● Research · verifiable math

Verify the engine. Don't trust it.

We publish enough for you to verify the math is honest — academic citations for every factor, live self-tests that run on every page load, measured crisis performance, and public prediction intervals. What we don't publish: the exact formulas, weights, and constants that compose it all. That's our edge.

LIVE

Factor methodology

Which academic papers each factor comes from, what they measure, and when they fail. Architecture without constants.

LIVE

Mathematical self-test

Structural invariants the engine must satisfy — posterior sums, correlation bounds, weight constraints. Runs live on every request.

LIVE

Measured backtest

Historical IC, Sharpe, drawdown. No curve-fitting — out-of-sample walk-forward CV with honest failure disclosure.

LIVE

Crisis stress test

How the engine performed on 2008, 2020, and 2022 crashes. Coverage rates and IC — including the failures.

LIVE · MATURING

Forward track record

Out-of-sample window opened 16 May 2026. 1d + 7d horizons measured live; 30d and 90d roll in as their windows mature. No backtest dressed as performance.

What we don't publish

The specific formulas, factor weights, calibration constants, composition order, pattern thresholds, and amplifier bounds that turn the published architecture into a working system. This follows the AQR / Renaissance precedent: publish enough to be peer-evaluated and criticised, withhold enough that replication takes months of engineering, not days of copying. Qualified investors and funds can request a detailed methodology walkthrough under NDA — founders@framler.com.

Research — Framler working papers | Framler