So I have some vague and rambly ideas about cody ceci and player evaluation models and I apologize ahead of time for the long and ramblyness.

Gonna preface this with some general pfaff about kinds of certainty because every time I talk about model weaknesses or weirdnesses people get the wrong idea in the same wrong way.

In very broad terms, the "traditional way" of evaluating is built from the bottom up, like most of how humans learn things. Some things of which we are extremely certain, on top of them things of which we are pretty sure, on top of them the tendentious stuff, then the hot takes.

Model-based knowledge I think of as more inside-out. A big mess onto which some structure is imposed; where the myriad assumptions are all fairly well-satisfied it lines up right, out at the edges things get weird, the madness encroaches from the outside.

Less architecture and more map-making, if you like.

Anyhow, enough of that waffle, what about cody ceci. First, depth chart. He's had seven full (ish) seasons in the league. Never been primarily a third-pair guy; started as a #4 mostly, three seasons as a #2 and last season back to a #3-#4 with Toronto.

Now I have had the pleasure* of watching, at a rough guess, all but maybe ten or twenty of his games with Ottawa, somewhat fewer of his games with Toronto. I mention this as something like the opposite of an appeal to [my own] authority.

The reason that I don't trust any eye-test is not because I think most people judge what they see poorly, although they do. Canute was surely a very powerful king but that wasn't the reason why he could not command the tide. Still, my eyes are unusually useless.

That said, from the earliest days I saw what everyone watching cody ceci has seen: to observe him playing the puck is to be in a state of grievous distress. It leaps out at you. As if the man were beset by spirits.

But as I say, I do not trust my eyes anymore than anybody else's, no eyes of any skill can judge impact. That kind of complexity requires math (and sadly therefore computers).

So my "main" model is the one that estimates impacts on 5v5 shot rates, which are the most important aspect of hockey, if not all of it by any stretch. What about our lad Ceci.

The way the model works is that it tries to estimate a player's impact as of a specific instant, namely, the final moment of each regular season. After every season I update the estimate from as-of the previous off-season using the results from the season just finished.

So, rookie season, playing mostly with the finger-pointing slow-moving perpetually-stubbled ghost of chris phillips, ceci comes out looking pretty good - maybe not great defensively, but with some real offensive promise. Watching him you can see he can clearly skate.

The next season he splits his time between Jared Cowen and Patrick Wiercioch. (The top pair is EK witha rotating cast of Phillips, Borowiecki, and then mostly Methot) The results could not be more different.

Model estimate at the end of this is that Ceci isn't really contributing much offence, defensive weaknesses start to stack up; nerds are starting to talk up Wiercioch's results but traditional types prefer Ceci, he skates better and throws hits. Puck use remains butchery.

Next season (15-16) Ceci plays with the same two as before but also a chunk of Dion Phaneuf and also four exciting games with Mike Kostka. The on-ice results are dire all-up but the crease /does/ remain clear and he scores ten himself.

Model evaluation at the end of the year: -2% xG/60 impact on offence to his own team's shots, +5% xG/60 to his opponent's shots. Starting to form a type by way of impact: a few points, physical, awful with the puck on his stick unless actually shooting.

Next year (16-17) the Senators think they finally have their top four sorted. Karlsson-Methot are the top pair, Ceci-Phaneuf is the second pair. Apart from the odd blip where they switch spots in icetime that's the whole season.

Part of why I highlight Phaneuf is that he's a known commodity and his end-of-season impact matches exactly during his time in Toronto and later in Ottawa (namely, blah offence, rough defence).

That's just what you want from a well-tuned model that tries to measure aspects of human performance: slow evolution over time, since human ability also changes slowly over time.

Senators with Ceci on the ice are dire offensively and blah defensively and Ceci personally cops most of the model blame. Forward teammates are a mixture of good (Stone, Pageau) and bad (Pyatt, Kelly) and other (Dzingel, Hoffman), and breadth of QoT makes me feel at ease.

That doesn't quite fit the "type" for Ceci, though; which is a little unusual; file that away.

That season was the one that the Senators went to the conference finals btw, the following off-season featured a LOT of recrimination and finger-pointing about "SEE we ARE good" versus "is it not CLEAR how much of that was LUCK". Much imbroglio

The Certainties on defence the following year (17-18) are Karlsson on the top pair and Ceci on the second pair. (Both righties, never together except for 26 hilariously bad (net -62% xG/60 lol) hail-mary-oh-god-we're-gonna-die minutes). The rest of the defence corps is A Mess.