Jann/investment-sandbox
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Closes #SYS-DEPLOY-001 - Integrate Two-Stage Engine and Alpha Regressor Matrix

Browse Source
This commit is contained in:
Antigravity Agent
2026-06-17 19:47:01 +02:00
parent 615521ed99
commit dcb59c17f0
11 changed files with 640 additions and 127 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
DEV_LOG.md
View File

@@ -372,6 +372,28 @@ This document tracks all modifications, npm packages, active compilation states,
* **Active Bugs**: None.
* **Type Checker Status**: Verified 100% clean type verification (`npx tsc --noEmit` returns exit code 0).
---
## [2026-06-17] - Alpha Unit Activation & Pure Quantum Fusion (#SYS-DEPLOY-001)
### Added
* **Microstructure & On-Chain Ingestion Engine**: Constructed [etl.py](file:///c:/Users/jannr/.gemini/antigravity/scratch/investment-sandbox/backend/core/etl.py) extracting:
* **On-Chain Layer**: Reconstructs UTXO sets. Computes Young-to-Old Supply Velocity ($V_{\text{supply}}$) and Adjusted SOPR (aSOPR), tracking STH/LTH-SOPR at the 155-day boundary.
* **Perpetual Derivatives WebSockets**: Computes Open Interest-to-Market Cap Ratio ($\Theta_t$), Implied Liquidation Distance ($D_{\text{liq}}$) using Gaussian smoothing kernels, and Rolling Z-score of compounded 8h funding rates ($Z_F$).
* **Microstructure Pipeline**: Splits Cumulative Volume Delta (CVD) into institutional vs. retail cohorts to evaluate Divergence ($Div_{\text{CVD}}$), and estimates rolling Kyle's Lambda ($\lambda_{\text{Kyle}}$) price impact.
* **Stationarity & Memory Transformation**: Applied Dickey-Fuller (ADF) optimal $d^*$ search (target p-value < 0.01) for Fixed-Width Fractional Differentiation (FFD) and causal Rolling Median Absolute Deviation (MAD) scaling in [pipeline.py](file:///c:/Users/jannr/.gemini/antigravity/scratch/investment-sandbox/backend/core/pipeline.py).
* **Markov-Switching GJR-GARCH Volatility Regime Router**: Integrated Student-t innovations MS-GJR-GARCH regime gated matrices routing microstructure features during Calm states, and prioritizing On-Chain/Derivatives features during Turbulent states.
* **PIMP & Boruta Pruning Filters**: Implemented Permutation Feature Importance (PFI) vs. randomized null distributions ($M=50$ target permutations, significance $p < 0.05$) and Boruta shadow feature pruning.
* **Two-Stage Selective Inference ML Engine**: Standardized XGBoost/RF directional classifiers (Stage 1) gated by secondary correctness Reliability Meta-Learners (Stage 2) with strict execution confidence thresholds ($\theta_{\text{conf}} = 0.55$) triggering Zero-Exposure states (`0.5` probabilities) upon failure.
### Fixed
* **Visual Calibration Math**: Verified JSX braces wrapping and double-escaped backslashes render beautifully in `CryptoDemo.tsx` calibration dropdown.
### Active Bugs / Compile Status
* **Active Bugs**: None.
* **Type Checker Status**: Verified 100% clean type verification (`npx tsc --noEmit` returns exit code 0).