Turn a hunch into
a tested model.
Just ask your AI.

BetFlux gives your agent the sports data and know-how to turn a simple prompt into a tested model.

nfl-totals · codex

›_ OpenAI Codex

directory: ~/nfl-totals

skill: betflux ✓ ready

It looks like NFL total overs are dominating weeks 1–2. Show me the percent of overs that won, broken down by home/away.

● I’ll pull graded NFL totals for weeks 1–2 from BetFlux and split the overs by home and away.

● Read the BetFlux Agent Skill

└ Ran betflux agent-guide

Loaded dataset schemas, query patterns, and grading semantics.

● Find weeks 1–2 games

└ Ran betflux games --league NFL \
  --date-from 2026-09-10 --date-to 2026-09-21

32 games · Week 1: Sep 10–14 · Week 2: Sep 17–21

● Load graded over selections

└ Ran betflux get market-results --league NFL \
  --date-from 2026-09-10 --date-to 2026-09-21 \
  --side OVER --format csv > overs.csv

Kept TOTAL and TEAM_TOTAL overs. Pushes and indeterminate grades excluded.

● Split overs by home and away

└ Ran python overs_by_side.py --main-lines

Game totals have no home/away side, so team-total overs are split by HOME / AWAY. Game totals kept for reference.

● Weeks 1–2 overs

Illustrative NFL over results for weeks 1 and 2, main lines only.
Market	Won–Lost	Over %
Game totals	20–12	62.5%
Home team totals	19–13	59.4%
Away team totals	14–17	45.2%

Game-total overs are the clearest edge at 62.5%. Home team totals lean over too; away team totals don’t. Thirty-two games is a small sample, so keep checking it.

Saved overs_by_side.csv + report.md

Game-total overs look like the edge. Build a projection for week 3 and show me the highest-EV overs at current odds.

● I’ll pull week 3’s current totals from BetFlux, project each over from weeks 1–2, and rank them by expected value.

● Find week 3 games

└ Ran betflux games --league NFL --status SCHEDULED \
  --date-from 2026-09-24 --date-to 2026-09-28 --format csv > week3.csv

16 games · Sep 24–28

● Pull current odds

└ Ran betflux get sportsbook-lines --league NFL \
  --date-from 2026-09-24 --date-to 2026-09-28 \
  --market-type TOTAL --side OVER --format csv > week3_odds.csv

Latest game-total over prices loaded across 4 books.

● Project overs and price EV

└ Ran python project_totals.py --train overs.csv --shrink

Shrank the early over rate toward the long-run base rate, then priced EV at each game’s best available over.

● Highest-EV week 3 overs

Illustrative week 3 game-total overs ranked by expected value at the best available price.
Game	Over	Best	Model	EV
KC @ BUF	47.5	+100	54.8%	+9.6%
DET @ GB	49.5	−105	55.9%	+9.1%
LAR @ SF	44.5	+105	53.1%	+8.9%
CIN @ BAL	51.5	−110	55.0%	+5.0%

Four overs price above zero EV at the best available number. The model leans on a 32-game trend, so treat this as a watchlist and re-check the lines before kickoff.

Saved week3_overs.csv + report.md

Codex reads the BetFlux Agent Skill, finds the NFL games from weeks 1 and 2, downloads graded total overs with the BetFlux CLI, and splits the over win rate by home and away team totals.

● Weeks 1–2 overs

Illustrative NFL over results for weeks 1 and 2, main lines only.
Market	Won–Lost	Over %
Game totals	20–12	62.5%
Home team totals	19–13	59.4%
Away team totals	14–17	45.2%

Game-total overs are the clearest edge at 62.5%. Home team totals lean over too; away team totals don’t. Thirty-two games is a small sample, so keep checking it.

Saved overs_by_side.csv + report.md

Asked for a week 3 projection, Codex pulls each scheduled game’s current odds from sportsbook-lines, projects every game-total over from the weeks 1–2 trend, and ranks them by expected value.

● Highest-EV week 3 overs

Illustrative week 3 game-total overs ranked by expected value at the best available price.
Game	Over	Best	Model	EV
KC @ BUF	47.5	+100	54.8%	+9.6%
DET @ GB	49.5	−105	55.9%	+9.1%
LAR @ SF	44.5	+105	53.1%	+8.9%
CIN @ BAL	51.5	−110	55.0%	+5.0%

Four overs price above zero EV at the best available number. The model leans on a 32-game trend, so treat this as a watchlist and re-check the lines before kickoff.

