FlashAlpha Methodology

Abstract

FlashAlpha is a computed-analytics layer over the US listed options market. From a market-data snapshot it derives a per-contract Greek set under a Black–Scholes–Merton (BSM) framework, fits a per-expiry implied-volatility surface, and aggregates the resulting sensitivities into dollarized dealer-exposure metrics — gamma (GEX), delta (DEX), vanna (VEX), and charm (CHEX) — and into derived levels such as the gamma flip, call/put walls, max pain, and the volatility risk premium.

This document states the framework, the conventions, and the design rationale behind those computations at a level sufficient for quantitative due diligence. It is written for readers fluent in BSM and options Greeks; it is not an introduction to the concepts themselves, and it deliberately does not publish the exact SVI-calibration internals (solver tolerances, parameter seeds and bounds, acceptance thresholds) that would constitute a turnkey clone of the surface.

In scope · out of scope

• In scope: the pricing model, IV solving and surface construction, exposure definitions and sign conventions, derived-level construction, 0DTE treatment, data-hygiene rules, internal validation, and operational access terms.
• Out of scope: order-level microstructure modeling; the exact SVI-calibration constants; market-data licensing terms; and any statement constituting investment advice. The exposure aggregation is a transparent dollar sum and is fully described (§4); it is not a withheld recipe.

Design principles

1. Snapshot determinism. Given the same input snapshot and the same parameters, a computation is reproducible. Caching (below) affects freshness, not determinism.
2. Leak-free history. Every historical statistic — percentiles, z-scores, smoothed surfaces — is constructed using only information available up to its as-of timestamp. No forward-looking data enters a historical value. This is a hard requirement for backtest integrity.
3. Explicit dealer convention. Exposure metrics encode a single, stated assumption about dealer positioning rather than leaving the sign ambiguous (see Exposure metrics).
4. No-arbitrage discipline (wing-bounded). SVI slices are calibrated with the Roger Lee wing-slope bound and a non-negative-variance floor enforced at fit time. This constrains the wings and ATM variance; it is not a full butterfly/calendar arbitrage check (see §3).

1. Data lineage

Coverage. The analytics universe spans 6,000+ US equities and ETFs plus major index symbols (e.g. SPX, VIX, RUT, DJX). The live universe is enumerated by probing each symbol's option chain and retaining those with a sufficient count of liquid contracts. Minute-resolution history is a deliberately smaller set of high-liquidity names (e.g. SPY, QQQ, and several large single-names) plus SPX; the flagship SPY history extends to 2018-04-16 at one-minute resolution. No options history is available before April 2018.

Cadence and timestamping. Live endpoints serve from a short-lived cache (~15 seconds), so a live value reflects the market as of at most ~15 seconds prior. Intraday Greek snapshots are computed at minute resolution across the regular session (09:30–16:00 ET). Open interest, the fitted SVI surface, and macro series are end-of-day stamped. All timestamps are emitted as ISO-8601 in the as_of field; the historical at= selector resolves against Eastern wall-clock time.

Open interest: settled vs effective. Settled open interest is the end-of-day figure from the official broadcast and is the basis for end-of-day exposure. Intraday, FlashAlpha also exposes a flow-aware effective open interest used by the Flow endpoints: settled OI plus an intraday estimate of net opening interest, in which each unit of side-classified trade volume (each print tagged buy / sell / mid by its position relative to the prevailing quote) contributes a calibrated 0.43 contracts of estimated OI change. Effective OI is floored at zero; an unclamped simulated OI is retained for diagnostics. The 0.43 factor was fit against end-of-day OI residuals — estimated intraday OI is explicitly a model, not settled fact.

Point-in-time integrity & restatements

A systematic backtest is only valid if a historical value is known to be as originally observed. Three properties govern this; FlashAlpha's current posture on each, stated plainly:

• Restatement: not versioned. The historical store is as last pulled. Quote and open-interest rows are written append-or-overwrite with no as-reported version history; a day already loaded is skipped on re-ingest, and a forced re-pull overwrites the prior value in place (fitted SVI/forward tables upsert by key). A late upstream revision that is re-pulled will therefore silently change historical results for that day. Reproducibility-critical work should pin to a fixed extract.
• Survivorship: retained, not adjusted. Rows are never deleted on delisting or expiry, so expired contracts and delisted underlyings remain queryable for any past at=. The dataset is not survivorship-adjusted in the index-membership sense, and there is no automatic symbol-change remapping beyond index roots. Cross-sectional studies must account for this.
• Leak-bounded derived stats. Percentiles and z-scores are constructed using only data strictly prior to their as-of timestamp (see §5) — the one axis guaranteed by design.
• Known historical gaps. In the minute-resolution store, per-contract bid/ask sizes and volume are not populated (carried as zero), intraday smoothed IV is gated null (backtest mode), and prior-day OI-change joins are absent. A per-symbol coverage report exposes missing-EOD, missing-SVI, and uncovered-calendar gaps.

