How GammaRips Works

The rule, in one breath

You should be able to restate the whole selection rule from memory. That is the point. Six steps, run once a night, in this order:

1. About 3,500 optionable US names. The starting universe. It is refreshed weekly.
2. Traded 3M+ shares that session. Thin names come off first. A stock nobody traded does not have an option market either.
3. Chain carries 25+ listed strikes. A shallow chain is not a market you can work an order in.
4. Top 100 by combined liquidity rank. The rank is the z-score of chain dollar volume plus the z-score of share volume. Nothing else goes into it.
5. Bullish names only. A hard gate since 2026-06-11. Calls only.
6. One out-of-the-money call per name. The contract inside each name is chosen on contract liquidity, not on how unusual the print was.

What comes out is a pool of roughly 40 to 50 contracts, published every morning on /signals and, in structured form, on the MCP for connected agents. This funnel went live 2026-08-24.

The two safety rails

Two rails run each morning before the paper cohort's entry. They remove risk, they do not hunt for winners. And they screen the cohort's entry, not the published pool:

• Earnings exclusion. The cohort drops any ticker reporting earnings during its hold. Holding options through an earnings report is a documented way to lose money. The published pool can carry earnings-window names, so run the same check on any candidate you take.
• VIX ≤ VIX3M. Short-term fear must sit at or below long-term fear. If VIX runs above VIX3M, the rail fails closed and the cohort stands down for the day.

Some days the rails clear the board, or the pool comes up empty. The cohort stands down. No forced trade. Skipping is correct behavior.

What we publish

The product is the data layer: the pool, the surfaces, and the methodology. There is no pick.

The daily clock

Executability is the only measured improvement

The funnel used to select on unusual activity. It turned out to select contracts that were harder to trade than ordinary liquid ones, because unusual volume in a thin name is one large print, not a market you can exit. We measured the old funnel against the liquid universe on the same tape:

The conditions travel with the numbers. 60 trading days ending 2026-08-14, on minute-path tape, counting a real print inside the 09:55 to 10:15 ET window. Those are study rates measured on that window. The funnel went live 2026-08-24 and has not been forward-validated yet, so this is not a live fill rate and we will not present it as one.

Better fills are not better returns. What changed is that your agent can get in and out at the price it sees. That is worth paying for on its own, and it is the only improvement we measured.

What we do not claim

No selection edge. Two pre-registered studies on 2026-08-22 tested this. One measured the pool against matched random optionable contracts on the same tape: indistinguishable. The other looked for an unusual-activity signal inside the liquid universe: none found. So this is not a list of better contracts, and we will not call it one.

Buying the whole pool under a fixed exit loses money. The whole-pool composite is negative, we measured it, and we publish it on the Track Record. That single fact is why there is no pick endpoint anywhere in this product. A pool average is not a strategy. The winners are in the pool, the excursion data shows they were reachable, and finding them is analysis. That work belongs to your agent and to you.

So what is the value? Two things. Your agent can actually fill the contracts it reasons about. And it gets the historical surface, wins and losses, to reason with.

What is unusual options activity?

Unusual options activity (UOA) is options volume that runs far above normal levels for a particular stock. It can mean institutional traders, such as hedge funds, pension funds, or large trading desks, are building new positions.

The usual indicators are the volume-to-open-interest ratio (fresh activity against positions already on the books), dollar flow (how much capital moved), and directional imbalance (calls against puts).

Here is how GammaRips uses it, plainly. Since 2026-08-24, membership in the pool is liquidity-based. Flow does not decide who gets in. We still measure it and we still publish it, because it is useful context for an agent reading a name. It is a column in the data, not a gate in the funnel.

The private paper cohort

One name a day gets paper-traded as a measurement instrument. It is the operator's private signal and it is not published. There is no public pick, no pick card, and no pick endpoint on the MCP. What is public is the whole pool's outcome record on the Track Record.

The cohort picks its one name with a randomized bracket tournament. Three independent brackets each shuffle the pool into a fresh random order and reduce it in batches of 10 or fewer. A language model advances the top 2 from each batch, round after round, until one name is left. The three bracket winners are then compared: 3 of 3 agree is high confidence, 2 of 3 is medium, 1 of 3 is low. No memory, no rubric, no weights. Every candidate is checked for hindsight data before the model sees it, and any error fails closed.

The tournament runs one deliberately rigid exit bracket, so the method is measured under fixed, unfudgeable rules:

• Entry: 10:00 ET at market.
• Stop: −30% option price.
• Target: +40% option price.
• Hold: the same trading day. Nothing carries overnight.
• Exit: 15:45 ET at market if stop and target both go untouched.
• Conservative tiebreak: if one bar touches both stop and target, the stop wins.

One honest caveat, straight from the Lab: a fixed bracket like this is a measurement instrument, not a strategy. Our own research shows the same contracts produce very different outcomes under different exits. That is exactly why the MCP ships an exit-rule simulator instead of a rule to copy.

The tournament pattern itself is published as a methodology playbook on the MCP (run_your_own_tournament), so a connected agent can run it against your objective, horizon, and risk tolerance instead of the engine's fixed one.

What is agentic trading, and how do you try it?

Agentic trading means using an AI agent (Claude, ChatGPT, or one you build) as your own market analyst instead of following someone else's calls. You don't ask it for a pick. You give it real data. It reasons over the whole surface: today's pool, how similar contracts actually resolved, and how stressed the market is. Then it lays out the picture. The judgment, the sizing, and the trade stay yours.

The catch most people discover the hard way: a chatbot without data will happily improvise. Ask a raw model about a ticker's options flow and you get confident fiction. Its knowledge froze months ago, and no options-flow data exists in any training set. The fix is not a smarter model. It is a connected one.

Here's the on-ramp, cheapest step first. Step 1 (free, no account): browse today's pool and the Track Record yourself. That is the same data your agent would reason over. Step 2 (free, no card): point any MCP-capable agent at our server's anonymous tier and let it taste the pool preview, daily reports, and methodology playbooks. Step 3 (the full data layer): with Agent Access, your agent queries the complete outcome history, opportunity surfaces, and exit-rule simulator, and can even run our bracket-tournament selection pattern against your own objective. Setup for all three takes minutes, and the For Your Agent page walks you through it.