Data & Methodology

Two questions run through the whole section. Where does each number come from, and which of those numbers can you read through an accessor? Not every computed quantity is exposed, so every page below separates what the platform computes from what the Signal language lets you read, and names an unexposed quantity plainly rather than inventing an accessor for it.

The pipeline

Every quantity descends from one quote and a handful of point-in-time inputs. The chain forks after the inversion: the Greeks come off the same solved volatility, the forward comes off put-call parity, the dealer-exposure aggregates come off the Greeks and open interest, and the earnings split comes off a fit across expirations.

Each branch has its own page. Start at how every number is produced for the branch-by-branch walk and the single consolidated table of what is computed versus what is readable.

Three guarantees that hold everywhere

Three properties hold on every number in this section, regardless of which page produced it.

• Point-in-time, no look-ahead. Every read takes the most recent value whose timestamp is at or before the read minute, and returns null before a series' first observation. A number computed at 11:00 saw only what was knowable at 11:00. This is the same backward as-of fill that aligns mixed-frequency Signals on the union spine, and a daily value becomes readable only at its session-close label.
• Per-contract marks, not an arbitrage-free surface. Each contract's IV and Greeks are solved from that contract's own price. Nothing couples one strike to its neighbors or one expiry to the next, so the marks are locally correct and globally unconstrained. Smiles can cross, term structure can invert. A fitted arbitrage-free surface is not computed; the published marks are per-contract.
• Every option.* read is 2D bilinear interpolated. An option read is interpolated in days to expiry and in delta or log-moneyness across the contracts that bracket your target, so you read a point off a continuous surface rather than a single listed strike. When the chain is too sparse to bracket the target, the read is null, never extrapolated.

Each page states the default that shapes its own number (the quote basis, the adjustment switches, the window semantics, the fill mode) before it shows the number.

What each page covers

• How every number is produced: the branch-by-branch map of the whole pipeline, and the one consolidated table of every methodology quantity marked computed-and-readable or computed-but-not-readable.
• Data coverage: the symbols, the 1-minute resolution, the session, and the history depth. The single page that states coverage; every other page links here.
• Data inputs and point-in-time guarantees: the six raw feeds, the minute-timestamp convention that aligns them, and why spot is the raw close.
• Implied volatility: the bid/mid/ask solve, the quality grading, the parity recovery, and where the marks stop being reliable.
• Greeks: the five readable Greeks, the per-Greek unit scalings, why 0DTE Greeks explode at the time floor, and vanna's unexposed role.
• Implied forward, carry and rates: one forward per expiry from parity, the continuously-compounded rate behind it, the carry and borrow diagnostics, and forward-centered moneyness.
• Dividends: discrete escrow instead of a yield, the point-in-time projected schedule, and why only regular cash dividends enter options pricing.
• Dealer exposure: GEX and VEX: the gamma and vanna channels, their shared per-1%-move denomination, and the 0DTE exclusion.
• Earnings volatility and implied move: the variance decomposition that splits ambient from event vol, the implied move, and ex-earnings IV.
• Stock price adjustments: the four price series, the dividend-adjustment look-ahead trap, and why options pricing still uses the raw close.
• Fills, marks and slippage: how a backtest prices every entry, exit, and per-minute mark, and the honest ledger of what realism is modeled.

How we model fills and margin

A backtest turns the data above into a result through two engines, fills and margin.

Fills. You choose a fill mode on every backtest: mid prices at the quoted midpoint (the optimistic bound), bid_ask crosses the full quoted spread (the pessimistic bound), and the Backtest.ai fill model books a calibrated estimate of where between mid and the touch the order actually fills, the same pricing paper and live trading use. A size-aware capacity layer applies in every mode, a lot clamp at sizing time plus an additive market-impact penalty at fill time, and it connects to position sizing and capital, where the requested lots are chosen. A fixed per-contract commission applies, settable to zero. Your equity curve is marked at mid the whole time you hold, whatever your fills paid. See fills, marks and slippage for the mechanism and backtest assumptions and realism for picking a mode.

Margin. Two models, selectable per run. REG-T strategy-based margin decomposes a position into recognized structures and charges the least-cost set. TIMS portfolio margin nets offsetting risk across the whole book under a grid of price shocks instead of charging each structure in isolation. See margin and auto-decomposition.

Trust that lives in other sections

Two dimensions a trader checks before risking capital are documented outside this section.

Validation. How a signal becomes a model, walk-forward by default, and how the platform reports out-of-sample fit rather than an in-sample number, is the validation layer. That is where a result is stress-tested against the look-ahead and curve-fit failure modes this section's point-in-time rules are built to prevent.

Research to live. The path from a backtested strategy to a running account, through a point-in-time snapshot, is the snapshot model. It carries the same no-look-ahead discipline forward into live operation.

What this methodology does not do

Stated plainly, so nothing here surprises you later:

• No partial fills, fill probability, or adverse selection. A fill is all-or-nothing at the resolved lot count and always executes at the modeled price.
• No arbitrage-free surface. The published IV is a per-contract mark, not a fitted surface.
• No in-backtest dividends and no early assignment. Dividends enter options pricing through escrow, but a backtest does not credit or debit a dividend to an open position, and it does not model early assignment.
• One symbol per run. A signal or backtest runs against a single symbol at a time.

Every number reaches you through the DSL accessors documented under data accessors. Where a quantity on the pipeline has no accessor, the page that produces it names the quantity plainly rather than inventing one.