Most of the stack is free, and a spreadsheet does more than most paid products. What each category of tool is for, and where paid data actually starts to matter.
- Auteur
- EdgeMarket
- Publié
- Temps de lecture
- 6 min de lecture
Search for prediction-market tooling and you will find listicles that are really pricing pages with a paragraph in front. This is not that. The honest answer is that a prediction-market stack has five jobs, that three of them are done adequately by things that cost nothing, and that the two remaining ones are where data actually has to be bought.
What follows is the categories, what each is for, and how to tell whether the one you are looking at does the job.
What the venue already gives you
Polymarket's own interface is a good execution screen. It shows the current price, the book, the market's rules, and your open positions. For placing a trade and reading its terms, it is sufficient, and any tool that presents itself as a replacement for it is solving a problem you do not have.
Where it stops is history and comparison. The interface answers what is this market doing now. It does not answer what did prices in this shape do afterwards, what has this address done before, or what did I pay across my last hundred trades. Every tool worth adding exists to answer one of those three questions, and the ones that do not are decoration.
The five jobs
1. Converting between price, probability and payoff
A contract priced at 0.62 is a 62% implied probability, a payoff of 1 per contract, and — if you think in sportsbook terms — decimal odds of 1 ÷ 0.62, or 1.6129. Traders lose track of the equivalence constantly, and the mistakes it produces are arithmetic mistakes rather than judgement ones.
This job is trivially automatable and should never be paid for. Our odds converter is open with no account, as is the rest of the calculator set. A spreadsheet with three columns does the same thing. The point is to stop doing it in your head.
2. Sizing
The second job is deciding how much. For a binary contract bought at price c where you believe the probability is p, the Kelly fraction is (p - c) / (1 - c) — which shrinks fast as c rises, because at 0.90 you are risking ninety cents to make ten. A position sizer turns that into a number; discipline turns the number into a fraction of it.
Also free, also a spreadsheet, and skipped by almost everyone. Sizing is the mechanism that converts a real edge into a drawdown you cannot sit through, and it is covered in more detail in why most prediction-market traders lose money.
3. Your own trade record
The third job is the one nobody wants and the only one that produces feedback. Log every trade with the price you paid, the probability you believed at the time, and the eventual resolution. Then group by the price you paid and compare your realised frequency against the average price of the bucket.
That is a calibration table, and it is the only measurement of your own judgement that is not filtered by what you happen to remember. It is a spreadsheet. It is also strictly more valuable than every other tool in this article combined, because it is the only one measuring you.
The method is the same one we apply to our own emissions on the public register, including the bands where the measurement is unflattering.
4. Wallet-level history
Here is where free stops being enough. Every Polymarket trade settles on Polygon, so a block explorer will show you an address's transactions. What it will not do is reconstruct, per address, which markets those transactions belong to, at what prices, and how the markets resolved — and without that reconstruction, an address's history is a list of hashes.
This is a genuine data problem: joining trades to markets to resolutions across a corpus that runs to more than 1.2 million markets in our own store — around 130,000 of them still active — and more than 2.1 million classified trades. It is doable with the public API and enough patience. It is not doable in an afternoon.
It is also the category where the most nonsense is sold, because a table of addresses sorted by a score looks authoritative regardless of whether the score means anything. What "smart money" means, and what it does not goes through what has to be measured before a ranking is worth reading — and what we found when we audited the obvious ranking on our own data.
5. Alerting
The last job is being told when something happens while you are elsewhere. This is the category most likely to be oversold and most likely to be misused, because an alert stream that fires constantly is indistinguishable from no alerts at all — you stop reading it in a week.
Our own alerting has raised 335,737 alerts since 2 February 2026, measured on 17 August 2026. That figure counts thresholds crossed, not messages delivered, and the distinction is the whole design problem: the useful product is the filter, not the firehose.
What a stack looks like at each stage
| Stage | What you need | What it costs |
|---|---|---|
| Reading markets | The venue interface, the resolution rules, a converter | Nothing |
| Trading your own view | The above, plus a sizer and a trade log | Nothing |
| Judging other people's positions | Reconstructed wallet history with a stated method | Data |
| Reacting away from the screen | Filtered alerts on defined conditions | Data |
| Running your own analysis | Raw series you can join yourself | Data, and an API |
The first two rows are the ones that change outcomes. If you have not built them, buying the last three is buying speed on a road you have not chosen yet.
Where we sit
We cover two terrains — Polymarket and crypto market structure — and we sell measurement, not execution. Nothing here places a trade, holds a balance, or tells anyone what to do.
The public register is open to everyone with no account: the calibration table, the coverage series — 47.66% of the signals we have emitted were never verifiable, and in May 2026 that share reached 73.76% (measured 17 August 2026) — and the bands where our own signal is worse than the market's price. So is the calculator set and the live tape on the home page.
Beyond that, the split is by job rather than by volume. Operator, at 39 USD a month or 390 a year, opens the wallet ranking, the depth of an operator's history, and alert channels. Principal, at 79 USD a month or 790 a year, adds the things that only matter if you are computing something yourself: the level-2 book wall, the REST API and outbound webhooks. No plan sells a file: nothing in the product hands you a CSV today, and the pricing grid does not claim one. Observer stays free and stays useful. The full grid is on pricing.
The build order
- Read the resolution rules of every market you enter. This is free and it is the highest-return habit in the entire list.
- Convert prices to probabilities explicitly, every time, until it is automatic.
- Size from a formula and then take a fraction of what it says.
- Keep the log. Build your own calibration table after fifty trades.
- Only then add someone else's data — and when you do, read the method before the numbers.
Step five is where most people start, which is why the tooling market looks the way it does. A dashboard is a pleasant thing to buy at the moment you would rather not look at your own record.
Related reading: what a prediction-market order book does that a crypto book does not, and what goes wrong when you copy a wallet — the two places where a tool most often gets used to confirm something instead of measure it. The mechanics of the instrument itself are in how prediction markets work.