Saved week3_overs.csv + report.md

Analysis completeIllustrative results, not a betting recommendation

From one prompt to a working model

What will you build?

01 / 12

You bring the idea.
We bring the model-ready data.

BetFlux gives your agent the data and the know-how to build models, test strategies, and turn your questions into analysis.

Your agentIdeas → instructions

+ Agent SkillBetFlux CLIThe connection + the contextbetflux get …

Our datasetsNormalized. Connected.
Queryable.

01Give it a mission.

A sportsbook model, a fantasy ranking, a market hypothesis. Say what you want to explore.

02Teach it the market.

Our Agent Skill teaches your AI the datasets. Our CLI puts the data within reach.

03Build. Test. Refine.

Your agent writes the code, runs the analysis, and explains the results. You steer the next move.

The hard math is already done.

Datasets your agent can query directly.

Sportsbook odds

Moneyline

Price: −135
Implied: 57.4%
De-vigged: 55.3%

−100−125−150

−110 → −135

OpenLatest

Illustrative · de-vigged using opposite side +115

Every move matters.

Change-by-change odds history with de-vigged probability fields. One consistent schema across books and markets.

Line movementFair probabilities

Closing lines

Moneyline

OpeningClosing

−105

51.2% implied

−135

57.4% implied

Same selection, same sportsbook · illustrative example

Know where it closed.

Opening and closing prices, line-movement summaries, and book-only closing line value.

CLVMarket benchmarks

Market results

Settled selections

09/18WONWONLOSTWONLOSTWON

09/19LOSTWONWONLOSTWONLOST

09/20WONLOSTWONWONLOSTWON

Graded by settlement date · illustrative example

Give your model answers.

Closing lines joined to graded outcomes and settled values.

Outcome labelsBacktesting

Game-state timeline

Q2 13:451st and 10

GB 14 – DET 0

Moneyline

GB −400

Game state + sportsbook odds · illustrative example

Put the play in context.

Score, clock, and game-state observations on a shared timeline. Connect what happened to what moved.

Game contextTime-aligned features

Coverage varies by dataset, league, and provider. See availability and history →

Modeling used to mean scrapers.
Now it’s a conversation.

You call the shots. The agent does the work.

Building a model used to mean

With BetFlux

Used to mean

Coding experience and model-building skills

With BetFlux

You direct, your agent builds

Used to mean

Finding sources, scraping, writing your own de-vigging and results handling

With BetFlux

Every dataset already there, reachable by the agent

Used to mean

Hours of tweaking and testing

With BetFlux

Agent loops do the tuning

An individual at a three-monitor analysis workstation in a suburban home, with basketball playing on a TV in the background.

The sharps had the tools.
Now you do.

What stood between a good idea and a working model was weeks of scraping, de-vigging and backtest scaffolding. Now it’s a prompt.

No coding background. No data science degree. Just a question worth asking, and an agent ready to analyze.

1Ask in plain English.
2Test your thinking.
3Keep what holds up.

See the data behind
your next breakthrough.

Explore the datasets

Across the leagues.

NBA

NFL

MLB

NHL

NCAA Men’s Basketball

Across the sportsbooks.

Into prediction markets.

Bring your agent.

Your favorite AI with a new set of skills.
Connect once, then keep asking better questions.

OpenAI

Codex

Anthropic

Claude Code

Cursor

Agent Skills

Your agent, too.

Works with agents that support skills and terminal access.

Stop guessing.
Start asking more.

New leagues, new books, and CLI releases.
One email when there’s something to say.

Already have an invite? You’re up. →

Turn a hunch intoa tested model.Just ask your AI.

What will you build?

You bring the idea.We bring the model-ready data.

01Give it a mission.

02Teach it the market.

03Build. Test. Refine.

The hard math is already done.

Every move matters.

Know where it closed.

Give your model answers.

Put the play in context.

Modeling used to mean scrapers.scrapers.busy-work.data science.coding.Now it’s a conversation.

The sharps had the tools.Now you do.

See the data behindyour next breakthrough.

Across the leagues.

Across the sportsbooks.

Into prediction markets.

Bring your agent.

Codex

Claude Code

Cursor

Your agent, too.

Stop guessing.Start asking more.

Turn a hunch into
a tested model.
Just ask your AI.

You bring the idea.
We bring the model-ready data.

Modeling used to mean scrapers.
Now it’s a conversation.

The sharps had the tools.
Now you do.

See the data behind
your next breakthrough.

Stop guessing.
Start asking more.