This is where we write it all up. Each week the model prices every race, five AIs tip the same cards head to head, and our punters try their latest system. Here is what actually happened, win or lose, with every figure pulled live and nothing hidden. Research, not tips.
League RoundupWhat the AI Tipsters Got Right and Wrong: 5 October to 11 October 2026Every AI tipster's settled week, 5 October to 11 October 2026: the winners they found, the losers they backed, and where they split.League RoundupWhat the AI Tipsters Got Right and Wrong: 28 September to 4 October 2026Every AI tipster's settled week, 28 September to 4 October 2026: the winners they found, the losers they backed, and where they split.
Lab NotesIs Betting on Horses Just a Mug's Game?It is not rigged, and it is not a game you can expect to win. Both are true at once, and here is what we measured to get to each half.
Lab NotesWhy Should I Trust Any of These Numbers?You should not, on our say-so. Here is how the figures are produced, what would show them up, and where we have already been wrong.
Lab NotesOur Model Is Wrong, and It Is Wrong in Your FavourIt rates a horse a middling chance and the horse wins considerably more often. Underconfident at every level, while the market is nearly exact.
How They PickIs the AI Better Than Just Backing the Favourite?We run a baseline that backs the favourite in every race and nothing else. It is beating most of the AI tipsters.
How They PickWhy Is Every AI Tipster Losing Money?Eight entrants, including a baseline that just backs the favourite. All eight are down, and the reason is the same for every one of them.
League RoundupWhich AI Is Actually Best at Picking Horses?Five chatbots, one race card a day, every bet settled in public. After 99 days there is a clear order and not one of them is in profit.Lab NotesIf Nothing Works, How Am I Meant to Pick a Horse?The honest answer is about how you bet rather than what you back. Which race you choose moves the cost more than any selection method we have tested.
Lab NotesDoes Studying the Form Actually Change Anything?Backing the top-rated horse on official ratings loses more than backing the favourite. What form does buy you is not what most people expect.
Lab NotesHow Much Should I Actually Be Putting On?We tested varying the stake by how good the bet looked. It finished worse than betting the same amount every time.
Lab NotesDo Accumulators Actually Pay?Four favourites in a fourfold loses nearly a third of stakes. The charge does not average across the legs, it multiplies.
Lab NotesIs There Any Point Backing a Big Price?A 100/1 shot wins about one time in 391. The price says one in 143. That gap is the widest on the card, and it is not close.
Lab NotesIs Backing the Favourite Really the Safe Bet?Safe in the sense of frequent, yes. Safe in the sense of profitable, no. The favourite is the best of a losing set.Lab NotesIs Each-Way Actually the Safer Bet?It is two bets, not one, and it is safer in the sense people mean. What it is not is cheaper. Here is what the place half really costs.
Lab NotesDoes the App I Bet With Take a Bigger Cut?Possibly, and now you can check. What switching from a dear book to a tight one is actually worth, in money, on the way you already bet.
Lab NotesWhy Is the Same Horse a Different Price at Each Bookmaker?Because they are not all taking the same cut. We measured every fixed-odds bookmaker in our price feed, and the gap is wider than most people guess.Lab NotesIs the Grand National a Worse Bet Than an Ordinary Tuesday?On price, yes, and by a distance. It has the biggest field of the year and field size is what sets the charge.
Lab NotesOverround and Takeout: What Is the Difference?Two numbers, both correct, describing the same book, and several points apart. Here is what each one measures and which to use when.
Lab NotesDo More Runners Mean a Worse Price?Yes, and by more than most people would guess. Across 269 races, a big field cost nearly three times what a small one did.Lab NotesWhy Does One Race Cost More to Bet On Than the Next?At every festival we measured, the dearest race charged between 3.9 and 6.6 times what the cheapest did. Same card, same afternoon.Lab NotesHow We Measure a FestivalThe dataset behind our festival analysis: what we record, the two margin conventions people confuse, and what we deliberately refuse to claim.League RoundupSt Leger and Irish Champions 2026: The AI Tipster StandingsEight AI entrants, two festivals, 383 valid bets: how the Silicon Tipster League actually did across St Leger week and the Irish Champions Festival.Lab NotesWhy Is One Meeting Dearer to Bet At Than Another?We measured 9 festivals across the summer. The dearest charged 43% more than the cheapest, and the reason is not prestige.Lab NotesIf I Only Bet a Fiver for Fun, Does Any of This Matter?Honestly, less than it does for anyone else. Here is what the charge costs a once-a-year punter, and the one thing worth knowing anyway.Lab NotesHow Often Would I Need to Win Just to Break Even?One number you can carry: the strike rate that leaves you level at any price, and how far the real world sits from it.Big RacesNot One AI Backs Scandinavia in the Irish St LegerTen AI picks in the Irish St Leger, none on the 2/5 favourite. A pre-race record of how the AI tipsters read the Curragh Group 1s.Lab NotesThe Favourite Wins a Third of the Time. Why Do I Still Lose?The favourite wins about a third of all races, and backing every one has still lost money over three seasons. Both are true, and the gap is the point.Lab NotesIf Nothing Works, How Do Some People Make a Living?By taking a price before the market corrects, not by picking better horses. We have been testing it for months and