Versioned scoring models
Each model is a JSON file of weights and scoring curves under config/scoring-models/. New versions can be run side-by-side against the same observations and measured against real angler feedback — so improvements are compared, not guessed. The accumulating feedback dataset is what an empirical catch-probability model would eventually train on.
v1.0.0
Hand-tuned baseline weighting fish-count momentum and flow most heavily. The reference model every later version is compared against.
Factor weights
v1.1.0
Experimental re-weighting that leans on water temperature and weather, on the hypothesis that thermal comfort predicts feeding windows better than raw counts. Run side-by-side with the baseline to compare against angler feedback before promoting.
Factor weights