How SYZYG Works
This document describes the data pipeline, scoring methodology, signal taxonomy, regime classification system, and governance controls used by SYZYG. It is designed for transparency — not marketing.
Data Pipeline Architecture
SYZYG ingests market data through a medallion architecture (Bronze → Silver → Gold) with governance enforced at each layer. Raw data is never deleted. Every transformation is versioned.
Raw ingestion from primary sources. Unmodified CSVs, API responses, and feed snapshots. Stored as-received with ingestion timestamps.
Cleaned, validated, and normalized. Schema enforcement, duplicate removal, missing data flagging. Anomaly detection runs here.
Derived analytics: scores, regime classifications, signal outputs. Every Gold table traces to Silver inputs with version lineage.
Production delivery layer. Gold outputs pushed to Supabase PRO with RLS policies. Freshness SLA enforced (≤15 minutes).
SYZYG Scoring System
Each security receives a composite SYZYG Score derived from multiple independent signal planes. The score maps to a proprietary celestial classification that determines regime context and execution class.
The composite score aggregates independent measurement planes. Each plane contributes directional bias and is weighted according to the current regime context.
Regime Classification & Truth Table
SYZYG classifies market conditions into regimes using a two-dimensional matrix: Market Regime (macro environment) × Signal Regime (per-security score). The intersection determines the Execution Class — a governance-enforced behavioral constraint.
The Truth Table is a locked governance document. It defines what actions are permissible under each regime-signal combination. Two invariants are enforced:
Realized volatility classified using Yang-Zhang estimator with quantile buckets. Determines whether the market environment favors trend-following or mean-reversion strategies.
Four-quadrant risk matrix: COBRA (high vol, trending), CALM (low vol, range), GIANT (high vol, choppy), NOVA (low vol, breakout). Each quadrant modifies signal weights.
Identifies dominant mechanical forces: dealer gamma positioning (long vs short), vanna/charm pressure direction, and institutional flow imbalances.
Governance Controls
SYZYG operates on a principle of Deterministic Restraint: when data freshness, coverage, or integrity falls below thresholds, outputs are automatically suppressed and the system enters Safe Mode.
All required inputs must be ≤15 minutes old at decision time. Stale data triggers a BREACH event and suppresses affected outputs. No exceptions.
Automated enforcement layer that prevents publication during staleness, API failures, schema anomalies, or regime uncertainty. The system protects users from itself.
Continuous monitoring of model accuracy proxies and distribution shifts. When baseline deltas exceed thresholds, breach events are emitted and outputs are downgraded.
Every output traces to its inputs, transform versions, regime context, and gating state. Sufficient for audit reconstruction and incident forensics.
Research Decision Ledger stores every Decision Artifact with cryptographic integrity. No hindsight edits. No unverifiable claims. Append-only by design.
Defect taxonomy, SPC control charts, root cause analysis, and preventive controls. Signal accuracy measured and improved continuously, not assumed.
Automation & Pipeline Execution
SYZYG runs on Google Cloud Platform infrastructure. The daily pipeline is a 9-job cron chain executing Monday through Friday, with governance checks at each stage. Publisher v6 pushes scored data to Supabase every 30 minutes during market hours.
Shariah Compliance Screening
SYZYG provides Shariah compliance screening across all covered securities. The methodology follows established Islamic finance principles with quantitative thresholds applied to financial ratios.
Known Limitations
Transparency requires acknowledging what the system cannot do. These limitations are operational constraints, not future roadmap items disguised as caveats.