problem that you’re rightly pointing out (that may be helpful for Flanker): Predicting low base-rate events is one of the hardest problems in forecasting and statistics because the thing you're trying to predict happens so rarely. Only a tiny fraction of high school football players will ever be drafted into the NFL. Estimates vary by year, but it's well under 1% of high school players. That creates several challenges. 1. Most predictions should be "no" Suppose 100,000 high school players exist and 200 eventually get drafted. Drafted: 200 (0.2%) Not drafted: 99,800 (99.8%) A model that predicts "not drafted" for everyone would be 99.8% accurate. This is called the accuracy paradox: a model can appear highly accurate while being completely useless for identifying the rare outcome. 2. False positives dominate Even a very good predictor can generate many more false positives than true positives. Imagine a scouting model that: Correctly identifies 80% of future NFL draftees Incorrectly flags only 5% of non-draftees That sounds excellent. But with our example: Actual Outcome Number Future draftees 200 Non-draftees 99,800 The model finds: 160 true future draftees (80% of 200) 4,990 false positives (5% of 99,800) So among the 5,150 players flagged as NFL prospects: 160 are real future draftees 4,990 are not Only about 3% of the flagged players actually make it. This surprises people because the model sounds strong, yet most positive predictions are still wrong. 3. Small differences matter enormously When outcomes are rare, tiny differences in probability become meaningful. For example: Player Draft Probability Average HS player 0.2% Strong prospect 2% Elite prospect 10% National blue-chip recruit 30% A 10% chance sounds low, but it's actually 50 times the baseline rate. People often interpret probabilities in absolute terms ("10% isn't much"), while forecasters care about them relative to the base rate. 4. Many important factors haven't happened yet A freshman's NFL prospects depend on events years in the future: Physical growth Injuries Coaching Position changes College choice Motivation Academic eligibility Competition level A model must predict not only current ability but also an entire developmental trajectory. This creates huge uncertainty. 5. Selection effects distort the data The players who reach college football are not random. For example: High school players are selected into college programs. College players are selected into starting roles. Starters are selected into combines and scouting evaluations. Prospects are selected into the draft. At each stage, the population changes. A variable that predicts NFL success among all high school players may not predict success among elite college players, and vice versa. 6. Rare events are especially sensitive to luck When outcomes are common, randomness averages out. When outcomes are rare, luck plays a larger role. Two equally talented players may have very different outcomes because: One tears an ACL. One gets recruited by a better college. One develops later physically. One enters the draft in a stronger or weaker year. For low base-rate events, chance often explains more of the outcome than people realize. 7. The "needle in a haystack" problem Most machine learning methods learn patterns from examples. If only 0.2% of observations are positive: There are relatively few successful cases to learn from. The model can easily overfit. Small quirks in the positive cases can look like meaningful patterns. This is why predicting rare diseases, financial fraud, terrorist attacks, startup unicorns, and NFL draft picks all share similar statistical challenges. A useful intuition Imagine trying to identify future NFL draftees among 100,000 high school players. The challenge is not finding someone who looks athletic. Thousands do. The challenge is distinguishing: the 200 who will eventually be drafted from the thousands who currently look almost as promising but won't make it. That's the core difficulty of low base-rate prediction: when the event is extremely rare, even strong signals leave substantial uncertainty, and most "promising" cases still won't experience the rare outcome