A quant desk that moves by the minute.
Screening backed by up to twelve years of option data.
Real-time stock and option flow, dealer gamma positioning, and an AI desk that reasons over the same data — so you can decide inside the move, not after it. The strategies Themis screens for are backtested on up to twelve years of real option data, then graded GO, CAUTION or LOTTO under a published rubric. You always know which one you're trading.
Sixty-eight graded. Eleven cleared.
This is the grader running. The app's strategies carry tiers and measured figures — including the thirty-one that came out LOTTO. The losers stay on the board; the dated write-up behind each verdict is in the app.
Grader finished: 68 strategies tested — 11 GO, 26 CAUTION, 31 LOTTO.
9 of the 11 GO rows are quoted on gated entries — the letter, the regime, the tenor. 55 rows are quoted pooled, every signal and no gate, which is why most of them sit lower.
31 rows commit capital and must beat SPY as a program to hold GO. 14 margin overlays are judged on per-cycle expectancy plus a stress battery at a notional cap; 23 defined-risk tickets on per-trade expectancy.
28 annualized — 19 of them the idle-in-SPY whole-calendar blend, marked BLEND — 26 per trade or per event, 5 as a percent of width, 5 as an edge over buy-and-hold. Same column, incomparable dollars, so every tile prints its cohort beside the figure, and the 20 whose denominator is not already in the figure's own suffix print that too.
Every figure is net of friction — the 10% pessimistic tier, or measured NBBO fills where the archive holds the quotes. LOTTO is what is left once the spread is actually paid: 31 structures that look fine at the midpoint and are not.
A dated scan, with expiry outcomes.
Each name has a letter grade and component grades for timing, economics and evidence, alongside credit, yield, break-even, a scenario for a 20% decline, and its gate score. This dated shelf pairs the entry assumptions with the calculated expiry outcome.
Names are shown because this shelf has settled: every position on it cleared more than 30 days ago, so this is a record of what the gate produced, not a live pick. Every figure beside a name is that position's real one. Loss −20% shows the position's scenario P&L for a 20% decline in the underlying, in dollars, using the displayed entry assumptions. It is not the maximum possible loss; losses can be larger. Settled return = P&L of one 100-share covered straddle held to expiry, as a percent of the share cost at the shelf's price: entry at the shelf's price and mid credit, ATM strike = the listed strike nearest that price, settlement at the expiry-day close (assigned leg settled at intrinsic value); gross of friction, assignment costs and dividends. Credit/yr is the entry-time credit yield annualised arithmetically. It does not estimate the return from repeating the trade. Both yield columns and the settled column use the share cost at the shelf's price as their denominator. Not settled: SATS (no daily close on or before 2026-09-18 in the archive). This is the shelf as it stood on Aug 3, 2026; every position on it had settled by Sep 18, 2026. The live shelf, re-graded every close, is in the app.
See what's forcing the move, not just the move.
Daily and historical options activity, split into the flow that actually pays for direction — plus where dealers are pinned into hedging. Both are the real views from the app.
Bullish and bearish are computed from what the premium actually pays for — call buys and put sells on one side, put buys and call sells on the other — so a name can be heavily traded and still net to nothing. The net column is the one that tells you which way the money leaned. Every session is kept, so you can replay any day in the archive. 91% of this session's premium carried a definitive aggressor read; the rest is shown as neither, never imputed. Leveraged and inverse ETFs are excluded from every number on this page; this session that removed 31 names (1,460 prints) before anything was counted.
Dealers are positioned to suppress moves. Likely mean-reverting, range-bound action.
5,492 contracts · 27 expiries · all expirations ≤ 400d · strikes within ±20% of spot · nightly per-contract snapshot · dealer-short convention
A desk of agents. Every one has a job and a receipt.
AI workflows combine the market snapshots, research and portfolio context available to each task. The research archive spans up to twelve years; a report's sources and dates show the evidence used for that report. These are the workflows available today.
[symbol] — lean bear, moderate confidence, recommended structure: skip. The drift setup is real but the report lands inside the window and the credit does not pay for the print. Re-run after the release.
Then it watches what you actually did.
Screening is the easy half. The part that decides whether you keep your gains is what happens after entry — sizing, greeks, drift, and an honest accounting of the trades that went wrong.
Broker-synced positions with live greeks, per-leg health, and P&L attribution by strategy and regime.
Review NAV, portfolio exposures and scenario losses alongside the inputs used for position sizing.
TradingView-grade charts sitting next to the screener and the audit notes. No tab-hopping mid-decision.
Grounded in your actual book. “You're short vega into CPI” — not “consider diversifying.”
News, filings, social and flow on one rail. A push has to clear severity ≥ HIGH, $10M of premium, and a six-an-hour ceiling.
The whole book re-read every close, in writing — concentration, tail exposure, pin risk, and the rolls you still have time to make.
Before a GO play becomes an order, three axes are checked in writing — validators, chain data quality, event timing. FOMC in three days or a strike with no real contract behind it is a DANGER row before the ticket, not after.
A multi-agent read on one symbol and one question — the bull case, the bear case, and what would change either — with the engine tier and its cost shown before you run it.
The seat's holdings, strikes and written reasons stay inside the app — a public page carries the shape of the book, never the book.
Every paper fill is logged with the structure, the credit and the reason it was taken. You can read a manager's thinking trade by trade inside the app, and place its order list yourself if you agree with it.
Then it turns the same machinery on a book.
Broker-synced positions, portfolio-level greeks, and a scenario grid that prices the whole book across moves and horizons. Three AI managers review it independently — read-only, suggestions for you, never trading on your behalf.
Shown on the paper book of Opus · Covered-Premium, this month's #1 chair — no customer's book, and no operator's, ships on a public page. Connect a broker and every panel below runs on yours.
Covered-family = covered = governor flag covered_premium_only or wheel_cc_only, or option_autonomous without the generalist flag; everything else = other.
This month's chairs — chosen for 2026-09 on a separate 63-mark window (40-mark tenure floor); each chair's alpha below is over its own chair window, not the common window above.
Only A/B-graded covered straddle / partial-coverage names, with the short legs held to expiry.
Only A/B-graded covered straddle / partial-coverage names, with the short legs held to expiry.
Only A/B-graded covered straddle / partial-coverage names, with the short legs held to expiry.
Every cell is the whole book repriced — stock, short options and greeks together, at flat implied volatility — so you can see where the position actually breaks before you size it.
Pricing is coming soon.
Themis is in private testing. Plans and the free trial open when the desk is ready for the public — the research and the grader are already published above.
Educational tool — decision support, not investment advice. Options carry risk of loss, and backtests are not promises.