2. Greeks

Framework. Greeks are computed under Black–Scholes–Merton. Each contract is priced from its implied volatility (below), and analytic sensitivities are evaluated on that price. FlashAlpha publishes a 15-Greek set:

• First order: delta, gamma, theta, vega, rho.
• Second order: vanna (∂Δ/∂σ), charm (∂Δ/∂t), vomma, dual delta.
• Third order: speed, zomma, color, ultima.
• Additional: lambda, veta.

Greeks are closed-form (analytic BSM partials) in double precision; theta, charm, color, and veta are reported per calendar day (the annualized value ÷ 365), vega per one volatility point, and vanna per unit of σ. Implied volatility is recovered per contract by a Newton–Raphson solve with automatic bisection fallback against the bid/ask mid, to a tolerance of 1×10⁻³ vol points over the bracket [0.1%, 1000%]; a contract whose price admits no solution is dropped rather than forced. The full Greek set is returned on the option-quote endpoint; aggregate exposures (§4) are built from these per-contract values.

3. Implied volatility & the surface

Per-strike IV. A contract's implied volatility is the σ equating its BSM price to the bid/ask mid, solved as in §2. The raw per-contract IV is surfaced directly; a model value (svi_vol) from the fitted slice is surfaced alongside it, so a consumer can compare the market-implied point against the smooth.

SVI slices. FlashAlpha fits a per-expiry SVI (stochastic-volatility-inspired) slice in Gatheral's raw parameterization. Each expiry's total implied variance is five parameters — a, b, ρ, m, σ — as a function of forward log-moneyness k = ln(K/F):

w(k) = a + b [ ρ(k − m) + √((k − m)² + σ²) ]

Slices are fit independently per expiry — no SSVI or global term-structure coupling. The forward F is recovered from put–call parity on near-ATM mids; tenor is calendar time to the 16:00 ET close on a 365-day basis. The fitted parameters, per-expiry forward, and ATM variance are exposed on the surface and advanced-volatility endpoints; svi_vol is the fitted slice evaluated at the contract's own forward log-moneyness.

No-arbitrage — what is and isn't enforced. Two static-arbitrage controls are applied during calibration: the Roger Lee wing-slope bound b(1±ρ) ≤ 4/T, which keeps the large-strike wings non-arbitrageable, and a non-negative ATM total-variance floor. FlashAlpha does not run a separate butterfly (Gatheral g(k) ≥ 0) or calendar-spread arbitrage check, and does not flag per-slice violations on the serving path. Consumers requiring a provably butterfly/calendar-arbitrage-free surface should validate independently. [FOUNDER: the public “arbitrage detection” wording on the surface endpoint overstates this — align the marketing copy with the wing-bound reality.]

Inputs & filters. The fit uses OTM quotes only (puts below the forward, calls above), each from its bid/ask mid; quotes with no bid, or a bid–ask spread wider than 50% of mid, are dropped; per-contract IVs are banded to roughly 5–200% before fitting; a minimum of five surviving points is required, within a moneyness window that widens with tenor.

Two distinct surface objects — do not conflate them. (a) The SVI slices above: analytic, forward-moneyness, per expiry, wing-bounded. (b) A rendered 50×50 grid used for visualization and historical surface queries, which bilinearly interpolates raw BSM IVs (not the SVI fit) over a log-spaced tenor axis and a linear spot log-moneyness axis (ln(K/S), roughly ±25%). The grid extrapolates flat beyond the outermost quoted strike or expiry; the SVI slice instead extrapolates per its fitted function under the wing bound. Tail Greeks therefore depend on which object you read.

4. Exposure metrics

Exposure metrics translate the per-contract Greek surface into a dollarized picture of dealer hedging demand. Each is a per-contract Greek scaled by open interest and the 100-share contract multiplier, with metric-specific spot factors below, summed per strike and across the chain. Spot is the underlying's NBBO mid for equities/ETFs, or the index level for index symbols.

