Methodology

How Spread Foundry sources data, builds candidates, and scores quality.

Data Sources

Spread Foundry pulls equity and options market data from third-party providers. The default cadence is end-of-day (EOD) with provider-specific intraday refresh where licensed. The Service does not guarantee real-time data unless a specific view is explicitly labeled as such.

Snapshot Modes

• EOD:post-close assembly of the day's official marks and derived spreads. This is the canonical snapshot for overnight research.
• Intraday: periodic refresh during regular hours, where provider entitlements allow. Latency varies by provider.
• Same-session:an intraday-like snapshot run off the current session's data before close.

Each snapshot carries a status (READY, DEGRADED, PARTIAL, UNAVAILABLE) and a publish timestamp visible in the app header.

Snapshot Quality

Every published snapshot carries an explicit snapshot_class (tradable / degraded_tradable / analytic_only / unavailable), two structured reason fields — resolution_reason (the worst signal present anywhere) and dominant_reason (why the snapshot landed in its class) — the modal quote_basis and mark_basis, and an execution_intent classifier. Per-spread rows inherit a worst-leg tradeable_classrollup. The platform annotates rather than excludes: every structurally valid candidate is generated and surfaced — the default view shows all classes, quality-ranked, with suitability badges (off-regime, stand-aside, jump risk, incomplete) — and the explorer's “hide low quality” toggle is an explicit, reversible narrowing, never the default. The full contract spec — enum values, threshold defaults, and the per-leg classifier algorithm — lives in the snapshot-quality specification.

Spread Candidates

Candidates are generated from the option chain for each eligible underlying against a configurable set of strategy families (vertical credit, vertical debit, iron condor, etc.). Eligibility is driven by liquidity, width, and moneyness filters; the exact filter set is profile-scoped and versioned — the current profile and its version are visible in the app header strip.

Scoring

Each candidate is scored on multiple axes (edge, liquidity, risk). The scoring model is deterministic per profile version and is not a prediction of outcome — it is a ranking signal over the candidate set for the published snapshot. Scores are not adjusted for portfolio, tax, or commission context and do not incorporate your positions.

Limitations

• Quote-basis detail (bid/ask spread width, mark source, degradation reason) is being surfaced progressively.
• Historical backtest performance is not exposed in the MVP.
• Corporate-action adjustments depend on provider data; late adjustments can cause brief re-scoring after the canonical EOD publish.

Changes

Methodology changes — new filters, new scoring components, provider switches — are reflected in the visible profile version. Major changes will also be called out here.