How to Use Our Downloadable Predictions

Share this page

You're looking at a match prediction and you want the actual numbers — not a screenshot, not a summary, but the raw probability distributions behind the charts. On PredixSport, you can download that data as JSON or CSV from any match analysis page, across NBA, Football, and Tennis.

This guide covers what's in the files and four practical ways to put them to work.

Where to Find the Download

Open any match analysis page — NBA, Football, or Tennis — and you'll see this bar:

Download Probability Data

Full distributions as structured data (JSON or CSV)

JSON CSV

Sign in with Google (free), click JSON or CSV, done. 200 downloads per month, no cost.

What's Inside the Data

Each file contains every prediction we generated for that match. Here's a real example — Alcaraz vs Sinner, Monte Carlo Masters 2026:

JSON
{
  "sport": "tennis",
  "match": {
    "player_one": "Carlos Alcaraz",
    "player_two": "Jannik Sinner",
    "player_one_rank": 1,
    "player_two_rank": 2,
    "tournament": "monte_carlo",
    "surface": "clay",
    "tournament_type": "masters_1000"
  },
  "predictions": {
    "player_one_win_probability": 0.4648,
    "player_two_win_probability": 0.5352,
    "predicted_winner": "Jannik Sinner"
  },
  "distributions": {
    "total_games": {
      "expected_value": 23.1,
      "mode": 23,
      "pdf": { 12: 0.44, 13: 0.79, ... probability for each outcome },
      "cdf": { 12: 0.86, 13: 1.65, ... cumulative under probability },
      "cdf_over": { 12: 99.14, 13: 98.35, ... probability of going over }
    },
    "games_spread_p1_minus_p2": { ... same structure },
    "aces": { ... same structure },
    "double_faults": { ... same structure }
  },
  "source": "predixsport.com",
  "exported_at": "2026-04-13T17:09:23Z"
}

Three layers per distribution: PDF (probability of each exact outcome), CDF (cumulative probability of going under a value), and CDF-over (probability of going over). Same data behind our interactive charts — raw model output, nothing stripped.

What's available per sport:
  • NBA — Win probability, total points, spread, plus per-player distributions for points, rebounds, assists, and 3PM
  • Football — Match result probabilities, total shots, corners
  • Tennis — Total games, games spread, aces, double faults

Four Ways to Use This Data

1 Fantasy Sports Analysis with AI Assistants

If you play fantasy basketball, you already research matchups, recent form, and projected stats before setting your lineup. Our data adds something you probably don't have: full probability distributions for every starter — not just expected values, but the shape of the uncertainty around them.

The workflow takes about two minutes:

Step 1 — Download

Open an NBA match analysis page and grab the JSON or CSV. You get per-player projections for points, rebounds, assists, and 3PM — each with a complete distribution.

Step 2 — Upload to ChatGPT, Claude, Gemini, or Perplexity

Attach the file to a conversation. Works on desktop and mobile — just drag and drop or tap the attachment button.

Step 3 — Ask what you need

The AI reads the full distribution and can answer specific questions. You can also paste in extra context — injury news, box scores from last night, matchup notes from other sites — and it'll factor everything together.

Example prompt after uploading

"I uploaded today's NBA prediction data from PredixSport. For the Lakers vs Celtics game: which player has the widest distribution for points? Who has the highest probability of exceeding 25 points? Cross-reference with the latest injury reports."

This works for daily fantasy lineups, season-long roster decisions, or just getting a sharper read on tonight's games before tip-off.

2 Prediction Market Research

Prediction markets like Polymarket publish crowd-sourced probabilities for real-world events, including sports. The interesting question isn't what the market says — it's whether the market is right.

Our data gives you a second opinion. An independent model-based estimate you can compare against what the market is pricing. When the gap is large, it's worth understanding why.

How to use it

Download the match data, upload it to ChatGPT or Claude (with web browsing enabled), and ask it to pull the current market prices for the same event. The AI can check Polymarket in real time and tell you where our model disagrees with the crowd.

Example prompt

"Here's PredixSport's prediction data for tonight's NBA games. Our model gives the Celtics a 71.2% win probability. Check the current Polymarket prices for this game. How big is the gap? What might explain it — injuries, public sentiment, something else?"

You get a structured comparison based on full probability distributions, not just a single number. And because the AI can browse the web, it can pull in context you might have missed.

3 Automated Agents and Daily Pipelines

This one is for people who build things. If you've used OpenClaw, OpenAI Agents, or LangChain, you already know the pattern: give an agent a data source, a set of tools, and a goal, and let it run on a schedule.

Our prediction data fits right into that. You can build an agent that fetches our latest predictions every day, reads the probability distributions, checks external sources (news, market prices, historical results), and sends you a summary of whatever it found interesting. No manual work after the initial setup.

Example: daily briefing agent

Your agent runs every morning. It downloads today's predictions, compares the distributions against yesterday's (did the model shift significantly on any game?), checks for late injury news that might not be reflected yet, and flags games where model confidence is unusually high or low. Then it sends you a Slack message or email with the three most interesting things it found.

Why OpenClaw works well here

OpenClaw lets you define agents that can browse the web, call APIs, and run on a cron schedule — exactly what you need for a daily sports analysis pipeline. Point it at our JSON files as a data source, give it access to a few external sites for context, and you have an autonomous research assistant that runs while you sleep.

Pipeline structure
# Runs daily — adapt to OpenClaw, OpenAI Agents, LangChain, etc.

1. Fetch     → Download today's match predictions (JSON)
2. Parse     → Read probability distributions (PDF, CDF, CDF-over)
3. Compare   → Check against external sources (markets, news, stats)
4. Filter    → Keep games with high confidence or unusual patterns
5. Notify    → Send daily briefing via Slack, email, or Telegram

The JSON schema is consistent across all sports and all matches — same field names, same structure, every time. That makes it straightforward to parse without writing custom logic for each sport or match type.

4 Feature Input for Your Own ML Models

If you build your own prediction models, you can use our probability distributions as input features. Instead of replicating our entire training pipeline (hundreds of features, tens of thousands of matches), you take our model's output and feed it into yours. In ML terms, this is model stacking — and it's a standard technique for a reason.

What you get as features

Expected values, distribution modes, the full PDF shape (which you can convert to standard deviation, skewness, or any summary stat you want), and CDF values at specific thresholds. That's a lot of signal from a single download.

The CSV loads directly with pandas.read_csv() — one row per outcome per distribution, with columns for probability, cumulative under, and cumulative over. No cleaning needed.

Python
import pandas as pd

# Load — ready to use as-is
df = pd.read_csv("predixsport-lakers_vs_celtics.csv")

# Filter to total points distribution
total_pts = df[df["distribution"] == "total_points"]

# Extract what you need
expected = total_pts["expected_value"].iloc[0]
pdf = total_pts[["outcome", "probability"]]

# Use as features in your model
features["px_total_pts_ev"] = expected
features["px_total_pts_std"] = compute_std(pdf)

JSON works too, via Python's json module. scikit-learn, XGBoost, PyTorch, whatever you're using — the data fits.

Get Started

The download bar is on every match analysis page — NBA, Football, and Tennis. Sign in with Google, pick JSON or CSV, and the file is yours.

Upload it to an AI assistant for a quick read. Plug it into a daily agent pipeline. Compare it against prediction markets. Feed it into your own models. The data is structured, consistent across sports, and doesn't need cleaning.

Start Downloading

Pick a sport, open a match analysis page, sign in with Google, download. Free account, 200 files per month.

Share this page

Get daily predictions straight to your phone

Join Our Telegram Channel