Sign convention & dealer-positioning assumption

The dealer-positioning convention — calls treated as dealer-long gamma, puts as dealer-short — is applied explicitly to GEX only: call gamma contributes positive, put gamma negative, so net GEX's sign is the net dealer gamma position. Positive net GEX implies dealers are net long gamma and hedge against the move (buy weakness, sell strength — price-dampening); negative implies net short gamma, hedging with the move (price-amplifying). DEX, VEX, and CHEX do not apply a separate call/put flip: their sign comes from the Greek itself — delta is already signed (negative for puts), while vanna and charm are symmetric across calls and puts, so the sign rides on the Greek's value. This is a deliberate, stated choice.

GEX and DEX

• GEX (gamma exposure) = signed sum of Γ × OI × 100 × spot² × 0.01 (calls +, puts −). The 100×0.01 collapses to 1, so this is simultaneously the raw gamma-dollar figure (Γ·OI·spot²) and the dollar delta re-hedge for a 1% spot move. Net GEX is the chain-wide sum; its sign sets the dampening/amplifying regime above.
• DEX (delta exposure) = Δ × OI × 100 × spot — the signed delta-dollar (share-equivalent) inventory; the sign rides on delta.

Aggregate vanna & charm — VEX and CHEX

Vanna exposure (VEX). Vanna is ∂²V/∂S∂σ = ∂Δ/∂σ — how a contract's delta moves when implied volatility moves. Dollarized per contract as vanna × OI × 100 × spot (vanna per unit of σ), VEX measures how much the dealer complex's net delta — and thus its hedging demand — shifts per unit change in implied volatility. Because spot and vol are empirically correlated, a vol shock mechanically re-hedges the delta book even with spot unchanged; VEX is most consequential around volatility-regime transitions and large option-expiry concentrations.

Charm exposure (CHEX). Charm is ∂²V/∂S∂t = ∂Δ/∂t — how delta drifts purely with the passage of time, holding spot and vol fixed. Dollarized per contract as charm × OI × 100 (charm per calendar day; note CHEX carries no spot factor), CHEX measures the delta the dealer complex must rebalance simply because options age: calls slowly gain delta and puts shed it, intensifying sharply into expiry. This makes CHEX a structural read on end-of-day and weekly-expiry rebalancing flow, independent of any new information.

Aggregation. Per-strike values accumulate the contributions of every contract at that strike; chain totals are the straight sum across strikes. Expiries are pooled (a strike's bucket sums across all its expirations) — there is no DTE or near/far weighting. The aggregate is a transparent dollar sum, which is precisely what makes it reconcilable against the per-strike breakdown. A flow-aware variant recomputes each exposure on effective rather than settled OI, with estimated intraday OI carrying the 0.43 confidence weight from §1.

5. Derived levels

Gamma flip (stable, repriced zero-gamma). The gamma flip is the spot level where net GEX crosses zero — the boundary between the positive-gamma (dampening) and negative-gamma (amplifying) regimes. Rather than read a noisy zero-crossing off the current strike grid, FlashAlpha reprices net GEX across candidate spot levels — backing each leg's implied vol out of its stored gamma, then recomputing gamma at each candidate underlying — and walks a ±8% band around spot in fine steps to the first regime-consistent sign change, refining it by bisection. A smoothed per-strike interpolation (Gaussian-smoothed net GEX, significance-floored) is the fallback when the repriced root is unavailable.

Call wall / put wall. The call wall is the OTM strike (K ≥ spot) of greatest call gamma exposure — resistance where dealer hedging caps advances; the put wall is the OTM strike (K ≤ spot) of greatest put gamma exposure — support. Each is a single argmax by default; in dense at-the-money clusters (e.g. 0DTE index chains) a Gaussian-smoothed argmax is used instead, so a wall doesn't latch onto one noisy strike. Ties resolve to the lower strike.

Max pain. Max pain is the candidate settlement strike that minimizes the aggregate intrinsic value payable to option holders across the chain — equivalently the strike of greatest aggregate writer expiry value — weighted by open interest (×100), not volume. Each distinct strike is tested as the hypothetical settlement and the minimizer returned. Callers pass a single expiry's contracts (e.g. the 0DTE set) when an expiry-specific value is wanted; the routine itself pools whatever set it receives.