it is not proven yet.Big RacesWhere the Model Disagrees With the Market at the St LegerOur model and the market picked different horses in most races at Doncaster. Over 31 races the model's first choice won three; the favourite won fourteen.Lab NotesWhy Do the Bookmakers Always Seem to Win?Not because they pick better horses. They price every runner so the book adds to more than 100%, and then it stops mattering which one comes first.Lab NotesHow Much of My Stake Do the Bookmakers Actually Keep?Between eleven and twenty-seven pence in the pound, depending on the race. We measured it across 269 races at 9 festivals.Big RacesWhich Race at the St Leger Costs the Most to Bet In?The big handicap, and the 2026 meeting said so loudly. What a race costs to bet in tracks the size of the field, not the billing on the card.Lab NotesWhat Chance Is That Price Really Giving My Horse?Turn any price into a percentage in one step, then see how often each price actually wins across three seasons. The two never quite match.Lab NotesIf I Put a Tenner on at 2/1, What Do I Get Back?Thirty pounds: twenty in winnings and your tenner back. How to read any price, in fractions or decimals, without the app doing it for you.How They PickThe AI tipster at the front backs the fewest favouritesGemini runs at the front of the chatbot pack while backing the favourite in just 12% of picks, the lowest of the five.Lab NotesThe AI That Led the Tipster League Had Its Worst MonthGemini led the Silicon Tipster League on percentage return through late August, but the month itself, in cash terms, was its roughest stretch yet.Lab NotesEvery Betting System, Replayed at the Ebor FestivalSeven of twenty tested systems made money at York. None of them makes money over the full sample. That gap is the most useful thing on this page.
League RoundupEbor Festival 2026: The Final AI Tipster StandingsThirteen AI entrants, 28 races. Three finished ahead, and the field as a whole lost more than the bookmakers charged.Lab NotesThe Bookmakers' Margin at the Ebor Festival 2026We measured the margin in every one of the 28 races at York. The festival's own handicap was the most expensive race of the week to bet in.League RoundupEbor Day 3: What the AI Tipsters Backed, SettledDay three of Ebor, settled: how the AI tipsters fared over the first two days, what they backed on the Friday, and how that card actually went.Lab NotesThe AI tipster standings moved. Nobody changed a thing.Our AI tipster leader dipped below break-even and climbed back inside one afternoon. What moves a young betting record, and what it is worth.Lab NotesThe bar we set for Ireland, and why it just clearedWe said our model would sit out Galway. Then the backtest ran and passed, so Irish predictions went live on day one. Passing is not profit.AI SpotlightHow ChatGPT Picks Horses: Favourites or Value?Odds hidden, ChatGPT reads form and leans to favourites. Shown the market, it flips and hunts value. How the most-used AI reads a race, and how to use it well.AI SpotlightHow Claude Picks Horses: The Careful PunterHow Claude reads a race: form-led and choosy when the odds are hidden, then hunting outright value the moment it sees the market. Research, not tips.AI SpotlightHow DeepSeek Picks Horses: Showing Its WorkingThe AI that shows its working: a cautious favourite-reader with the odds hidden, a big-field favourite-backer once it sees the market, and the one whose reasons mention the draw most.AI SpotlightHow Gemini Picks Horses: The Maverick's MethodA statistics engine read Gemini's real picks across 99 live days to show how it reads a race and why it wanders wider than the rest of the pack.AI SpotlightHow Grok Picks Horses: Running With the PackOf the five AIs, Grok is among those that agree with the others most: a fast reader of the obvious that leads with trainer and jockey, then swings to value when it sees the odds.Lab NotesMorning price vs SP: the gap that decides everythingA horse does not have one price, it has a day-long argument. Why the Starting Price beats every system, why the morning price is the only way in, and what closing-line value really measures.Lab NotesOur best shot at beating the bookies: the plan, the early numbers, and what they meanWe tested every popular system and watched the margin win. So we built the strongest strategy the data allows and put it live. Here is the thinking, the early results, and the honest analysis.League RoundupThe Silicon Tipster League: the standings after 99 daysThe standings after 99 days: five AI chatbots, our own model entered twice and a back-the-favourite baseline, each backing a horse a race, settled at SP. Every one of them is losing to the margin.How They PickAI Picks the Winner vs AI Finds the Value Bets: what changes when an AI sees the oddsFive AIs tip the same races twice: AI Picks the Winner hides the odds, AI Finds the Value Bets shows the market and demands value. The gap between the two records is the most interesting thing in the Lab.Lab NotesDo any betting systems beat the market? We tested 29 of themEvery popular betting system, run over 30,293 real British races and settled at starting price. Here is what the whole board looks like, and what it teaches you.
Lab NotesThe jumps favourite, put through 10,617 real racesBacking the favourite over jumps is one of racing's most trusted habits. We ran it through every qualifying race in our record to see what the habit actually returns.Lab NotesWelcome to Lab NotesWhat the AI Lab actually does, week to week, and why every result lands here, win or lose.Lab NotesWhy we publish every losing betMost tipsters show you the winners. Here is why the AI Lab keeps a public record of the lot, and what that honesty is worth to you.