A standalone algorithmic trading fund, a mortgage-rate intelligence service, and a pre-registered research layer — each backfill-validated, each surfaced live. Discipline over narrative: every strategy proves itself before it trades real capital.
The Platform
Algo Fund
Standalone systematic algorithmic trading fund (Alpaca paper). Post-earnings drift, cross-sectional momentum, and structural-carry sleeves with real-fill NAV, Sharpe, drawdown, and per-strategy attribution.
Open fund dashboard →Rates Intelligence
Mortgage-rate direction (V1.5.1 regime-conditional + V1.2 calibrated) with an asymmetric LOCK / FLOAT / WATCH default and a public out-of-sample track record. For chadinvestorlending.com borrowers.
Open rates dashboard →Research Registry
The intelligence-layer cockpit: every live signal + pre-registration in one view — status, backfill calendar, overlap/collinearity map, and the disconfirmed graveyard. Decision-support only; never auto-trades.
Open registry →Research & Development
Every strategy and signal is treated as a falsifiable scientific hypothesis. Each is pre-registered with a SHA-256 hash and timestamp before any data is tested, held through a mandatory cooling-off period, then backfilled exactly once — no re-tuning, no retrofitting. Outcomes are recorded whether they confirm or fail. The result is a contemporaneous, immutable, auditable trail of technological research under genuine uncertainty.
54
Research programs catalogued
signals + strategies + pre-registrations
29
Hypotheses disconfirmed
tested, failed their bar, recorded
10
Live, validated
shipped after backfill validation
13
Under test / resolved
pre-registered, in the pipeline
Pre-registration
Every hypothesis is hash-locked (SHA-256) with a timestamp and a mandatory cooling-off period before testing. The record can't be edited after the fact.
Run-once discipline
Each backfill runs a single time against frozen parameters. No iterate-and-retry, no curve-fitting — a result is the result.
Adversarial statistics
Politis-Romano bootstrap CIs, Holm-Šidák correction, López de Prado purged k-fold CV with embargo, walk-forward out-of-sample splits.
Disconfirmation as evidence
Failed hypotheses are documented, not buried. The disconfirmed graveyard is the clearest proof of genuine experimentation under uncertainty.
Self-calibrating substrate
Bayesian IC shrinkage, regime-conditional information coefficients, slippage-feedback loops, and cross-strategy crowding penalties — all measured continuously.
Immutable, auditable trail
Pre-registration files, backfill outputs, and verdicts are version-controlled and dated — a contemporaneous record of the research process end to end.
How It Works
Regime detection
FRED macro (TEPPER) + 8-feature HMM behavioral regime
Signal generation
PEAD drift, cross-sectional momentum, structural carry
Risk governance
Per-position / per-strategy / gross caps, CVaR gate, circuit breaker
Execution
Alpaca paper, 5-shape routing, real-fill attribution
Self-calibration
Bayesian IC shrinkage, walk-forward OOS, slippage feedback
Intelligence layer
Pre-registered crash-precursor + valuation signals (decision-support)
Rates models
V1.2 / V1.5 logistic, public OOS track record
Discipline
Hash-locked pre-reg + cooling-off; run once, no retune