Atomic alpha signals are cheap, replicable, and increasingly commoditized. The durable edge lives one layer up: in how you fuse them, schedule them, and rank what to trust. Alpha Factory is that combinator, productized.
Most "signal" debates collapse three different things into one bucket. We treat them as recipe, ingredient, and scheduler, and only the scheduler is a moat.
The mathematical recipe that converts data into a directional read. Cheap to replicate; available on every quant blog.
The substrate a signal is computed on. Better ingredients raise the ceiling, but they're not a signal in themselves. Calling NLP alt-data "a signal" is a category error.
The layer that fuses many noisy signals into one bet (and a confidence), then ranks them on a scoreboard. Not a signal. Not a feed. The durable edge.
Where the four signal-pack types intersect the three data layers. Today, in the roadmap, and the speculative far edge.
| L1 · OHLCVprice · volume | L2 · 318-factorfactor library | L3 · alt-dataNLP · macro · on-chain | |
|---|---|---|---|
| IndicatorRSI · MACD · BB | ✓ | — | — |
| HMM · n-gramregime detect | ✓ | — | — |
| MLclassifier · regressor | ✓ | ◐ | ◌ |
| FactorFama-French · custom | — | ◐ | ◌ |
Two layers, by design. A Signal Factory produces individual alpha signals from any recipe in the matrix above. A Combinator fuses them into LONG / SHORT / NEUTRAL with a confidence score.
The combinator-is-moat thesis isn't a hot take. It's the only way to read the last three decades of quant history without contradiction. Two firms lived, two died at exactly the layer where the moat lives.
The canonical "many weak signals → one strong ensemble." Survived 1998 because the combinator did the work; no single signal was the bet.
Read caseA static statistical model with no combinator. Russia 1998 cracked the correlation assumption; $4.6B lost in four months, leverage 25→250:1, Fed-organized rescue.
Read caseThe combinator as craft: cross-asset linked pricing, risk embedded in the quoting engine, OCaml end-to-end. Trading revenue, not directional bets.
Read caseThe sharpest take-C evidence. A 13-year SaaS that sells infra (LEAN, brokers, data). The one signal-marketplace experiment, Alpha Streams, closed in four years.
Read caseSignals are commodities.
Combinators are moats.
Build on the moat.
