batch 3 grid + honest verdict (chinoll leads 6-2-7)

README.md

@@ -58,13 +58,25 @@ The exact instruction is printed on every row — judge it yourself:
 - **Tied**: black hat (both land it), city-street background (both), watch
   removal (both barely change it — a hard edit for everyone).
 
+**Batch 3** — 5 different portraits, the hardest surgical categories
+(eye color, freckles, eyeglasses) plus environment and style:
+
+![v3 portrait head-to-head batch 3](eval/real_grid_v3c.jpg)
+
+- **chinoll wins all three small facial edits**: "bright blue eyes" (armV3
+  leaves them dark), "round eyeglasses" (armV3 adds none), and "winter with
+  snow" (chinoll's snowy scene is more convincing).
+- **Tied**: black-and-white (both desaturate cleanly), freckles (both subtle).
+
 ## Verdict
 
-**Across all 10 portrait rows: armV3 2 wins, chinoll 3 wins, 5 ties.**
-armV3's wins are the small surgical/accessory edits and dramatic lighting —
-the exact categories Reddit said the incumbent fails. chinoll's wins are
-outfit and scene swaps. The two models are competitive overall; armV3 is the
-better choice when the edit is small, surgical, or lighting-driven.
+**Across all 15 portrait rows: chinoll 6 wins, armV3 2 wins, 7 ties.**
+armV3's wins are sunglasses and dramatic studio lighting. chinoll leads on
+facial-attribute edits (eye color, eyeglasses), scene swaps (beach, snow),
+and outfits. **Honest bottom line: chinoll is the stronger all-round editor
+today; armV3 beats it specifically on accessory edits and lighting-driven
+changes** — and we're not there yet on facial attributes. The grids above are
+the evidence; judge them yourself.
 
 Full per-candidate images:
 [akhaliq/wiki-eval-images](https://huggingface.co/datasets/akhaliq/wiki-eval-images).