Deadwood handles the infrastructure.You handle the product.

Local git workflow

Train locally, commit manifests, let floppydisk.cc pin artifacts, flip production in git, settle promotions on-chain.

Step 1: You train locally
├─ Run: deadwood train --config .deadwood.yml
├─ Local model trains on your data
└─ You validate it works

Step 2: You commit
├─ git add .deadwood.yml
├─ git commit -m "feat: trading_taste_v2 model, 98.2% accuracy"
└─ git push

Step 3: floppydisk.cc detects change
├─ Sees new model definition
├─ Trains trading_taste_v2 on remote GPUs (optional)
├─ Stores final weights on Filecoin
└─ Model is versioned, permanent, immutable

Step 4: You deploy
├─ Edit .deadwood.yml: set trading_taste_v2 to production: true
├─ git commit -m "deploy: trading_taste_v2 to production"
├─ git push
└─ Your app immediately uses new model (no downtime)

Step 5: Avalanche records it
├─ Transaction: "user_xyz swapped to trading_taste_v2"
├─ Timestamp: [timestamp]
├─ Model hash: [hash]
└─ Now immutably recorded

Result: New model in production. Perfect auditability. No manual deployment. No infrastructure thinking.