Why I rebuilt my old auction script with real statistical families
DBB2 is here!!!!
Most NBA analysis still treats player value like isolated seasons. A player pops, a player disappoints, a player gets hyped, and the conversation usually starts over from scratch.
I built DBB2 because I think that misses the more durable question: what kind of statistical family does a player actually belong to, and how often have we seen that shape before?
DBB2 is built from more than 30 years of NBA player-season data. It maps player seasons into recurring statistical archetypes, looks at centroid stat blocks, and uses rarity framing to separate common shapes from unusual ones. The point is not to pretend every player fits perfectly into one box. The point is to stop acting like every season is a brand-new mystery when a lot of the same player types keep showing up.
A scoring guard with a real creation gap is usually not a totally new puzzle. A boring rebound-and-defense frontcourt profile that keeps helping rosters is usually not a totally new puzzle either. A rare stat line is not automatically valuable, but it is worth understanding as a descriptive fact before you start building takes on top of it.
That is the lens here.
DBB2 is not trying to replace scouting, context, or common sense. It is a taxonomy and pattern-recognition engine. The stable part is the cluster ID and the statistical shape, not the nickname. The labels can stay provisional. The recurring family is what matters.
That is what I’ll be publishing from here: historical player families, archetype patterns, rarity notes, transition ideas, and the statistical shapes people keep misreading.
Less surface-level noise. More recurring structure.
DBB2 is supposed to do one thing better than surface analysis: place modern players and player seasons into recurring statistical families that actually survived era changes.
That means the useful question is not whether a player is exciting, polarizing, or famous. The useful question is what family the numbers place him in, how rare that shape actually is, and what history says tends to travel with it.
Today’s anchor example is Charles Oakley. Charles Oakley maps to SF_07 with a centroid block of 9.4/6.9/1.5/0.8/0.8 and a representative season of 1997-98 at 9.0/9.2/2.5/1.6/0.3.
A rebound-first, low-glamour frontcourt archetype that looks boring on the surface but keeps recurring in useful fantasy seasons.
The evidence pack for this post is simple:
54-cluster count
archive scope
sigma concept
1-2 concrete cluster examples
caveat that labels are provisional
Historical examples in the same family: Derrick Coleman 2002-03, Dickey Simpkins 1998-99, Michael Kidd-Gilchrist 2016-17. Recent echoes from the current archive: Precious Achiuwa 2021-22 (9.1/6.5/1.1/0.5/0.6), Jabari Walker 2023-24 (8.9/7.1/1.0/0.6/0.3), Richaun Holmes 2021-22 (10.4/7.0/1.1/0.4/0.9). The clearest rarity signal is STL: 1.557 (+1.72 sigma vs SF).
This is where the DBB2 lens matters. A rare shape is descriptive before it is predictive. A clean cluster assignment is useful before the nickname is. And a recurring archetype is most valuable when it helps explain why similar looking seasons can produce different reactions in public conversation than they do in the historical record.
Safe framing for this post: This is the taxonomy DBB2 uses to describe recurring statistical families, not a claim that every player fits neatly into one box.
That is the difference between a taxonomy and a take. The taxonomy gives you a stable statistical family. The take is what you build on top of it.
DBB2 is the authority layer. Court Dominion and WinstAPlayer are application layers that sit downstream of this work. So this post stops at the evidence and the lens, even when the product implications are obvious.
It’s a pleasure to be here serving up DIMES!!!!