Upload zimg-fdloss/RECIPE.md with huggingface_hub

zimg-fdloss/RECIPE.md

@@ -22,7 +22,15 @@ holdout = sorted-enumeration idx % 12 == 0.
 3. **Optional TFC finishing dial** — short (250–500 step) pass tying the
    1-jump endpoint's SigLIP2 features to sg(own multi-step endpoint):
    redistributes quality between NFE operating points along a measured
-   exchange curve (~1.2 pts gained @1 per pt lost @8). Not net creation.
+   exchange curve (~1.9 pts gained @1 per pt lost @8, pooled over 5 runs).
+   Not net creation. **Mechanism caveat (measured)**: the dial as shipped
+   uses the standard stats root, whose warm-start moments come from the
+   iMF-stage model — a stale anchor relative to the checkpoint being
+   finished. Rebuilding the warm-start from the finished model's own
+   samples (honest variant, replicated at n=1000 and n=5000) erases the
+   @1 gain (z=−1.9 vs stale) and deepens the @8 cost (z=−3.7 vs
+   champion): part of the dial's @1 gain is stale-anchor replay, not TFC
+   alone. Keep the standard (stale) stats root when using the dial.
 
 ## Headline numbers (GenEval macro, 553×2 VLM-judged; FDr held-out judge mean)