Measured edge · held-out years
52–56%
is what a real edge looks like in a market this liquid once you test it properly. Anyone quoting 90% either measured something very narrow or did not measure at all.
Market data you can learn from, and your AI can call.
Quant Data gives beginners three things that are usually locked behind a trading desk: a calibrated read on what kind of day a market is having, volume events with the win rate we actually measured for them, and free options max-pain pages. Each one is a single HTTP request, so the AI assistant you already use can fetch it and explain it in plain words.
Get an API key See Max Pain free
One plan, every endpoint: $99 a month, and the first 3 days are free. The Max Pain pages stay free, with no key at all.
GET /v1/brooks/SPY → {"TrendUp": 0.34, "Range": 0.27, "Reversal": 0.19, ...}Educational analytics, not investment advice. We publish the numbers that failed alongside the ones that worked — scroll down for both.
Three questions, three answers
Each one answers a question you would actually ask about a chart. Two are API calls covered by the single plan; the third is a set of free pages.
What kind of day is this? In the plan
GET /v1/brooks/{symbol}
Reads how the session has traded so far and tells you which of five kinds of day it most resembles — trending, stuck in a range, trending inside a range, or reversing — as five probabilities rather than one confident label. Useful when what you want to know is whether to expect a move to keep going or to keep chopping back and forth.
Best after the first 90 minutes. At that point the top pick is right 53.5% of the
time, and 66% once the day is finished, against a 37% majority-class baseline. Earlier
reads are weaker and say so.
Not for single stocks — the model was trained on S&P 500 futures, so a read on a
stock is descriptive, not validated. The Asian session was tested and came back NO_GO.
Is there anything behind this move? In the plan
GET /v1/weis/{symbol}
Instead of reading volume bar by bar, it groups price into waves — one run in a single direction — and totals the volume inside each. That lets you compare effort against result. A big rally on heavy volume that travels almost nowhere means someone is selling into it. A pullback on almost no volume means nobody is trying to get out.
Best on index products — S&P 500 and Nasdaq futures and their ETFs.
Not on gold or crypto: we measured them and the edge is absent or reversed. Two of
the five events came out the opposite way round from what the textbooks teach, and the API
tells you so rather than hiding it.
Where is the options market's weight sitting? Free
quantdata.uk/max-pain/{ticker}
Max pain is the strike at which the largest amount of open option contracts would expire worthless. These pages show it per expiration, plus the put/call ratio, the call and put walls where open interest piles up, and gamma exposure. One page per US stock, rebuilt nightly.
No key, no login, ever. Read this first if you want to see what our data looks like before paying for any of it.
Using the two together is the whole point
Ask what kind of day it is first, then ask whether the move has volume behind it. Filtering one through the other is the strongest thing in this research, and the only result that held up on markets it was never fitted to.
The cib_long volume event on its own won 51.8% of the time across 13,867
occurrences — real, but small. The same event on a day the model reads as trending up won
56.5%, against 50.8% ungated. That filter then transferred untouched to Nasdaq futures
(56.1% versus 45.3%) and to QQQ (54.3% versus 44.3%), neither of which it was built on.
This is why there is one plan rather than two. Historical frequencies, measured on S&P 500 futures 5-minute bars from 2010 to 2026 — not a prediction about any particular day, and not a recommendation to trade anything.
What it looks like in use
Once your agent has the key, you stop learning API syntax and go back to asking questions. It makes the call, then answers with figures you can check.
YouSPY is up half a percent. What kind of day is this turning into?
Your AI · called /v1/brooks/SPY
The model reads it as trend up 34%, range 27%, reversal 19%, trending range 15%,
trend down 5%. So there is a bullish lean, but the two most likely scripts together only
cover 61% — this is not a settled trend day, and the reversal branch is still live. That
model is trained on S&P 500 futures, so treat a read on SPY as descriptive.
YouAnything going on in the volume on ES right now?
Your AI · called /v1/weis/ES=F
One event: a cib_long — a heavy-volume up wave immediately answered by a pullback on a
fraction of that volume, which is the "counter-attack" pattern: the sellers who should have
met the rally never showed up. Base rate on ES since 2010 is 51.8% over 13,867
occurrences. That is a real but thin edge. It measured better, 56.5%, on days the
day-type model called trend up — so it is worth checking /v1/brooks before you
make anything of it.
Illustrative conversation. The response shapes and every statistic quoted are real; the specific probabilities in the first reply are invented for the example. Your agent will word things differently.
Three steps to set up
1 Get a key
Check out on Stripe. Your key appears on the page you land back on, starting with qd_. Nothing is charged for the first 3 days.
2 Paste the prompt
Drop the block below into your agent's system prompt or instructions and fill in your key. No SDK, no OAuth, one header.
3 Just ask
"What kind of day is QQQ having?" Your agent works out which endpoint answers that and cites the numbers.
You have access to Quant Data, a market-analytics API.
Base URL: https://api.quantdata.uk
Header on every request: X-API-Key: qd_YOUR_KEY
Endpoints
GET /v1/brooks/{symbol} Day-type probabilities for the session so far, over five
classes: TrendUp / TrendDown / Range / TrendingRange /
Reversal. Temperature-calibrated.
