Studio Strategies, Part 1: Build Your First Strategy

Start here: one idea, three parts

You do not need to write code to build a strategy in Studio. You do need to explain what it should do. This series takes you from a plain-language idea to a graph, a backtest, and more demanding experiments. Part 1 builds a small BTC strategy. Part 2 tests it and reads the results. Part 3 adds market making, confirmation signals, weather markets, and perpetuals.

Every threshold and result in these lessons is an educational example, not a recommendation or a measured return. The goal is a strategy you can explain and investigate. Start with the categories your account can access; some feeds and blocks require a particular plan or market permission.

1. Write the idea before adding blocks

“Buy when BTC looks strong” leaves too much unanswered. Try this instead: “In BTC 15-minute prediction markets, buy an outcome only when the model estimates its probability at least five percentage points above the best asking price, with more than two minutes remaining.” This says what to trade, what counts as a signal, and when to stay out.

The hypothesis is that the quoted outcome sometimes underprices the model’s estimate. The model can be wrong, and the apparent gap can disappear before an order arrives. Those are things the experiment must investigate. A winning-looking graph alone cannot establish an advantage.

• Market: BTC, one 15-minute horizon. Keep the first experiment narrow.
• Entry: a favorable price gap, enough time remaining, and a sensible quote range.
• Exit: hold a filled outcome until settlement for this first lesson.
• Invalidation: insufficient history, stale inputs, or an unverified settlement reference means no signal.

2. Understand the numbers you are comparing

A prediction-market outcome priced at 0.60 costs $0.60 per share. Under ordinary binary settlement, a winning share pays $1 and a losing share pays $0. The price suggests a market probability; it does not tell you the true probability. A model estimate of 0.68 against an asking price of 0.60 gives a raw estimated edge of 0.08: eight percentage points, or eight cents of expected value per share before costs, if the model is calibrated.

Now change the ask to 0.75. The same 68% estimate gives a negative raw edge of −0.07. A likely winner can still be expensive. Compare the model with the price you can actually trade, not just a number above 50%.

Keep three units separate. A probability difference of 0.05 is five percentage points. A BTC move of 5 basis points is 0.05%; from $100,000 that is $50. A tick is the venue’s smallest allowed price increment. Five ticks, five basis points, and five percentage points are different settings.

3. Meet the main block families

A feed supplies information. A condition asks a yes-or-no question about it. An action tells the strategy what to do when a path reaches its end. Studio organizes the library so these jobs are easier to find. You do not need a block from every group.

• Signal engines: Random-walk mispricing compares a model with a quote. Price momentum measures a recent move. Relative traded volume asks whether executed trading is unusually active. Weather blocks work with forecasts and event definitions.
• Execution: Bid/ask spread and quote filters check trading conditions. IOC tries to fill immediately at its allowed price and cancels the unfilled remainder. Post-only submits a resting maker order and must not cross the spread.
• Order management & timing: Market time remaining helps avoid late entries. Order age and Order price distance help decide when to cancel a resting quote.
• Randomization: randomized decisions can provide a comparison experiment. Randomness does not create predictive information. Leave it out of the first strategy.

4. Entry, Cancel, and Exit have different jobs

Entry decides whether to open a position when the relevant lifecycle is flat and has no active order. Cancel deals with an outstanding order. Exit deals with a confirmed position. Keep the distinction clear: canceling an unfilled bid does not sell shares you already own.

A condition routes false to the left and true to the right. Put the Do nothing action at any endpoint where nothing should happen; we call this Hold in the recipes below. Hold means “take no action”; it is not a stop-loss, a cancellation, or a sale. Each path from the start of a graph to its final action can contain at most five nodes, including that action.

5. Create a focused universe

Open Studio and select Graph. In More → Select universe, choose Prediction markets → Crypto prices. Select BTC and the 15m market horizon, then Save universe. This chooses the market candidates; it does not automatically build your rules.

Open More → Setup, give the strategy a clear name such as “BTC 15m — five-point value experiment”, and save the setup. The 15m horizon is the contract duration. Later, the model’s 5m volatility history will mean a different thing: how much recent price history it uses to estimate uncertainty.

6. Build the Entry graph

In Entry, add the following conditions in order. Connect each true branch to the next block and each false branch to Hold. Click a block to edit its settings. The final true branch ends in IOC. Three conditions plus the action make a four-node trading path.

Random-walk mispricing uses the evaluated outcome and the contract’s settlement reference. You do not need to invent a separate BTC price target or treat every token as UP. If the exact settlement source or required history is unavailable, the model cannot supply a valid signal.

• Market time remaining: greater than 120 seconds. At 120 seconds or less, Hold.
• Random-walk mispricing: Volatility history = 5m; Polymarket comparison price = Best ask; comparison ≥; Minimum favorable edge = 0.05.
• Current outcome price: choose the ask field and a range from 0.10 to 0.90. This keeps extreme-priced outcomes out of this particular experiment; it is not proof of liquidity.
• IOC: price offset = 0 ticks. Review the sizing shown for the action and the strategy’s minimum capital. The request can fill partially or not at all.

7. Finish Cancel and Exit deliberately

Set Cancel to Hold because this basic IOC example leaves no resting order after its immediate attempt. Set Exit to Hold because the experiment keeps a filled outcome until settlement. This is an explicit research choice: if the outcome loses, the purchase cost of those shares can be lost. It does not give you a protective exit.

Walk through an invented moment: 300 seconds remain, the model estimates 0.68, and the ask is 0.60. All three entry checks pass. If the estimate falls to 0.63, the edge becomes 0.03 and the graph holds. If only 90 seconds remain, the first condition stops the entry even if the price gap looks attractive.

8. Check size and save a draft

An entry rule is incomplete without a size. Studio uses subscription-capital sizing and venue minimums. For Polymarket, the round-allocation setting participates in a share budget that also depends on the selected assets and market slots; it is not simply the same percentage of cash spent on every trade. Review the actual configuration before interpreting exposure.

For arithmetic only, 10 shares at $0.60 cost $6 before fees. That position pays $10 if it wins and $0 if it loses. This illustration is not a promise that entering a particular allocation percentage will create a 10-share order. Keep your experiment’s size and universe consistent when comparing backtests, and save a draft before changing the idea.

A useful Quant AI prompt, then your next step

You can also ask Quant AI: “Build a BTC 15m prediction-market research strategy. Entry: time remaining greater than 120 seconds, random-walk favorable edge at least 0.05 against the best ask using 5m volatility history, and ask between 0.10 and 0.90. Use IOC with zero tick offset. Hold in Cancel and Exit; the example holds to settlement. Use supported subscription-capital sizing and explain the minimum capital.”

Inspect the generated Graph against the checklist above. AI is a drafting aid; read every branch, action, and size. You should now be able to explain a trade, a skipped trade, and what happens after a fill. In Part 2, you will run the 24-hour backtest and find out what its report can—and cannot—tell you.