How our numbers are made

Everything is built on the Delta Exchange India public API — the venue where these contracts actually trade, in INR margin. Live pages re-read the chain on every request (with short server-side caches, typically 1–5 minutes). Deribit’s DVOL and global open interest appear on the Overview only as market context and are labeled as such.

A recorder snapshots every BTC/ETH/XAUT chain hourly: ATM IV per expiry, straddle marks, expected move, open interest, walls and net gamma. Those snapshots power IV History, OI change and wall history. A separate live recorder keeps a trade tape (every print, April 2024 → today) that powers the Simulator, Flow History and the Radar’s backtests.

Option IVs are the exchange’s mark IV; where a strike quotes both sides, we use the call/put mean. ATM IV per expiry is the mark IV at the strike nearest spot with both sides quoted. Greeks are plain Black-Scholes at mark IV (r = 0), matching the venue’s own convention. The expected move is the ATM straddle mark divided by spot — the market’s own price for movement over that expiry, no model of ours involved.

IV percentile windows: for BTC, short-tenor and ~30d percentiles are computed against 2.2 years of reference data — 810 daily IVs reconstructed from the venue’s own trade tape (Black-76 inversion of near-ATM trades, vega-weighted daily medians; validated against an independent monthly-tenor series) — plus the live hourly record collected since 21 Jul 2026. BTC ~7d, and all ETH/XAUT percentiles, currently rest on the live record alone (~5 weeks and deepening hourly); read those as “vs recent weeks,” not years. Wherever a stat includes the long reference, the page says so.

Three different 0–100 numbers appear near each other. They measure different things:

• Regime conviction (Overview, Brief) — how cleanly current conditions match one playbook (premium selling, premium buying, carry/calendar, directional). It is the dominant bucket’s share of scoring points: low conviction means signals conflict, not that the site is unsure of its data.
• Structure fit (Playbook, Builder fit cards) — how well one specific structure suits the current regime, from a transparent weighted heuristic over IV percentile, VRP, term structure, trend and skew.
• Expiry Radar score — which listed expiry deserves attention today, weighted from IV percentile, term-structure position, tradeability (OI + active strikes; the venue publishes no bid/ask depth) and tenor fit. It is a heuristic, not a validated signal — which is why the Radar also shows measured historical edge separately.

The Radar’s Historical Edge block is measured, not scored: for every historical BTC expiry on the tape, the suggested structure is entered at the last traded prices at/before the entry-DTE checkpoint (no look-ahead; a leg must have actually traded within 48h), held to expiry, settled at intrinsic against the 12:00 UTC (5:30 PM IST) settlement, fees included. Win rate, average P&L, profit factor and drawdown are reported per tenor with confidence stars scaled by sample size. When profit factor is below 1 the block now says plainly that the strategy lost money net despite a high win rate.

The Movement engine (Movement Odds) was validated walk-forward over 800 days with 500 out-of-sample days, Brier-scored against a ladder of benchmarks (base rate → raw option-implied → haircut implied → models). The shipped forecast is the only variant that beat raw implied pricing out-of-sample, its full calibration table is published on the page, and every live forecast is logged and scored against what actually happened 24 hours later.

Independent replication. The two findings that carry the most weight here were re-tested on data this site does not otherwise touch: 8.9 years of 1-minute Binance history (2017–2026, 4.7M bars per asset), a different venue covering six years our own tape never saw. The day-of-week cycle held — Friday movement ran 21% below average on BTC (lower in 9 of 10 years) and 24% below on ETH (10 of 10), p < 0.00001. The volatility forecasts held — out-of-sample R² of 0.47–0.64 across 8.9 years versus the 0.49–0.63 published here from 2.3 years, so those figures are not a small-sample artefact.

A control test on tokenized gold then changed the explanation. Gold was expected to show no weekend pattern; it showed the same one at the same size, and gold’s cash market is closed all weekend. The cycle is therefore driven by traditional market hours — which crypto inherits despite trading 24/7 — rather than by anything crypto-specific. We consider that a stronger basis for the finding, since exchange opening hours are structural rather than a crowded position that decays.

Delta’s standard options taker fee is min(0.03% of notional, 10% of premium) per option, per side — the rate every backtest here is costed at, because it was in effect across the sample. The payoff builder shows entry fees against the credit; the Radar backtests include fees; the Simulator charges them on every fill. We show entry fees rather than guessing exit fees (which depend on the exit price) — a round trip roughly doubles the entry number. Fees matter more than most parameter choices at this venue’s premium sizes.

Promotion, as of August 2026: Delta is running an “Options Carnival” discount that cuts the taker fee to min(0.010% of notional, 3.5% of premium) — a two-thirds reduction. The historical numbers on this site keep the standard rate (correct for their sample period); live economics going forward are cheaper while the promo lasts. Because fees are the largest cost on far-OTM premium selling, this is not a rounding error — but it is promotional and Delta can revert it at any time, so treat the discount as temporary rather than the new baseline.

• One venue for options. Every options-derived number here comes from Delta Exchange; most global BTC option open interest sits elsewhere, so gamma/OI maps understate the whole market. No public source publishes a long crypto-options trade tape, so this one is not second-sourceable. The findings that rest on price rather than options — the day-of-week cycle and the volatility forecasts — were re-tested on 8.9 years of independent Binance data (2017–2026) and held.
• Gamma sign is a guess. Dealer-positioning conventions from equity markets may not transfer to crypto — the gamma page carries this warning prominently and shows the mass split by side instead of pretending to know.
• No order-book depth. The public API exposes no bid/ask depth history, so “tradeability” proxies use OI and active strikes.
• Short IV history (see above) behind percentiles, for now.
• Backtests are the past. Fee-in, no-look-ahead replays of what happened — not predictions of what will.

Everything on this site is educational analytics — never investment advice, a signal service, or a prediction. You place, size and own every trade yourself.