Tafiti Labs

An analytics & research company

Research,
measured before
it is believed.

Tafiti Labs studies how to forecast honestly, how markets really price risk, and how to build AI that checks its own work. Then we test it in products people use.

00.5100.51FORECAST PROBABILITYOBSERVED FREQUENCYCalibratedOverconfidentPerfect calibration00.5100.51FORECAST PROBABILITYOBSERVED FREQUENCYCalibratedOverconfidentPerfect calibration
Fig. 1 · SchematicA forecast is calibrated when events it calls 70% likely happen about 70% of the time. Overconfident forecasts flatten away from the diagonal.

1Tafiti, in Swahili.

What we are, and what we never do

Never the house.

No product of ours takes the other side of a bet, runs a game of chance or awards a prize.

Paid for analysis, never for outcomes.

What we charge is the same whether a user wins or loses. Nothing we sell profits from a loss.

Measured, or it doesn’t ship.

A claim reaches a page or a product only after it has been tested. What’s unproven is labelled unproven.

Privacy is a hard wall.

Our research uses aggregate, anonymised data only. Nothing personal is ever published.

Three questions run through everything we build.

All research

01

Forecasting & calibration

When a system says 70%, does it happen 70% of the time?

We build forecasts that are honest about uncertainty, and evaluation that can’t be fooled: independent samples, one method per claim, and gates that are allowed to say no.

Read more about Forecasting & calibration

02

Market measurement

How well do real markets price risk?

Prices are forecasts too. We measure margins, disagreement between books and internal consistency, starting with African-facing sportsbooks, which the market-efficiency literature has largely skipped.

Read more about Market measurement

03

Trustworthy AI systems

Can a language model be trusted with facts that matter?

We design agents that gather their own evidence, check every extracted claim against it, and refuse rather than guess. Code, not the model, owns anything that touches money.

Read more about Trustworthy AI systems

Five rules we don’t bend.

Every result, in a paper or in a product, has to pass the same rules. They were learned the hard way, and they apply to our own work first.

Box 1 · The discipline

  1. Measured, or it doesn’t exist.

    Every claim names its data, its sample size and its method.

  2. Count what’s independent.

    Statistics are computed on independent events, never inflated by repeated rows of the same one.

  3. One method per claim.

    A result that only appears under a favourable method is a bug report about the method.

  4. Gates are allowed to say no.

    An evaluation that admits nothing isn’t broken. Empty can be the honest answer.

  5. Abstain rather than guess.

    What we can’t resolve, we name. We don’t estimate it.

Nothing published yet. On purpose.

Every paper is posted to arXiv before it appears anywhere else, this page included. The first manuscripts are in preparation.

Publications
TitleAreaVenueStatus
No entries yet · papers appear here once they’re on arXivNo entries yet.Papers appear here once they’re on arXiv.

Research, put to work.

The lab builds products to test its research in the real world. Each carries its own name; Tafiti Labs stands behind all of them.

All work

Panenka

By Tafiti Labs

AI football betting assistant on WhatsApp, Telegram and the web. Priced predictions and booking codes across eleven bookmakers in Nigeria, Ghana, Kenya, Tanzania, Zambia and South Africa.

Live
usepanenka.com

Who’s behind the lab.

Tafiti Labs was founded by Ololade Babatunde and George Thamer Francis.

Meet the people

Write to the lab.

Research questions, partnerships and press.

hello@tafitilabs.com

Tafiti Labs · Lagos, Nigeria