Two things went wrong here today. Both are worth writing down properly, because a service that asks you to trust a number has to explain that number when it moves.
The 50/1
This morning our X thread carried U S S Charleston at 50/1 with “3pt Win” printed underneath it. That should never have gone out, and if you saw it and thought it looked wrong, you were right.
We publish inside a price band. Anything shorter than evens or longer than 7/1 is not a selection we are prepared to stand behind, because outside that range our rating stops telling us anything useful. The website has applied that band from the start. The X thread never has. The only filter that thread ever applied was which signals are allowed to publish at all, and price was simply never checked. So for as long as the thread has existed, it has been able to carry a price the website itself would have rejected.
Yesterday we brought the top of the band down from 11/1 to 7/1. That made the gap wider. It did not create it. This has been sat there the whole time and it took a 50/1 shot to make it visible.
It is fixed. The price band now applies to the thread the same way it applies to the site. On today’s card that takes it from eighteen selections down to eleven.
Stakes are coming off the thread
The other half of that tweet was the stake, and that is going too.
The point figure next to each pick was our internal staking plan, which weights selections differently depending on which signal fired. Printed next to a 50/1 shot it reads like an instruction to pile in, and that is not a reasonable thing to put in front of someone. From tomorrow the thread carries the time, the course and the price, and nothing else. The stake still sits behind the tracked record, so nothing about how we score ourselves changes. It is just not something we should be broadcasting.
The record moved on its own
Yesterday’s price change was meant to apply going forwards. It also reached backwards, which nobody asked it to do.
The record page rebuilds every morning. For the earliest days, before we started saving a copy of each morning’s card at the moment it was published, the page works out what the card would have been by re-running the selection rule. The rules about which signals counted back then were pinned to what was actually live at the time. The price band was not. So when the top price came down yesterday, every pick between 8.0 and 12.0 from the back end of May quietly dropped out of the record.
Six days that had been settled for two months changed. The published profit rose by about 8.75 points overnight, on no racing whatsoever.
That is corrected as of today. The record now reads 92.73 points and 13.8 percent, where yesterday it said 101.48 and 15.8. The lower number is the true one. The band those old days are filtered by is now frozen at what was genuinely live at the time, so no future change to our rules can reach back and move them again.
What we changed so it cannot happen quietly
The uncomfortable part of this was not the size of the error. It was that nothing flagged it. The record went up, and up is the direction nobody questions.
So there is now a separate ledger sitting alongside the site. Every pick that has ever appeared on the record goes into it once and is never removed. Results update as races settle, because that is legitimate, but the selection itself is written down permanently the day it appears. If a rule change ever drops a previously published pick again, the pick stays in the ledger, gets flagged, and the gap between the ledger and the site is reported as a number. We tested it by simulating a change that removes forty past picks. The ledger kept all of them and reported the divergence rather than absorbing it.
Had that existed yesterday, this post would have been written yesterday.
The short version
Our thread published a price it should not have, with a stake attached that should not have been there. Our record was overstated by 8.75 points for about twenty four hours. Both are fixed, the record is corrected downwards, and there is now a check in place that makes the second one impossible to miss.
If you follow the thread, you will notice it is shorter from today and no longer carries stakes. That is deliberate, and it is an improvement.