Volatility risk premium (VRP). The headline VRP is ATM implied volatility minus 20-day close-to-close realized volatility, VRP = IV_ATM − HV₂₀ (5/10/30-day horizons are also published; realized vol is additionally available via Parkinson, Garman–Klass, Rogers–Satchell, and Yang–Zhang estimators, annualized on 252 days). ATM IV is the average IV of the four nearest-to-spot OTM contracts in the front expiry of at least seven days. The percentile and z-score are computed over a trailing window of up to 60 daily VRP snapshots and are leak-bounded: the dated/historical path uses only snapshots strictly before the as-of date (minimum five points). This is what makes the VRP percentile safe inside a backtest.

6. 0DTE handling

Same-day-expiry (0DTE) contracts are selected explicitly and treated as a distinct regime because their Greeks behave extremely near the bell. The 0DTE endpoint returns pin-risk, expected move, gamma acceleration, and dealer-hedging reads.

• Pin-risk score (0–100). A weighted composite — open-interest concentration 30%, spot–strike proximity 25%, time remaining 25%, gamma 20% — around the magnet strike (the strike of greatest |net GEX|). The sub-scores: OI = top-3-strike OI share (saturating at 50%); proximity = closeness of the magnet to spot (max within 0.5%); time = closeness to the 16:00 close on a 6.5-hour day; gamma = the magnet's |net GEX| against ten times the median strike. Computed for a single target expiry.
• Expected move. A one-sigma move = spot × IV_ATM × √(hours-to-close / (365×24)); the published bounds are spot ± this remaining-session sigma, so the band contracts mechanically through the day. (The general, non-0DTE expected-move endpoint uses spot × IV × √(days/365).)
• Gamma acceleration. The ratio of ATM BSM gamma at the remaining time-to-close to ATM gamma at remaining-plus-seven-days — how much sharper hedging becomes into expiry.
• Theta decay. The per-hour time-decay figure, surfaced because 0DTE theta is dominated by intraday passage.

0DTE Greeks are hydrated at minute resolution across the session. [AUDIT-DEPENDENT: intraday 0DTE Greeks can be null/zero on some historical dates pending the intraday-0DTE-Greek hydration upgrade; confirm current coverage before claiming continuous historical 0DTE Greeks.] Settlement note: time-to-expiry is always measured to the 16:00 ET close, so AM-settled index options (e.g. SPX monthly SET) and cash-settlement nuances are not special-cased — for an AM-settled expiry the final-day clock is overstated. Single names are priced European (no early exercise).

7. Edge cases & data hygiene

• Stale / cached quotes. Live values reflect up to ~15 seconds of cache latency by design.
• Low open interest. A min_oi filter excludes thin strikes from aggregates on request, so a handful of stale low-OI contracts don't distort a level.
• Broken IV. Per-contract IVs feeding the surface are banded (~5–200%) to drop nonsensical solves.
• Insufficient data. Historical surface requests for dates with too few liquid OTM contracts return an explicit insufficient_data error rather than a fabricated fit.
• Backtest gating. Because the SVI surface is EOD-stamped, intraday historical queries return a null smoothed IV flagged as backtest-mode rather than leaking an end-of-day fit backward into an intraday timestamp.
• Coverage transparency. A coverage report exposes gaps (missing EOD, missing SVI, uncovered calendar dates) so consumers can see where history is incomplete.
• Wide markets. A liquidity endpoint reports spreads and execution scores per expiry, surfacing where quotes are too wide to trust.

Several further hygiene behaviors, stated explicitly:

• Stale quotes: during the session a stock quote older than five minutes triggers a refetch. There is no explicit trading-halt / LULD detection — a halted name is governed only by the staleness refresh, not excluded from aggregates.
• Corporate actions: there is no split or special-dividend normalization of strikes/OI; dividends enter only as the continuous yield q in pricing. A split mid-history can therefore produce a discontinuity unless the upstream feed pre-adjusted it.
• Mid-price: the midpoint is (bid + ask) / 2, falling back to the single available side when one quote is missing; a last-trade fallback is used only on the read path when no mid exists. It is not size- or volume-weighted.

Per-snapshot quality telemetry

Rather than ask consumers to trust a snapshot blindly, the API attaches machine-readable quality signals (on the 0DTE and screener paths) that a pipeline can gate on:

• Data-quality score = 0.4×(share of contracts with a valid IV) + 0.4×(share with valid delta & gamma) + 0.2×freshness, where freshness decays linearly from 1.0 at 0s to 0 at 300s of snapshot age.
• Greek-smoothness score = 100·e^{−(mean adjacent-strike IV jump)/0.02} — a cross-strike smile-smoothness proxy (a ~2% mean jump halves it); it runs low on thin, jagged near-expiry chains by design.
• Snapshot age (seconds since the freshest contract update) and contract count behind the aggregate; a liquidity block adds per-expiry spreads and an execution score.

