Who to target in the Bournemouth defence? Only got 4.5m? Go for Smith, gets less points than the rest but a cheap enabler for the fixture swing Want an attacking full back for assists? Truffert Mr Defcon? Hill Everything combined and consistent? Silva #fpl #fplcommunity

We backtest and refine on many seasons worth of historic predictive data and inferences from efficient sharp betting markets. I'd like to think projection models influence the talk of the town, rather than the other way round.

It depends how you define viable. He's almost certainly in the "best" team from a projected points POV, but your objective might be to increase your risk relative to the field. It's a bigger drop-off to go without him than it has been in recent seasons, and certainly has been more punishing so far, but I'd still say that's viable if that's your aim.

If money were no restriction, then I think most people would agree Isak over Jackson. It might not seem like budget is an issue now, but that can change quickly and it uses transfers to change your team structure around, so I wouldn't rule it out that you might need the money later.

We've built it to be predominantly automated now, factoring in a range of things like positions and roles and who else is likely to be on the pitch alongside them. We are monitoring the situation constantly though as you can imagine. Teams who play both a back-3 and a back-4 are a little fiddly for the wide centre-backs sometimes, especially for defcons, so that's probably the type of player we've changed the most on in our approach. No arguments here ❤️

When I ran solves for chip plans across the first half of the season in GW1, the worst chip plan I could come up with only scored ~10 projected points less than the best chip plan I could find. Of course that was just with a snapshot in preseason, and the landscape is dynamic, so it's likely the hindsight optimal best and worst strategies would have diverged more than that. I think the same still holds true for the upcoming stretch though. It will be more important how you play the chip line, than which chip line you use.

I think this depends more on how you are rating the more expensive mids and attackers. Thomas is not really better than the best defdners around 5.0-5.5, but he runs them pretty close for a decent price saving. He seems to be the standout option at 4.0 The benefits of a 4.0 defender are that you have more funds available elsewhere, and if they get injured and need to sit on your bench for a few weeks, they are not wasting money in doing so. The downsides are that if you do need to replace them, you either need spare money in the bank already, need to free up some funds, or need to accept a bit of a downgrade, although O'Shea and Davis are both pretty viable, and we have a lot more 4.0 defenders who play minutes than we used to.

There is no singular point at which a player becomes good or bad, this is binary thinking, which our brains use as a shortcut. As I said in an earlier post, if you think Palmer's underlying projections are too high, then one (or both) of two things must therefore be true: - Chelsea's team attack is rated too high - Chelsea's others attackers are rated too low x.com/SolioAnalytics/status/… We are currently projecting him at 0.32 non-pen goals p90 for the next 12 gameweeks, and he put up 0.27 last season in the Premier League, and 0.38 the season before.

Running the calculation this year again, we have a fairly low value for team value increases in the current state of play - only ~0.21pts per gameweek for every extra £1.0m of team value. Historically we have seen this number often nearer ~0.3 If we were to assume Bruno Fernandes is never viable and Saka remains injured for the rest of the season (extreme assumptions, for illustrative purposes only), then this value does drop a little, but only to ~0.19 The point at which team value is meaningless tops out a little lower there though at 114.0m, instead of 119.6m

I would say common mistakes (made by me also) are attributing too much value to small samples of data and metrics which hold little predictive value

This depends a lot on the landscape at the time. We can think of all plans as sitting somewhere within a region on a plane with one axis representing total projected points, and one axis representing amound of relative risk to the field. The shape of this region depends on the state of play at the time, including which players are projected to score well, and which players everyone else owns. You can see for example from two regions we mapped out last season for FH33 and FH34 that these can be of different shapes. In GW33, it was much more costly to dial up your relative risk than it was in GW34, but it was quite cheap to dial the relative risk down. (See how far to the right or left the "raw optimal" orange dot is on the upper frontier of the shaded reagion, and how steep the drop-off is moving to the left or right of it). I think the best approach is to try a few of the presets and see what changes, to better understand the cost of tuning your relative risk, and whether that is a cost you are willing to take based on your objectives. With regards to optimistic vs safe, those are more to do with absolute risk and plan fragility. A plan that depends on a lot of the landscape remaining the same might project more points if those assumptions hold, but a plan that is more cautious will be able to adapt to big changes more readily. Sometimes the fragile plan will come off, and sometimes the robust plan will still hit a run of injuries and fall behind anyway. When we tested over the course of a full season with different ways of valuing your free transfers back when the new rolling to 5 mechanic was introduced, similarly skilled managers could differ by around 15-20 projected points over 12 or so gameweeks in extreme examples, just from whether their team was more or less fortunate with injuries, rotations, changes in player or team strengths, etc. And that's before any variance from real points over projected points!