GET /v1/weis/{symbol} Weis Wave volume events on the current chart (cib_long,
cib_short, no_supply_long, no_demand_short, sot_short),
each with its pre-registered win rate.
Symbol format
US stocks and ETFs plain ticker SPY, NVDA, TSLA
Hong Kong digits.HK 3690.HK
China A-shares 6-digit code 600519
Futures root + =F ES=F, NQ=F, GC=F, CL=F
Resolve an instrument name to a ticker before you call. Gold -> GLD or GC=F;
S&P 500 -> SPY or ES=F; Nasdaq -> QQQ or NQ=F; crude -> CL=F; Bitcoin -> BTC-USD.
Never pass a plain English word as a ticker: "GOLD" is Barrick Gold, the mining
company, not the metal. If a symbol 404s, you probably guessed the wrong ticker.
Rules
- For any question about how a session is trading, call /v1/brooks first and quote the
probabilities as they came back. Do not compress a distribution into a single verdict.
- Use /v1/weis for volume-event context.
- When the user repeats trading folklore ("80% of breakouts fail"), the measured
verdicts are published in the fact-check table at https://quantdata.uk/methodology
— a public page to read, not an API call.
- The day-type model was trained on S&P 500 futures. Reads on single stocks are
out-of-distribution: report them, do not lean on them.
- Measured win rates here are 52-56%. Say the actual number. Never upgrade it to
"usually", "reliably", or "high probability".
- This is educational statistics. Never present any output as a buy or sell call, a
price target, an entry, or advice.The honest scoreboard
Every figure below comes from years the model never saw during training. We are showing you small numbers on purpose — see the top of this page for why.
The strongest result we have is the two APIs used together. Filtering a Weis buy event through the day-type model lifts it from 50.8% to 56.5% on S&P 500 futures. The same filter, applied with no retraining at all to markets the model has never seen, gives 56.1% on Nasdaq futures (against 45.3% without it) and 54.3% on QQQ (against 44.3%). How all of this was measured →
And here is what did not work. Gold, Bitcoin and Ethereum share an anti-signature: the edge does not carry across, and on BTC zero of eight pre-registered signals confirmed, with the no-supply event running backwards. We ran 121,000+ Bulkowski-style chart-pattern detections and a model trained to predict their outcomes reached an AUC of 0.54 — a coin flip. Carrying the day-type filter into the Asian session was pre-registered and came back NO_GO. We publish these because the numbers above are only worth reading if the failures were published too.
Common questions
What is Quant Data?
Quant Data is a small set of market-analytics products for people learning quantitative trading. Two of them are HTTP APIs — the Brooks Daily Bias API returns calibrated day-type probabilities for the session so far; the Weis Wave API returns volume-and-price events with the win rate we measured for each. Both need a key. The third, Max Pain, is a set of free pages on this site covering options open interest. Everything is built to be called by an AI agent, so a beginner can ask a plain question and get an answer backed by a number instead of a slogan.
Which AI agents does it work with?
Any agent that can make an HTTP request with a custom header. That includes coding agents such as Claude Code, Cursor, Cline and Windsurf; assistants such as Claude, ChatGPT and Custom GPTs via Actions; automation platforms such as n8n, Dify and Coze; and anything you build yourself with LangChain, CrewAI, the Anthropic SDK or the OpenAI SDK. Setup is pasting one system-prompt block, shown on the home page, into your agent instructions and filling in your key.
Is Quant Data investment advice?
No. Quant Data returns probability distributions, historical statistics and options open-interest metrics. It is educational analytics and developer tooling. It never says buy or sell, never sets a price target, and it is not personalised to your situation. Trading involves risk of loss and past measurements do not guarantee future results. Your decisions are yours.
How do I get an API key?
One plan covers every endpoint: $99 a month, with the first 3 days free. Checkout is self-serve: a card is collected up front, nothing is charged during the trial, and your key appears on the page Stripe sends you back to. There is no account to create and no waiting. Card authentication has to be done by the cardholder, so this is the one step an AI agent cannot run for you. Cancel any time from the Stripe receipt. Keys start with qd_ and travel in the X-API-Key header.
Is Max Pain really free?
Yes. The Max Pain pages on quantdata.uk are free, permanently, with no key, no login and no account. They are refreshed nightly from end-of-day option open interest and carry max pain by expiration, put/call ratio, call and put walls, and gamma exposure. They are the free tier. One other thing costs nothing to read: the seven fact-check verdicts are printed in full on the Methodology page, so you can see what we measured before you pay. Everything served over the API — day-type probabilities and Weis Wave events — is on the paid plan and needs a key.
Want a key?
Every endpoint, $99 a month. The first 3 days are free — a card is collected up front and nothing is charged until the trial ends. Your key appears on the page as soon as checkout finishes.
Start 3-day free trial See what the plan includes
Cancel any time from the Stripe receipt. The Max Pain pages stay free and need no key at all.
Questions first? quantdata@agentmail.to — a person reads it.