Derivative reference

Use this when you want the compact trust ceiling first. For page-by-page instructions, definitions, and the full workflow manual, use Guide. For staged theory and study tracks, use Learn.

Open Guide Guide: methodology Guide: data sources Open Learn

Topic	Current implementation	What it does not establish
Signal timing	Signals are computed from daily bars and shown as LONG / SHORT / FLAT with persistence-based strength.	No execution latency model, no intraday fill guarantee.
Backtest return model	Position on bar t is applied to forward return t→t+1. Commission and slippage are deducted via turnover.	No market impact model, no partial fills, no borrow fees.
Out-of-sample	Optional terminal holdout split reports OOS metrics on the last X% of observations.	Still only one terminal holdout. It does not establish regime invariance or protect against repeated tuning on the same idea.
Walk-forward validation	Sequential holdout folds now include a purge gap between train and test windows and report fold count, average Sharpe, worst-fold Sharpe, and positive-fold ratio.	Still lighter than combinatorial purged CV, nested tuning, and full live execution validation.
Benchmark	Default is same-ticker buy-and-hold; you can override with a single ticker or a weighted basket such as SPY:0.60, IEF:0.40 on the backtest, compare, and portfolio pages.	Still not a full optimizer: no risk-parity constructor, no automatic factor benchmark builder, and no benchmark search/ranking workflow yet.
Paper scoreboards	Scoreboards rank forward-tracked paper strategies and portfolios with governed forward scores that scale for live-record maturity, modeled frictions, validation quality, and owner caps.	Still not a brokerage-linked live track record, and start-date selection remains a product-layer rather than broker-audited constraint.

Data Sources

Provider order

Yahoo Finance → Nasdaq Data Link → Stooq. Results attempt a primary provider first, then fall back through secondary vendors, with provenance carried into the result view.

Continuity rule

If trustworthy market history cannot be established, the run is surfaced as non-decision-grade rather than being allowed to masquerade as research evidence.

Residual risk

Adjustment mismatches, stale rows, missing dividends, survivorship bias, and exchange-specific quirks can still distort a result even when a provider succeeds.

Ticker coverage contract

Free keeps the narrow reliability contract: core liquid symbols only. Pro unlocks extended best-effort symbols and weighted benchmark baskets, not blanket long-tail coverage.

Provider expansion is telemetry-driven. Escalate new provider work only after at least 20 resolve attempts, 6 unsupported events, and repeated pressure in the same symbols.

risk disclosure

Past performance does not guarantee future results. Backtests are model outputs conditional on data quality, cost assumptions, and implementation details. They do not establish regime robustness, crowding resilience, or live execution quality.

product status

Alphrex is still an evolving product. Paper tracking and scoreboards are implemented at the product layer, but broker integration, hardened code sandboxing, and a full production-database rollout remain incomplete.

Methodology Summary

Data Sources