fithia2/dev/phase7_deliverables/live_rollout_and_capital_ramp.md

Live Rollout and Capital Ramp

Objective

This document defines how capital is introduced gradually into live trading. The strategy is assumed to be profitable enough in research and stable enough in paper trading to justify real capital exposure. The purpose of the ramp is to limit operational losses while verifying that live performance matches expectations.

Rollout Principles

• Start smaller than feels necessary.
• Increase capital only after stability thresholds are met.
• Treat unexpected operational behavior as more serious than short-term PnL noise.
• Favor “no trade” over ambiguous or degraded execution.
• Never ramp capital because of a short streak of wins.

Capital Stages

Stage A — Pilot Live

Purpose: confirm end-to-end live behavior at minimal financial risk.

• Max portfolio capital allocated: 5% to 10% of planned production capital.
• Max per-trade risk: 0.10% to 0.15% of total account equity.
• Max simultaneous positions: 1 to 2.
• Overlay features optional; may be disabled by default.
• Human approval required before order submission.
• No strategy branching. Use only the primary long continuation engine.

Promotion conditions:

• Minimum 20 live trades OR 4 live trading weeks, whichever is later.
• No unresolved severe incident in the observation window.
• Data freshness violations below threshold.
• Order/position reconciliation success on every trading day.
• Live slippage within acceptable band vs simulated expectation.
• Parser and feature quality review passing rate above threshold.

Stage B — Controlled Live

Purpose: allow limited autonomy while validating consistency.

• Max portfolio capital allocated: 20% to 35% of planned production capital.
• Max per-trade risk: 0.20% to 0.25% of total account equity.
• Max simultaneous positions: 2 to 3.
• Human approval optional for low-risk standard setups; required for anomalies.
• Attention overlay may be enabled only if Phase 5 research justified it.

Promotion conditions:

• Minimum additional 30 live trades.
• No medium-or-higher unresolved incidents for the last 20 trading days.
• Outlier slippage days investigated and closed.
• Drift checks stable for the observation window.
• Daily review completed on time for at least 90% of trading days.

Stage C — Production Live

Purpose: run the approved strategy under standard production governance.

• Max portfolio capital allocated: 100% of approved production capital.
• Max per-trade risk: determined by current risk policy, but may not exceed Phase 0 limits.
• Max simultaneous positions: as allowed by approved risk policy.
• Human approval required only for exceptions, config changes, or degraded mode.

Ramp Freeze Conditions

Capital increase is automatically frozen if any of the following occur:

• New unresolved high-severity incident.
• Reconciliation mismatch not resolved by the next trading day.
• Live execution drift outside threshold for 3 days in a rolling 10-day window.
• Parser confidence or gold-set accuracy drops below threshold.
• Data source instability causes candidate corruption or missing mandatory features.
• Strategy underperforms simulation beyond allowed tolerance after sufficient sample.

Ramp Rollback Conditions

Capital must be reduced to the prior stage if any of the following occur:

• Emergency kill switch triggered.
• Two severe incidents within 30 calendar days.
• Unauthorized config or code drift detected.
• Daily max-loss breach caused by operational fault.
• Strategy produces materially different positions than approved research spec.

Live Eligibility Filter

A trade candidate is eligible for live execution only if:

• It is produced by an approved strategy id and approved config version.
• Mandatory market, event, and risk features are present.
• Candidate timestamp is valid for session timing rules.
• Risk guard passes.
• Required approval state is satisfied.
• Data freshness SLA is satisfied for all mandatory sources.

Approval Matrix

• Standard candidate in Stage A: human approval required.
• Standard candidate in Stage B: optional if no anomalies.
• Candidate with degraded overlay, stale optional features, or manual override: human approval required.
• Candidate generated under degraded mandatory source mode: reject or no-trade.

Metrics for Ramp Decisions

The capital committee or operator should review:

• Realized expectancy.
• Win rate and payoff ratio.
• Slippage vs expectation.
• Fill quality.
• Average holding period.
• Risk utilization.
• Cancel/reject ratio.
• Incident count by severity.
• Review SLA adherence.
• Drift metrics.

Required Records

Every ramp decision must create an auditable record containing:

• date and effective stage
• prior stage and new stage
• approver
• summary of evidence
• performance window used
• incidents considered
• config and model versions in effect