What the 2020 models were fed

In 2020 the category had its two biggest nights, five months and a hemisphere apart. One of them took money. The other could not, and said so in its own announcement. The difference is not the technology.

Aintree, 4 April 2020

The real Grand National was cancelled. ITV ran the Virtual Grand National on ITV1 at 5pm on Saturday 4 April 2020, presented by Nick Luck with Alice Plunkett and Richard Pitman and commentary by Stewart Machin. ITV reported a peak audience of 4.8 million, about 30 per cent of the television audience that night, and an average of 4.3 million across the 30-minute programme (read 18 October 2026). The 2019 edition on ITV4 had drawn 737,000.

Potters Corner won it at 18-1 under Jack Tudor, then 17 years old, for Christian Williams, from Walk In The Mill at 16-1 and Any Second Now at 10-1, with Tiger Roll the 5-1 favourite fourth. Sporting Life gave Potters Corner at 16/1 where Racing Post and the Betting and Gaming Council give 18-1, and this desk records both (all read 18 October 2026). Of the 40 runners, 19 finished: ten fell, four unseated their riders and seven refused.

What went in

The input list is documented, which is unusual for this category. Racing Post names age, weight, form and weather conditions, with fallers, unseated riders and collapsing fences modelled, and says data for all 40 runners was entered and the computer generated the odds. Aintree and the Jockey Club describe the simulation as considering the declared runners and riders, their track records, the weather, the going, and their chances against past winners. One of the event's own sites adds a panel of experts to that list (all read 18 October 2026).

The mechanism underneath is the familiar one. The outcome is randomly determined but weighted by each horse's statistics: the software assigns probabilities and the graphics render the result that came out. The same sites put it bluntly — the computer favourite has a better chance than an outsider, but a random number generator decides each runner's result.

The one genuinely interesting engineering detail is about time. Inspired's standard virtual horses were built for races of two to three minutes. A four-mile National runs nine or ten. TVBEurope reported in April 2018, quoting Steve Rogers directly, that new horse algorithms were written so that movement shifts over the life cycle of the race, with horses tiring towards the end, and that the runners carry a built-in forcefield so they do not collide (read 18 October 2026). In 2020 that modelling was visible on screen: Tiger Roll led clear into the final circuit and faded under 11 stone 10 pounds to finish fourth.

What it could not know

Three things, each of them structural rather than a failing.

It could not know the going, because there was no race meeting. In 2018 the team assumed soft ground and used each horse's soft-ground record (TVBEurope, read 18 October 2026). In 2020 the going was an assumption about a day that never happened at a meeting that was cancelled.

It could not know the field. The 40 runners were the horses most likely to have contested Aintree, not a declared field, because nothing was declared.

It could not be checked. The weighting algorithm has never been published, Inspired describes it only in general terms, and this desk found no independent audit, no regulator statement and no methodology paper for it. An agent-based modelling paper notes in passing that synthetic horse-racing simulators are generally not public-domain research tools, which is the same observation from the other direction (read 18 October 2026).

And the race was pre-recorded. Neither the Virtual Grand National nor its companion Race of Champions was broadcast live; both were produced as late as possible and delivered to ITV as finished elements. Racing Post states flatly that the 2020 race had been pre-recorded, and the producers, Carm Productions, undertook to keep the result from leaking to the betting market through strict confidentiality (read 18 October 2026). Bets were taken on that race, capped at 10 pounds per customer per horse by agreement with the charity, and the Betting and Gaming Council announced over 2.9 million pounds in profits for NHS Charities Together, after reports in the days immediately following gave over 2.6 million (both read 18 October 2026).

Put plainly: the result existed on tape before the market opened, and the control that kept it there was a contract at a production company. That is the only integrity question in the whole of 2020, and it belongs to this race alone.

Flemington, 3 November 2020

The Lexus Melbourne Cup "The Race of Dreams" was announced on 8 September 2020, built by Inspired for the Victoria Racing Club with Network 10 as media partner, and broadcast across Australia on Lexus Melbourne Cup Day, Tuesday 3 November 2020, ahead of the real Cup. The field was 24 past Melbourne Cup champions, selected by a panel of experts from more than 100 past Cup stars; the two panel members this desk could trace are Gai Waterhouse and the caller Greg Miles, who has called 36 Melbourne Cups (all read 18 October 2026). The course was a 3D recreation of Flemington, and historical data and past form were used to estimate each horse's likely finishing position.

Makybe Diva won it, coming from deep in the field to beat Phar Lap in the home straight, with Carbine reported third by secondary coverage. Makybe Diva is the only horse to have won three Melbourne Cups, in 2003, 2004 and 2005, so the simulation returned the historically safe answer. The real 2020 Cup, run the same day, went to Twilight Payment.

The crucial sentence is in the announcement itself. No bets were offered, because the outcome was "already known in advance by a select few creating the race". The VRC instead invited people to run a sweep or post tips on social media. The Kentucky Derby Triple Crown Showdown on NBC on 2 May 2020 took the same position in the same words — "There is no betting on the virtual race" — and ran a pick-the-winner sweepstakes instead, with a morning line set by Churchill Downs' oddsmaker Mike Battaglia that made Secretariat 7-2 favourite. Secretariat duly won it (all read 18 October 2026).

The one that went wrong

Three weeks before the Cup, the same supplier built the Greatest Ever Cox Plate for Ladbrokes, broadcast on 23 October 2020 for the race's hundredth running. Fourteen runners: eleven Cox Plate winners plus Frankel, Secretariat and Deep Impact, none of which ever raced in Australia. Frankel beat Secretariat, with Winx — the virtual favourite, taken wide in the straight by her virtual rider — third. Makybe Diva, Bonecrusher, Surround and Rising Fast were left out. Racing Post reported that the result "sparked outrage from pundits and punters alike", quoting Australians calling it an insult and a joke (read 18 October 2026).

Nothing was wrong with the model. The damage was done by the field selection, which is a human decision taken before the engine is asked anything. That is the lesson the broadcast virtuals leave behind, and it is a drafting lesson rather than a technical one.

Scoring the record

The Virtual Grand National can be scored against reality, which is exactly what separates a real-event simulation from a scheduled virtual race. In 2017 it named the real runner-up. In 2018 it named the real winner, Tiger Roll, 22 hours early. In 2019 it named the real third. In 2021 its winner, Cloth Cap, was pulled up three out. In 2022 it named the real second and third. One correct winner in five comparable editions. Whether an edition ran at all after 2022 is not confirmed by any primary source this desk could find; the affiliate pages contradict each other and the official account still styles itself 2022.

This circuit makes no such claim and cannot be scored that way. There is no real Bellbird Steeple and no real field at Bellbird Rise for a draw to be right or wrong about. What a Gallop Post card can be checked against is its own published weighting table, which is a smaller claim and the only one worth making.