Assumptions & limitations

The modeling assumptions a consumer implicitly accepts, and the limitations that follow, stated in one place. A vendor that buries these is harder to trust, not easier.

• Dealer positioning is an assumption, not an observation. Exposure metrics assume the market-maker complex is long call gamma and short put gamma (§4). Real dealer books are not observable; when actual positioning diverges — heavy customer call buying, or dealers running the opposite book — the sign and magnitude of an aggregate can mislead. The metrics are a structural lens on hedging pressure, not a measurement of inventory.
• European pricing on American contracts. All Greeks use European BSM; single-name US options are American-style and carry no early-exercise adjustment. Least accurate for deep-ITM and hard-to-borrow names.
• Wing-bounded, not fully arbitrage-free. The surface enforces the Roger Lee wing bound and a non-negative-variance floor at fit time, but runs no butterfly/calendar arbitrage check (§3). Validate independently if you require a provably arb-free surface.
• Open interest lags. Settled OI is end-of-day, so settled-OI exposure intraday is a T−1 estimate unless the flow-aware effective-OI variant (0.43-weighted) is used.
• History is not versioned or survivorship-adjusted. Late upstream revisions overwrite in place on re-pull; delisted names are retained but not membership-adjusted; splits and special dividends are not normalized (§1).
• No halt exclusion; index settlement not modeled. Halted quotes are handled only by staleness refresh; time-to-expiry is always to 16:00 ET, so AM-settled index expiries are not special-cased (§6).
• Derived levels are heuristics. Max pain, walls, and the pin score describe where hedging pressure concentrates — not forecasts of where price will go.
• Live values are cache-fresh, not tick-real-time (up to ~15s on paid tiers; see Operational & governance).
• Reproducibility is deterministic given the input snapshot, but the raw snapshot is not returned, so end-to-end reproduction needs the underlying chain. A published reconciliation vector would close this gap.

8. Validation

Validation is internal and consistency-based; FlashAlpha does not benchmark against third-party vendors in this document.

• Market-vs-model IV. Each option carries both its market-implied IV and the SVI-smoothed IV, so a divergence between the two is directly observable and acts as a per-contract fit-quality signal.
• Independent Greek recomputation. A pricing/Greeks endpoint lets a consumer recompute the expected Greek for given inputs and reconcile it against the surfaced value.
• Aggregate vs per-strike reconciliation. Chain-level exposure aggregates are the sum of their per-strike components and can be reconciled by the consumer.

• Put–call parity is used, not separately policed. The per-expiry forward is recovered from put–call parity on near-ATM mids, so parity is embedded in the forward and therefore the surface. There is no separate parity-residual data-quality gate today; a published parity-residual monitor is the most natural next validation to add.
• Shipped quality signals. The per-snapshot data-quality and Greek-smoothness scores (§7) are the live, machine-readable validation surface a consumer can gate on.

9. Versioning & changelog

The API is served under a stable v1 namespace. This methodology document is itself versioned: the content changes are tracked in a changelog while the canonical URL (/methodology) never moves, so citations and deep links remain valid across revisions.

• Document version: 1.0 (2026-06-10).
• API surface: v1. See the API changelog for endpoint-level changes.
• [VERIFY: methodology revision policy — whether the SVI model / exposure definitions carry their own internal version, and how a methodology change is communicated to API consumers.]

Operational & governance

Methodology aside, an enterprise buyer evaluates access, limits, and terms. The current facts:

• Access tiers: Free, Basic, Growth, Alpha, and Enterprise. The SVI surface, advanced/realized-vol analytics, and VRP are Alpha-tier; exposure metrics and Greeks are available from the entry tiers. See /pricing.
• Rate limits: enforced as a per-user daily request quota (not a per-second limit), with X-RateLimit-* headers and HTTP 429 + Retry-After on exhaustion; Alpha/Enterprise are unlimited. The limiter fails open if its store is unavailable.
• Freshness by tier: the live cache TTL is tier-dependent — ~15 minutes on Free, ~15 seconds on paid, ~1 second on Alpha/Enterprise.
• Authentication: REST uses an X-Api-Key header (or apiKey query param); the MCP server additionally supports OAuth 2.1 (PKCE + dynamic client registration).
• SDKs: official clients for Python, JavaScript, .NET/C#, Go, and Java.
• Legal & commercial: Terms, Privacy, and Risk disclosure are published; redistribution/commercial and enterprise scoping (bespoke SLA, data-residency, compliance) are handled via commercial enquiry and the quant-team page.

