The forum post goes up on a Sunday night. “Twelve winners from thirty bets, 14% ROI, average price 4.0. Am I a tipster yet?” The replies arrive within minutes. Some say yes, post your selections. Some say no, the sample’s too small. Nobody can say how much of that gap could be luck.
An Archie score can put a number on it. It compares the winners you actually had with the winners your odds said to expect, and tells you how often luck alone would produce a gap that big.
ROI is profit divided by the total you staked.
The short version
- Work out expected winners: add up 1 divided by the decimal odds of every bet you made.
- Count your actual winners.
- Put both into the Archie formula below.
- A bigger score makes luck a less convincing explanation for the gap.
- It doesn’t tell you whether you’d make money, and a short record gives a shaky score.
What does an Archie score measure?
It measures whether the gap between your actual winners and your expected winners is bigger than luck usually produces.
It uses the same two numbers as the A/E ratio (actual wins divided by expected wins). A/E tells you how big the gap is. The Archie score tells you how much to trust it. What is A/E ratio? covers the first half.
Neither tells you about profit after commission. So why bother? Because the score has a narrower job: it helps you decide whether a record is worth testing further before you treat it as evidence of a method that works.
How do I work out an Archie score?
You need three numbers: how many bets you made (n), how many winners you actually had (W), and how many winners the odds expected (E).
Expected winners is the easy bit. Divide 1 by the decimal odds of each bet. A horse at 5.0 has an implied chance of 0.20. A horse at 2.5 has 0.40. Add those up across every bet. If your thirty horses added up to 7.2, the odds expected about seven winners.
The formula:
Archie = (W – E)² ÷ E + (W – E)² ÷ (n – E)
Worked example with the forum post: 30 bets, 7.2 expected, 12 actual.
- The gap is 12 – 7.2 = 4.8. Squared, that’s 23.04.
- 23.04 ÷ 7.2 = 3.2.
- 23.04 ÷ (30 – 7.2) = 1.01.
- Add them: Archie = 4.21.
Now turn the score into a rough chance. This table shows how often luck alone would produce a gap at least this big, in either direction, if the horses were priced fairly:
| Archie score | Chance luck alone gives a gap this big |
|---|---|
| 0.5 | about 48% |
| 1.0 | about 32% |
| 2.0 | about 16% |
| 3.0 | about 8% |
| 4.0 | about 5% |
| 5.0 | about 2.5% |
| 7.0 | about 1% |
The forum poster’s 4.21 means a gap that size comes up about 4% of the time from luck alone. That is interesting. It isn’t proof, and it’s nowhere near enough to stake a bankroll on.
Read that table carefully. It does not say there’s a 4% chance the record is luck. It says that if the horses were really priced fairly, you’d see a gap this big about 4% of the time. The two sound similar and aren’t.
Why do small samples lie?
I ran a simulation to show it. This is not real racing, it’s a bettor I made up. Every bet is at decimal odds of 2.9, priced fairly, so the true A/E is exactly 1.00 and there is no edge to find. I let that bettor place 100 bets, then 500, then 5,000, and repeated each size 20,000 times. Then I looked at how far A/E wandered, and how often the Archie score came out above 4.
| Bets | Middle 90% of A/E results | Share of records scoring over 4 |
|---|---|---|
| 100 | 0.78 to 1.22 | about 5% |
| 500 | 0.90 to 1.10 | about 5% |
| 5,000 | 0.97 to 1.03 | about 5% |
Two things come out of that.
First, A/E swings a lot on a small sample. With 100 bets and no edge at all, a bettor can easily look 20% above or below expected. That’s why a short hot record tells you so little.
Second, the Archie score keeps its false alarms steady whatever the size. About one record in twenty with no edge at all still scores over 4. That’s what makes the score useful: it already allows for how many bets you’ve had.
So a score of 4 or more is where I start paying attention. But one in twenty is not rare. If you tried fifty filters and kept the best, you’d expect two or three to reach 4 on luck alone.
Forum poster, take note. Thirty bets with a 14% return and an Archie near four may be a tipster. It may also be one of the lucky draws, and at thirty bets the two look the same. More bets is what tells them apart. In the simulation, 500 bets was where A/E settled to within about 10% either way, so that’s the sort of length I’d want.
What does it look like on real data?
Here is Race Advisor’s database: every GB and Irish race from July 2013 to 14 October 2026, backed at Betfair Starting Price (the exchange’s closing price) after 2% commission. VDW and Race Predictor are two of Race Advisor’s own ratings. Rank 1 means the runner a rating puts top in its race, and where two runners shared the shortest price, both count as the favourite. I worked the Archie scores from the rounded win rates and A/E figures, so treat them as approximate.
| Signal | Races | Win rate | A/E | Archie score |
|---|---|---|---|---|
| Betfair SP favourite | 167,257 | 34.4% | 1.00 | under 1 |
| VDW Rank 1 | 153,920 | 21.6% | 0.99 | about 10 |
| Race Predictor top horse | 167,314 | 25.5% | 1.01 | about 3.6 |
Read it one row at a time.
The favourite is the cleanest. Over 167,000 races its winners landed within a whisker of what the prices expected, and an Archie score under 1 means luck explains the gap with ease. That’s what a well-priced market looks like.
VDW Rank 1 has an A/E of 0.99, so it won about 1% fewer races than its prices implied. The Archie score of about 10 says that gap is bigger than luck usually makes. It’s also a tiny gap. A large score doesn’t make it a large one.
The Race Predictor’s top horse won slightly more than expected, with an A/E of 1.01. Its Archie score of about 3.6 sits just under my line of 4, so it’s a “keep watching” result, not a finding.
Notice that the score doesn’t tell you which way the gap points. VDW Rank 1 scores high because it fell a little short of its prices, not because it beat them. Always read the Archie score next to the A/E.
PR, another Race Advisor rating, puts the same horse top as the favourite in 81% of races, and its results were almost identical, so I haven’t listed it as a separate row.
That’s one 13-year stretch of racing. A different window could come out a little differently, so I’d read these as a picture of how the market has priced these signals, not a forecast.
What can an Archie score not tell you?
Three limits.
It isn’t a profit test. A high score makes luck a less convincing explanation for the gap. It doesn’t say the gap survives commission, or that you could get the price.
It assumes each bet stands on its own. If several of your bets share a race, or one factor drives many of them, the results move together, and the score then makes luck look less likely than it really is.
It’s no use against searching. If you tried fifty filters and kept the best, a big Archie on that filter is expected. Why systems fail when you backfit them covers that trap.
How do I use it on my own record?
Three habits.
- Record the decimal odds you actually took on every bet. Not a tipster’s advised price. Use the same price to work out E.
- Keep running totals. Wins (W), expected wins (E) and bets (n). The score updates after every result.
- Set a sample target before you start. My own rule: below 300 bets I’d call it noise, and I’d want about 1,000 before trusting a result. That’s a rule of mine, not a tested one.
The point isn’t to wait around. It’s to stop reacting to every short losing run as if it were evidence, and stop celebrating every short winning run as if it were proof. The score is cheap, it takes a few minutes in a spreadsheet, and it lets you enjoy a hot streak without betting the house on it.
To try it, build a simple spreadsheet of your own with columns for date, course, race time, horse, the price you took and the result. Today’s racecards are a good place to find the runners and prices to log. Check W against E weekly. This article doesn’t cover how to find selections, only how to judge a record once you have one. Testing a system before you bet it live goes further.
