Model transparency

How TahmVest thinks

We test models on the past without peeking ahead, say when we abstain, and keep helper labels separate from trained goals.