Shipped
AMRS
Adaptive Multi-Regime Score adds a regime-classification layer before any buy or sell decision. The purpose is to stop one static strategy from being applied to every market environment.
Shift Forge · Build in public
The current edge is not prediction.
It is state detection.
The Research Lab tests whether market internals, options activity, cross-asset behavior, and causal feedback loops detect a meaningful transition more reliably than SPY price alone.
Active program
Research modules
A tool does not become production logic because it sounds intelligent. It must improve out-of-sample detection beyond a simpler price-only baseline.
In development
Market State Lab
Reproducible research on SPY versus RSP, QQQ, large components, breadth proxies, sector leadership, VIX, and one- through ten-day lead-lag relationships.
Causal detection
Transition engine
Combines persistence, dominance flips, contradiction checks, delays, and known data distortions to identify where a system sits in its lifecycle.
Market State Lab · Study 01
Is SPY telling the whole story?
Internal divergence
- SPY versus equal-weight RSP
- SPY versus current large components
- Breadth and concentration proxies
- Sector leadership and rotation
Lead-lag testing
- One-day through ten-day horizons
- Whether divergences precede meaningful SPY moves
- Walk-forward or time-split validation
- Comparison against a price-only baseline
Options and cross-assets
- Price versus options positioning and volatility
- Rates, dollar, gold, oil, credit, and equity agreement
- Confirmation delays and false-positive control
- Whether signals persist or reverse quickly
Look-ahead controls
- Current-weight experiments labeled as current-weight
- No claim that today's index weights existed historically
- Time-split or walk-forward testing
- Assumptions and limitations published with results
The research question is fixed: does this signal improve detection of a verified market-state transition beyond SPY price alone?
Publication standard
What gets released
Method
Reproducible pipeline
Data sources, cleaning, assumptions, feature definitions, and validation windows must be documented.
Evidence
Results and failed theories
CSV tables, charts, benchmark comparisons, and negative findings are all part of the record.
Decision
Ranked usefulness
Signals are ranked by out-of-sample contribution, not by the attractiveness of one historical chart.
Limits
No unsupported certainty
Research findings remain experimental until repeated and integrated without materially increasing false positives.