the window forward tells you far more.
Read note6 min
Execution
Measuring execution against arrival
Implementation shortfall in plain terms: what it measures, why arrival price is the honest benchmark, and what to log.
Read note4 min