The Rise And Fall Of The Shumpert Knicks A Decade Of Sundowns Upsurges And Regrets

Building a Player Evaluation Model: What Actually Matters Beyond the Box Score

The modern approach to evaluating NBA guards requires moving past basic counting stats and looking at how players actually create and consume shots. I spent about eighteen months building an internal scouting model for a D-League affiliate, and the process taught me that most people completely overvalue certain metrics while underweighting others that matter way more in practice. When I was putting together the framework, I used a lot of case studies from the last decade. The most instructive one came from watching how the New York Knicks handled their wing players over roughly ten years. You had periods where guys like Iman Shumpert were giving you two-way impact that directly correlated with wins above replacement, then you had stretches where the same player was essentially invisible because the system wasn't built around his skill set. The pattern repeated itself across multiple eras. There would be a period of upside where the roster construction seemed to click, followed by a regression that had very little to do with individual talent and everything to do with shot creation geometry and spacing. This happened not just with the Knicks but with other organizations that went through similar cycles.

The Framework That Actually Works

Start with shot profile analysis before you look at efficiency numbers. Most evaluators lead with true shooting percentage, which is useful but misleading if you are comparing players who take fundamentally different types of shots. A guard who pulls up frequently off the dribble against closeouts will always have a lower effective field goal percentage than a guard who gets easy looks at the rim or in the corner. This does not mean the first player is worse. It means you need to adjust your expectations based on shot difficulty, which is where league-wide tracking data comes in. From there, add pace adjustment. A player who accumulates solid counting stats in a slow tempo environment will project differently than one who does the same things in a fast-paced system. I used to build pace-adjusted per-possession models for each player I evaluated. This typically took about twenty minutes per prospect once you had the template built, compared to the hour or more it would take to manually crunch the numbers by hand. Then layer on defensive impact, which is the part most people get wrong. Box score plus-minus is not the same as actual defensive value. You want to look at opponent field goal percentage at the rim when the player is on the floor versus when they are off. You want to see contest rates and switchability. For wing players specifically, the ability to guard multiple positions correlates far more strongly with team success than any traditional stat line shows.

A Real Problem I Ran Into

About six months into building these models, I hit a wall with a specific type of player. We had a guard who looked terrible on paper. His usage rate was high, his efficiency numbers were mediocre, and the basic advanced stats suggested he was a liability. But every game I watched him play, he was making plays that the model was not capturing. He was finding soft spots in zone defenses, drawing help defenders, and kicking to open shooters at a rate that should have shown up somewhere in the data. The issue was that our model was built around individual offensive production, not creation for others. The workaround was adding an assist rate metric adjusted for team shot volume, combined with a visual component where I'd tag each of his passes as either high-value (leading to an open three or a dunk) or low-value. This took an extra hour per game to watch and tag, but it revealed that his actual playmaking value was in the top quartile for his position group. Without that adjustment, we would have completely missed him.

Counter-Intuitive Things That People Miss

First, a player's shooting percentage on catch-and-shoot threes is far more predictive of future performance than their pull-up three-point percentage. Organizations that invest heavily in developing pull-up range for non-shooters often waste time on it. The conversion rate from catch-and-shoot to pull-up is not where the value lives. If a player cannot shoot off the catch, no amount of mid-range work will fix it at the professional level. Second, turnover rate correlates poorly with overall offensive value when you isolate ball-handling guards. A guard who takes on heavy responsibility will always turn the ball over more. What matters is the ratio of assists to turnovers and whether those turnovers come from forced plays or from bad decisions. I learned this the hard way after spending too long trying to filter out high-usage guards from our targets because their turnover numbers looked alarming on the surface.

Where This Approach Breaks Down

The model has real limitations. It works best for players who have accumulated at least a season of NBA or NCAA Division I data. With fewer games, the sample sizes in tracking metrics become unreliable. You end up with numbers that look meaningful but are really just noise. I have seen evaluators make confident projections based on as few as fifteen games of tracking data, which is not enough to establish a reliable signal. Another failure mode is when the player is changing systems. A wing who thrived in a motion offense with constant screening and cutting will look very different on paper when moved to a iso-heavy scheme. The model will reflect the decline without explaining the context, which can lead to premature drops in evaluation. You always need to watch the film even when the numbers are telling you one thing. If your resources are limited and you cannot build a full tracking-based model, the closest alternative is using Synergy Sports clips combined with regular box score stats. It is not as precise, but it gives you enough film evidence to catch the gaps that raw numbers will miss. A couple of hours of clip review per player gets you further than trusting the spreadsheet alone.

Putting It Together

The practical workflow runs like this: pull tracking data for shot profiles, run pace-adjusted per-possession calculations, add defensive contest and switching metrics, then watch game film to validate what the numbers are showing and catch the edge cases they miss. The whole process for one player takes roughly three to four hours if you have automated data pipelines in place. Without automation, it is closer to six or seven. The players you will identify as undervalued are the ones whose shot profiles and defensive versatility create value that basic stats do not reflect. The ones you will avoid are the high-usage guards whose efficiency numbers look decent but who take the wrong shots and generate no secondary creation for teammates. Both categories exist in every draft class, and the model helps you sort them out.