How to cite

@misc{flashalpha_methodology,
  title        = {FlashAlpha Methodology},
  author       = {{FlashAlpha}},
  version      = {1.0},
  year         = {2026},
  howpublished = {\url{https://flashalpha.com/methodology}},
  note         = {Accessed: 2026-06-10}
}

References

This methodology builds on established literature in volatility modeling and dealer-positioning analytics. Anchoring references:

1. Black, F., Scholes, M. (1973). The Pricing of Options and Corporate Liabilities. Journal of Political Economy. — Merton, R. (1973). Theory of Rational Option Pricing. Bell Journal of Economics.
2. Gatheral, J. (2004). A parsimonious arbitrage-free implied volatility parameterization with application to the valuation of volatility derivatives (SVI). Presentation, Global Derivatives.
3. Gatheral, J., Jacquier, A. (2014). Arbitrage-free SVI volatility surfaces. Quantitative Finance, 14(1).
4. Lee, R. (2004). The moment formula for implied volatility at extreme strikes. Mathematical Finance, 14(3). — basis for the SVI wing-slope bound.
5. SqueezeMetrics (2017). The Implied Order Book (gamma-exposure / dealer-positioning framework). white paper.
6. Pan, J., Poteshman, A. M. (2006). The Information in Option Volume for Future Stock Prices. Review of Financial Studies, 19(3).
7. Parkinson, M. (1980); Garman, M., Klass, M. (1980); Rogers, L.C.G., Satchell, S. (1991); Yang, D., Zhang, Q. (2000) — range-based realized-volatility estimators used in the realized-vol analytics.
8. Corsi, F. (2009). A Simple Approximate Long-Memory Model of Realized Volatility (HAR-RV). Journal of Financial Econometrics. — RiskMetrics (1996), EWMA (λ = 0.94).

Open methodology questions (remove before publication)

Almost everything in the body has been resolved against the source code. What remains before publication is short. Remove this section, the draft banner, and the noindex tag once cleared.

FOUNDER — decisions only you can make:

• §1 Data-lineage paragraph — describe the upstream feeds accurately without asserting a source license. (The investigation identified the actual vendors; that detail was deliberately kept out of this public doc and handed to you separately.)
• §3 Align the public “arbitrage detection” marketing copy on the surface endpoint with the wing-bound reality (no butterfly/calendar check is performed).
• Governance: decide whether to publish a Security/Compliance overview, an SLA, and a real Status page before promoting this doc to enterprise buyers.

AUDIT-DEPENDENT — gate on the data audit:

• §4 Exposure values (not formulas): publishing specific GEX/DEX/VEX/CHEX numbers should wait on the open-interest-attribution / GEX-divergence audit.
• §6 0DTE: confirm historical intraday 0DTE Greek coverage (null/zero pending the hydration upgrade) before claiming continuous history.

Resolved from source (now stated in the body): effective-OI 0.43 weight & its meaning; risk-free rate (single scalar; 4.5% live / 10Y historical) & continuous-dividend model; European-for-all exercise; closed-form Greeks, per-day theta/charm; IV solver (Newton + bisection, tol 1e-3, vs mid); SVI forward log-moneyness, independent per-expiry slices, Roger-Lee wing bound (no butterfly/calendar check); SVI-slice vs 50×50 raw-IV grid distinction; all four exposure formulas & the GEX-only sign flip; plain-sum aggregation (no proprietary weighting); gamma-flip repricing algorithm; wall argmax + tie rule; max-pain OI-weighting; VRP HV20 + trailing-60-day leak-bounded percentile; pin-risk weights & sub-scores; expected-move & gamma-acceleration formulas; index AM/PM not modeled; no halt exclusion; no corporate-action normalization; mid-price construction; quality-score & greek-smoothness construction; restatement/survivorship posture; rate-limit/auth/tier governance.

VERIFY — worth a founder confirm:

• Live vs historical engine duality: the customer-facing live path and the historical hydrator use independent BSM/IV implementations (e.g. live IV tol 1e-3 Newton+bisection vs historical bisection tol 1e-6). Confirm they are intended to be methodologically equivalent, and consider documenting the difference.
• §9 Methodology revision/versioning policy — how a methodology change is versioned and communicated to API consumers.