Does dynamic alpha survive costs?

36 regularized models · chronological validation · separate $800 accounts. Inspect exit outcomes, regimes & sizing

No edge proven; no live promotion. Fresh 48h turnover improved, but historical and newest-segment losses and weak forecast accuracy prevent deployment. The submitted controller, sizing caps and −10% / −15% safeguards are unchanged.

Loading hash-audited research…

Compare the frozen model

Net sums below add independent experiments, not one portfolio. Splits overlap; never add split totals. A cash control is included. All returns are modeled, not live.

Evaluation segment

Trading costs

Market

Baseline versus polynomial alpha and cash
Policy	Cases	Net sum	Turnover sum	Trades	Mean return	Worst DD

Inspect an account

Replay case

Closed modeled trades
Side	Entry UTC	Exit UTC	Exit reason	Net

What can we optimize?

Features use closed bars only; entry is the next open. Degree 1/2, Ridge penalties 10/1,000/10,000, 30/60-minute horizons and three cost-aware presets were selected only on three purged, chronological validation folds. No forward-looking price, IV, OI or book data is injected into past candles.

Validation-only block permutation importance; association, not causal impact. Correlated features and regime changes limit interpretation.
Feature	MSE increase (bps²)	Dispersion

All 36 validation configurations

Scores combine mean net return minus half absolute drawdown; no volume reward. Small counts are not proof.
Configuration	Objective	Mean return	Trades	Turnover

Fresh native Derive tick and quote evidence

Captured instrument

Native depth-priced frozen-alpha transfer checks; short captures only, not a two-day backtest
Instrument	Costs	Trades	Net	Turnover	Status

Real IV, Greeks, OI and orderbook depth are captured for context. Public OI does not reveal dealer positioning: GEX is unknown. Options minimum-size quote markouts are diagnostics, not executed or strategy-approved spreads. No multi-day option-price history or paired venue fills have been established.

Targeted 2–5-day option surface

Actual REST option quotes, IV, Greeks and OI; current context only, not historical training data. Empty/zero bids remain explicit.
Instrument	Delta	IV	OI	Bid	Ask	Tight at receipt?

Options friction at minimum size

These are diagnostic quote crossings, not strategy trades. Zero-duration rows measure immediate round-trip spread/fees only; they do not measure a future return. Risk and delta compatibility remain separate gates.

Minimum-size spread cost probes, never executed P&L
Buy / sell	Duration	All-in loss proxy	Immediate cost	Fits $4 risk?	Delta-compatible?

Next safe experiment

Test fewer return/trend factors, stronger shrinkage and regime/cost gates on a genuinely new rolling native quote archive. Keep signal prediction, trade eligibility and execution sizing separate. Do not loosen minimum-order or loss limits to chase turnover.

Download sanitized model, candidates and results

Paper only. No private account, signing key or trading controls. Submitted strategy 5f68255 unchanged.