For each forecast we make, we wait until reality catches up, then publicly measure how wrong we were vs. the dumbest possible alternative. If we're not at least 15% better than dumb, we don't publish — and we say so.
A (cell, horizon) publishes its forecast publicly only if its rolling 90-day MAE is < 0.85 × the naive baseline. Below, gated cells are flagged.
| Cell | Horizon | Window | MAE | Naive MAE | Skill ratio | Hit rate | Band cov | N | Published? |
|---|