iiinova — where your digital self keeps evolving

He's like the anti-Brady Tkachuk. I'm looking forward to your write-up, but it looks like one of the changes has been more value for setting/finishing skill? IIRC, Laine was around -2 sG before the updates

If/when I have children, the urge to cram in as many middle names as I can think will be hard to resist. One of my cousins has seven middle names and I've always been slightly envious of that

000

perfectly cromulent name, my son is also named wilbrod, we're sold out of wilbrod license plates

130

Has anyone ever done a deep dive into Nashville w/r/t Saros and Brunette's system? Correlation isn't causation, but it's extremely weird to me that Saros went from a consistent top 5-10 guy to at or below league average in his age 28 season, which was also Brunette's first season in Nashville.

000

morbin' time, morbin' time, morbin'

030

Well, I may be interested in contributing if I think of any features I'd like to add

000

I've got some experience with python, but it's not my strongest language. What kind of new features are you thinking of?

what language is the codebase in?

100

I never wrote it up anywhere except this thread, so it just lived here and on my laptop. I wasn't sure how to validate my results, but using the EDGE animations is a really good idea, I'll have to try that

310

plus a positional adjustment where forwards shots were blocked closer to the shooter than defensemen

100

I estimated unblocked density for each shot type over the previous 1,312 games, then drew a cone out from the net plus a small margin of error to either side of the net and estimated original location from a weighted average of the unblocked shot density

100

Aaron K @ladyotreesyrup.bsky.social · 29/09/2026

I'd be curious to see how it compares to my attempt here: bsky.app/profile/lady...

100

Same thing for a fenwick-based 5v4 xG model Log Loss: 0.299, AUC: 0.677, Calibration: 1.001 Power Play xG is much harder, lol

density charts for xG density distributions for each of the last 4 seasons

011

Aaron K @ladyotreesyrup.bsky.social · 28/09/2026

Density distribution for fenwick-based 5v5 xG model trained each new day of games on the previous 1,000 game played to dynamically incorporate changing tactics and event recording behavior Log Loss: 0.203, AUC: 0.777, Calibration: 1.001 Looking forward to tracking performance this season

Aaron K @ladyotreesyrup.bsky.social · 24/09/2026

"One day I had to pump gas — it was a few years ago."

Aaron K @ladyotreesyrup.bsky.social · 23/09/2026

yep, there are about 25k player ids

Aaron K @ladyotreesyrup.bsky.social · 22/09/2026

Obviously this is pretty trivial when most locations are a foot or less outside the rink, but it’s potentially interesting for understanding the process

000

Aaron K @ladyotreesyrup.bsky.social · 22/09/2026

More changes in PBP event recording: starting in 2019-20, the number of events recorded outside of the rink's dimensions drops to almost zero suggesting at minimum some validation for coordinates. Most outside of rink events from previous seasons were from the corners

line chart of share of full coordinate run of play events recorded outside of rink

rink map of events recorded outside of rink

This also makes me wonder in specific instances like Joe Thornton or especially Alex Tanguay how much of their measured shooting talent is just them being extremely picky shooters. Have you ever plotted shootiness against finishing?

100

I'd love to know rates of attempts given possession of the puck in a potential shooting position

I've also found there's (possibly?) a slight tendency for better shooters to attempt worse shots, probably because they know _they_ can score from there. There's some internal calculus in shot selection that looking at raw shot distributions obscures

100

I've thought about this re: behind the net shots having such a high shooting percentage. There's information in the fact that the shot was attempted at all given that it's a bad place to shoot from

210

Made a mistake on the labeling here, first chart is all attempts, all situations

2023-24 is a weird outlier in a lot of ways. Over 10% of goals (10/68) from below the goal line being scored from inside the net seems is nearly double any other season and over 3x in raw count. Also completely explains the spike in goals from below the goal line compared to surrounding seasons

line chart of percent of goal line fenwicks recorded inside net

line chart of percent of below line goals recorded inside net

table of goals scored from below the goal line and goals scored from in the net

Aaron K @ladyotreesyrup.bsky.social · 17/09/2026

Don't think this is breaking any ground, but some interesting context for behind the net shots/shooting percentages. From spot checking, some behind the net goals are recorded from inside the goal, as in dekes or jam plays, but those aren't filtered here)

Aaron K @ladyotreesyrup.bsky.social · 17/09/2026

what did john gibson do to deserve this

Table of own goals against a goalie, John Gibson has the most with nine, next closest are Jake Allen and Cam Talbot with six each

030

Aaron K @ladyotreesyrup.bsky.social · 15/09/2026

It's arguably the second-most straightforward comedy of the nominees!

homer simpson telling mel brooks that young frankenstein scared the hell out of him

030

Reposted by Aaron K

I've spent a fair amount of time looking into recording changes for PBP shot locations with the introduction of puck and player tracking, but those aren't the only events that have seen changes in recent seasons. Giveaways, takeaways, and hits have also been affected

Density map of takeaways for the 2020-21 to 2025-26 NHL regular seasons

Density map of hits for the 2020-21 to 2025-26 NHL regular seasons

184

Post script and for completeness: penalty distributions have barely changed across the last six seasons. Mostly taken in the defensive zone, and mostly taken in the slot

Density map of penalties for the 2020-21 to 2025-26 NHL regular seasons

That's actually what makes the most sense to me, as I'd guess most of those "hits" were more like quasi-interference plays off dump-ins that full body checks on a puck carrier

Notably, while recorded hits have decreased in each of the past 5 seasons, raw counts of offensive zone hits have stayed fairly steady, so this could be a case of the across the board reduction just being unevenly distributed. That is, DZ hits near the blue line not being counted as hits anymore

100

Finally, hits haven't undergone changes as significant as giveaways and takeaways, but there has been a slight shift towards hitting in the offensive zone and deeper in the defensive zone over the past two seasons. This could be stylistic/tactical in imitation of the Panthers

There is probably some re-classing of takeaways as giveaways, which would be consistent with the shifts in zones, but the shifts within zones are notable too and suggest to me that the location of the player losing the puck is being recorded rather than the location of the player receiving the puck

100

And as with giveaways, this isn't simply a matter of removing takeaways from one zone, but part of an across the ice reduction in takeaways (except for deep in the receiving team's OZ, where they've increased)

Density map of takeaway counts for the 2020-21 to 2025-26 NHL regular seasons

100

Takeaways are sort of the inverse, with most occurring in the receiving team's defensive zone, but in 24-25 there was a major shift towards the receiving team's offensive zone (and deeper in the zone) and open ice takeaways vanish (along with an overall decrease)

100

If we look at raw counts though, we see that this was not just adding instances of the offensive team misplaying the puck, but part of an huge increase across the entire rink surface

100

Previously, giveaways were concentrated almost entirely in the receiving team's offensive zone (i.e., defenders coughing up the puck), but in 24-25, a much larger proportion are recorded in the receiving team's defensive zone giveaways. This coincided with a large increase in giveaways/60

100

Quick notes: - distribution heat maps are scaled for each season to get an idea of changes in to the spread of events - giveaway charts are from the perspective of the team getting the puck, not turning it over to make it easier to directly compare with takeaways

100

Since these events are often incorporated into xG models to provide recent context, I think it's worth digging in a little into what those changes are, starting with giveaways

184