The table below shows different players projected career WAR through both the college and professional models. The results from the pro model shown here are fed in with the output from the college model as well as given the player's draft pick, but none of the player's pro stats have occurred yet. The pro model ends up using a player's draft position for their initial value far more than it uses the college model, but it still tends to give players a higher/lower than expected for the draft position ranking for players it was high or low on.

The architecture of the college player model is extremely similar to that of the pro one, so rather than repeating most of it, this section will focus on the differences between the two models.

The biggest difference between the two models is the timestep; while the pro model operates on a monthly tick, the college one only updates yearly. This was not a design choice; rather, this was do to not having access to monthly data. This ends up creating problems that the model is unable to account for, degrading its performance. The model is given the players' yearly stats as well as the average ranking of their opponents, but with all of the values put together it doesn't have a way to discriminate between a player who destroyed poor teams and struggled in conference play and a player who was solid throughout the entire year.

The other main difference is that the college model has far less data on a player's defensive skills than the pro model does. The college model is only given the player's listed position, while the pro model has the percentage of games for each position a player played at as well as an internal DRS-based defensive metric. This leads to the college model having little idea how good a player actually is defensively, other than that most college C will play C in pro ball. This results in a model that is far more likely to give top marks for C than it is for other positions.

This table is a collection showing how different teams' drafts evaluate through the pro model, compared to the draft capital that they used to acquire those players.

To determine a team's capital, each pick is assigned an expected WAR for hitters and pitchers. When a team drafts a player, they get the expected capital for the hitter/pitcher that they drafted. This model does not evaluate at which points in the draft should pitchers be drafted to get proper value, simply how well are teams doing with the players vs their expected results.

For each year's value, the team results are the sum of the WAR for the last output for each player that year, unless the player has exhausted prospect eligibility. In that case, it goes through a separate evaluation that takes in their projected MLB WAR in their remaining seasons of team control as well as their already accumulated WAR.

The biggest drawback to these results is that they do not account for trades; if a player was drafted by Team A and traded to Team B, their results count towards Team A, regardless of whether they were traded in their final year of team control or the offseason after the draft.

This table shows similar results to those in the Team Rank table, but these are shown on a per-player basis rather than a per-team basis. The description for that table shows the methods to generate these numbers.

When looking at the results, especially in older drafts, many of the results are marked with a *. This indicates that the player has exhausted prospect eligibility and now is being evaluated through an alternative method.

For this alternative method, there is a separate head in the combined neural net model that is predicting a player's MLB WAR by using the bucketing method used in the prospect model. The result shown is the sum of the WAR the player has already accumulated in the majors as well as the expected WAR that the MLB WAR model has for their remaining years of control.

One issue with this current version is that it doesn't actually track their control, rather just using the 6 years after prospect eligibility has been exhausted. This creates issues where individual players can both be overcounted and undercounted. The prospect eligibility metric used often has it exhausting later than it actually does for players who do lose it through service time rather than PA/IP, potentially giving a player an extra year. Conversely, a player who spends considerable time in the minors after losing prospect status could contribute an extra year, or even 2, more than this approach does. While this does lead to some inaccuracies on individual players, on aggregate this balances out.