Pillar — The College Years
Every rookie arrives with the prior chapters already written.
The game gives you a draft profile and an overall. It does not give you a four-year production curve, a conference, a system, or a name on the back of the jersey before Sunday. Hashmark does. We synthesize the college career — calibrated to the league the prospect would have played in — so the franchise has the back story that every real scouting department starts a draft report with.
What gets generated
For every incoming prospect, on first export, we generate a four-year college statistical history. The model is conditioned on the player's combine numbers, attribute ratings, traits, age, and position. Output is calibrated against real-world conference and position statistical distributions — a Power Five corner's curve is not a mid-major corner's curve, and a service-academy quarterback does not put up Air Raid numbers.
- Conference assignment, weighted to plausibility given the prospect's grade
- School and program archetype (Power 5 / Group of 5 / FCS / service academy)
- Position-appropriate stat lines, year by year
- Age-adjusted production curve (true freshman to redshirt senior)
- Awards, all-conference selections, postseason honors at probability
- Bowl game appearances and lines
- Injury history at probability, where it would shape the curve
- Transfer portal moves where the data supports it
Why calibration matters
SEC corners produce one statistical signature. MAC corners produce another. Texas A&M quarterbacks under the play-action regime put up different lines than Washington State quarterbacks under the Air Raid. If we generated college stats by sampling the league average, every prospect would look like every other prospect. By conditioning on conference and program archetype, we get a college career that reads the way college careers actually read — with the silhouettes you expect from each program type.
The model is calibrated against open public datasets at training time and held against a small founder-curated test set of real college careers to validate that the conditional distributions look right. The methodology is documented and versioned; generated stats are flagged as such throughout the product.
How it's used
Generated college careers anchor the prospect's draft profile in narrative space. The recap engine quotes them. The HashmarkIQ rookie estimate uses them as a prior. The Scout Archive timeline includes them as the pre-snapshot baseline. They are not decorative — they are a load-bearing input to the rest of the system.
Generated content is always labeled. Hover any college stat and the tooltip discloses synthesis, the model version, and the conditional inputs. The fourth wall stays up.