There aren't really any "great" defenders this week, and most of the bad ones are not all that far behind. Similarly in GW10, there are lots of great defenders, but also the not so great ones are still quite good so the gap isn't all that large. Unless these are defenders with massive issues in the medium term that you'd be selling, I don't think a hit is likely to pay off.

I personally am a big fan of the Bournemouth defence. They have kept many strong teams to low numbers of chances so far this season, conceding just 1.16 non-penalty expected goals per game with not an easy run of fixtures, up there with the best so far (outside of outlier great defence Arsenal). Silva has also been accumulating some good chances from set pieces too, 1.30 xG is 3rd among defenders so far this season. The other two are great picks on these metrics as well, good defensive numbers, good defcon rates, and some attacking threat too (0.70xG for JPVH and 0.55xG for Muharemovic, both in the top 20 for defenders) I think again it partly comes down to your own assessment of the team's defensive strengths, and which team's fixtures align best with the rest of your squad.

Both are good options, only a mistake if you were hoping to go with the crowd and you misread, or if you valued the price change and missed that last night.

For attacking returns, we've backtested a lot of the matchup ideas: positional matchups, playing-style clashes, one flank being weaker than the other. Once you control for team strength and individual player skill, a large part of the edge disappears. Some of it is real, but it's a pretty small factor, and you are quite at risk of overfitting and making things worse but putting your finger on the scale. I think it's generally finen to use matchup and situational angles in your own decisions or 50/50 calls, but when you're modelling hundreds of players every week and reducing error is the goal, overfitting is a bigger risk than missing a small edge. Defensive contributions are one of the areas most affected though. Team style (passing style, attacking tilt, how much of the ball teams have, etc.) influences how many defensive actions (or at least opportunities) the opposing CBs, FBs/WBs and MIDs get. We factor these into the fixture-by-fixture adjustments in the projection model directly for certain model components. Tread with caution I would say, the edges here are small, and general team and player quality will be the dominant factor.

Technically if you want to increase your chances of getting luckier, then you want to widen your distribution of outcomes, such that the odds of getting really lucky increases. You can do that by sacrificing some projected points for a more differential team. Thus, you are more likely to get really lucky! (See how the yellow distribution in the image has a fatter right tail, depsite a lower mean score) The downside is that you have also increased your chances of getting unluckier.

Hull City were an incredibly tricky case for team ratings this season. Their underlying numbers in the championship had them in a relegation battle rather than pushing for promotion, and then there was a high level of squad turnover in the summer. We've seen that work before (eg Sunderland last year), but we've also seen it not work, and Sunderland's underlying data last year was also not as stellar as their actual 7th placed finish. Their defence has been more of a surprise than their attack. For non-penalty expected goals conceded, they do sit 13th, a good chunk ahead of Sunderland, Ipswich and Palace down the bottom. We are updating our team strengths in line with this, but preseason indicators were very poor, so it will take time to adjust, and this is callibrated on historical accuracy. There have been cases where teams like this do not carry on the defensive performance levels that we have seen from Hull so far, but the more they keep it up, the more we can accept this might be their level. Offensively though they have been poor. The have generated under 1 non-penalty xG worth of chances per game on average, with only Coventry below them. I don't think there are many indicators to suggest they are being underrated in terms of attacking output.

New front 3 template is set. 😴 Jackson - Isak - Haaland It’s going to get boring soon when all three start returning and everyone’s celebrating the same points every week. 😂 #FPL

Who else bought Jackson last night? He had been bought by 502k managers as of 17:00 yesterday. Now that number is up to 815k. Over 300k new owners, and according to @LiveFPLnet , he could rise in price again as early as tomorrow night. #FPL

🚫 GW9: FIVE POPULAR PLAYERS TO AVOID BUYING Flagged players excluded. Recent data covers GW6–8 👇 🔴 Bruno £11.7m — 24.3% owned 2 points and 0.28 xGI. Chelsea away next: weak recent output for that price. 🔵 Palmer £9.8m — 27.4% 9 points and 0.49 xGI. United offer opportunity, but his underlying form remains modest. 🩵 Gvardiol £5.7m — 26.9% 5 points in 143 minutes. Brighton scored seven goals over the same three GWs. 🩵 Cherki £8.0m — 38.4% 1.17 xGI shows threat, but only 135 minutes played. Check cup involvement and team news. ⚪ Groß £6.2m — 46.3% 11 points and 1.04 xGI. City away makes this a purchase to defer—not necessarily a sale. 🧠 Popularity isn’t a reason to buy. Neither is one poor week a reason to sell. #FPL #GW9 #FPLCommunity