BSS = 1 − (model_brier / baseline_brier). Positive = brain beats always-mean baseline. >0.10 = useful.
Brier = mean (predicted - actual)². Lower better.
Win rate = % of historical resolutions that went the brain's predicted direction.
Symbols with <10 resolutions don't have a skill score yet (statistical uncertainty too high).

The global Brier Skill Score says "is the brain learning?" overall. Per-symbol BSS says where it's learning. The brain may have +0.20 skill on SPY but -0.05 on TSLA — averaging hides this asymmetry.

How to read it (not a sizing rule):

• Per-symbol samples are small (tens of resolutions), so these scores are noisy and the extremes are partly luck — the best of ~72 names will look strong by chance.
• The brain's only proven-level result is the overall walk-forward number (mostly drift, no significant timing edge) — no single symbol has a statistically proven edge here.
• A persistently weak/negative name is a reason for extra caution; a persistently positive one is worth watching. This is exploratory research, not advice.
• Symbols not yet on the board: still warming up; revisit after >10 resolutions.