Update integration handoff and CONUS retraining notes

models/wildfire_fm/README.md

@@ -15,6 +15,9 @@ Use `modeling_unet.py` to instantiate the compact U-Net architecture before load
 For tiled serving, use `tiled_inference.py` or an equivalent overlap/halo
 stitching procedure. Avoid independent non-overlapping 32-by-32 tiles because
 they can introduce tile-center and edge artifacts.
+For the integration question checklist covering channel order, normalization,
+CAPE selection, validity masks, static resampling, spatial aggregation, and
+CONUS retraining status, see `../../docs/hugh_handoff_status.md`.
 
 ## Training scope
 
@@ -22,6 +25,9 @@ The released checkpoints are California regional weights trained for the 5 km
 EPSG:5070 California grid. They should not be described as nationwide-trained
 weights. A machine-readable scope file is stored in `training_scope.json`, and a
 CONUS retraining recipe is provided in `../../training/NATIONWIDE_RETRAINING.md`.
+The CONUS template includes random-containing positive tile placement and
+optional train-split z-score normalization for continuous channels; no CONUS
+checkpoint is released yet.
 
 ## Input channels