We Switched On Three Betfair Experiments. Three Days Later, We Had 1,161 Live Observations.

For a long time, most of my Betfair research started with the same question: can I find something that makes money?

The Trade Carefully Research Framework changed that slightly. The better question is often: what happens if I test this idea consistently for long enough to actually learn something?

That distinction matters.

My original Master Collector accumulated 49,172 actual live bets. It lost around 1,350 units. That sounds like a spectacular failure until you understand what happened next. Instead of throwing the data away, I analysed it, segmented it, rejected parts of it and refined what remained. That research eventually contributed to Dog Pound, a much narrower implementation which has recorded more than 2,000 live bets and over 466 units of profit.

The losing data turned out to be useful.

So I've started doing more of it.

Three Questions, Three Collectors

On 25 September 2026, I switched on three new automated collectors using BF Bot Manager.

• The first is H01 – Global Second Favourite Horse Collector. It backs the second favourite shortly before the start of sufficiently liquid horse races around the world.

• The second is F01 – Global Match Favourite Football Collector. It backs the favourite in the Betfair Match Odds market shortly before kickoff.

• The third is T01 – Global Match Favourite Tennis Collector. It does essentially the same thing in tennis: identify the current match favourite shortly before the scheduled start and record what happens.

  
  

These aren't three new betting systems I'm trying to sell.

They're three questions.

The selections are backed automatically with tiny flat stakes. There is no staking progression, no chasing losses and no changing the rules because last Tuesday wasn't particularly enjoyable.

The objective is to accumulate prospective live-market evidence under clearly defined conditions.

1,161 Observations in Three Days

I exported the data again on 28 September.

The collectors had already generated 1,161 live bet records:

• Horse racing: 678

• Football: 275

• Tennis: 208

Of those, 1,144 had already settled.

The horse-racing collector alone had captured markets across Australia, the United States, Great Britain, France, New Zealand, Ireland and South Africa. The football collector was already picking up matches across numerous countries and competitions.

That's when the potential of this really hit me.

I'm not sitting at a computer choosing which results look interesting enough to record. The collectors simply keep collecting.

Tomorrow's observation gets recorded whether I like it or not.

What About the Profit?

At the point of this export, the three collectors happened to be showing a small combined positive P/L.

I'm deliberately not making a story out of it. In fact, I'm almost certain these collectors will lose money over time.

These are deliberately broad research hypotheses. There has been no retrospective optimisation to find the countries, competitions, price ranges or market conditions that happened to work historically. We defined a question, switched the collector on and started accumulating prospective evidence.

If they happen to make money, interesting.

If they lose money, also interesting.

Three days tells us almost nothing either way. If they were £10 ahead, that wouldn't prove anything. If they were £10 behind, that wouldn't prove anything either.

The temptation in betting research is to start interpreting results far too early. A strategy wins for a week and suddenly we've discovered an edge. It loses for a week and we start adding filters.

Do that repeatedly and eventually you're no longer researching the original idea. You're reacting to noise.

These collectors are intended to run for thousands of observations. The objective isn't to make money from the collectors. It's to create enough evidence to learn from them.

The Dataset Is the Asset

This is perhaps the biggest change in how I'm thinking about Trade Carefully.

The valuable output doesn't necessarily have to be a profitable betting strategy.

Imagine having 10,000 prospectively collected observations of the second favourite in horse races, recorded under the same rules. Or 10,000 football favourites. Or thousands of tennis matches.

We can then start asking better questions.

Does performance change with price? Country? Competition? Market liquidity? How do different cohorts behave? Are apparent patterns persistent or simply variance?

Some experiments may eventually reveal something interesting.

Others may lose.

Both outcomes produce data.

And if an idea turns out to be complete rubbish after 10,000 observations, documenting that properly may be considerably more useful than qdeleting it and starting again.

That's what the new Trade Carefully Research Lab is really about.

We're not launching three betting systems.

We're launching three questions — and giving the evidence enough time to answer them.