entry to exit.

A new OHLCV frame arrives from the broker feed for an instrument inside an active S1–S6 config. The feed is normalized across brokers and timestamped to the script timeframe.

The frame is run through the four alpha driver engines (TRND · BRK · MR · SWP) and the 17 subsystems specific to that script. Each engine computes its own candidate trade — only one engine carries the trade forward (engine attribution is preserved end-to-end).

The candidate is scored by a gradient-boosted classifier trained on the calibrated cohort. Output is a probability of profitable resolution given the feature stack (engine, subsystem outputs, recent regime).

A FinBERT-derived sentiment score is computed against the streaming news flow for the instrument and its correlated peers. The gate blocks trades that conflict with the prevailing sentiment regime — e.g. fading a strong narrative without a regime-aligned reason.

The active regime (computed from rolling features) is matched against the regime profile that calibrated this parameter set. Out-of-distribution regimes block the trade even when statistical edge is intact.

The volatility PDE classifier computes the active Λ-Vol state from four features (level · slope · dispersion · velocity) and maps the candidate to one of five states: TREND_CALM · TREND_ACTIVE · MEAN_REVERT · STRESS_HIGH · TRANSITION. Only state-compatible engines are allowed through.

The approved trade is sized through three risk layers. ELU = Margin% × Leverage × SL% / 100 is computed at the per-trade layer and aggregated upward. Order-flow quality at the moment of placement maps continuously to leverage — thin books reduce size, deep books permit standard sizing.

The order is routed through the OMS to the broker, with retry logic on partial fills and slippage capture. The position record is created with engine, subsystem stack, AI scores, and Λ-Vol state stamped — every piece of context that informed the entry is preserved for attribution.

While the trade is open, the regime is re-checked on every new frame. A regime shift mid-trade can override the exit ladder (early exit) — this is the system's primary defense against held-through-a-state-change risk. The 4 AI layers continue to monitor; an out-of-distribution regime triggers a managed exit.

The position unwinds in four legs at engine-specific allocations. Targets are anchored to entry-time volatility; the runner is trailed by the live Λ-Vol envelope. A regime-shift override collapses the ladder into an immediate exit at whichever leg is next.

The closed trade is written to the manifest with full attribution: engine, subsystem stack, AI 1–4 scores at entry, Λ-Vol state at entry and exit, leg-by-leg P&L, slippage and commission. Drift detection compares live edge against the cycle baseline; persistent drift triggers the next R-PRISM cycle.