Point the card at the BF16 comparison, including where it loses

README.md

@@ -65,18 +65,39 @@ harmless on this calibration set and fragile by construction.
 
 ## Samples
 
-`samples/` holds five clips made with `A3` driving `LTX25-distilled-DiT-Q3_K_M.gguf`
-— 768x512, 24 fps, video and audio muxed, 2.0-4.0 s, one seed per prompt, no LoRA
-and no latent upscale. Prompts and per-clip notes are in `samples/README.md`.
-
-`sample-00` is the representative one: an actual robot, dome head and segmented
-neck, and the BF16 encoder renders the same thing from the same card and seed.
-
-`sample-01` through `sample-03` do **not** follow their prompts — no robotic hand,
-no four-legged robot, no swarm of wheeled robots. The BF16 encoder fails those
-same three identically, across seeds and CFG up to 6.0, so it is a limit of
-LTX-2.5 distilled rather than compression damage. They are published unedited
-instead of being cherry-picked out.
+`samples/` holds every clip twice — once from the **BF16 original (26.264 GB)**
+and once from this **4-bit build**, with everything downstream of the encoder
+held identical: same DiT, same seed, same schedule, same VAE settings, one
+process. `compare-NN.mp4` stacks each pair, BF16 on the left.
+
+Two sets, because they answer different questions. `compare-NN.mp4` uses the
+vendor's `euler_ancestral`, which is what you actually get; it re-rolls noise
+every step, so the two builds return different *takes*. `compare-det-NN.mp4`
+uses deterministic `euler`, where the seed fixes the starting noise and the
+conditioning is the only thing left that can move a pixel — that is the set
+that attributes a difference to the encoder.
+
+Mean absolute error against BF16 over every decoded frame:
+
+| clip | `euler_ancestral` | `euler` |
+|---|---:|---:|
+| robot in rain | 0.0172 | 0.0182 |
+| dune | 0.0596 | 0.0580 |
+| forge | 0.0098 | 0.0129 |
+| night road | 0.0702 | 0.0252 |
+| smoke | 0.0128 | 0.0165 |
+
+**Four of the five deterministic pairs hold together** — same composition, same
+lighting, same timing, differing in surface detail. **The dune does not**: BF16
+renders a soldier in fatigues where the 4-bit renders a man in a business suit,
+from the same seed under a deterministic sampler. That difference belongs to
+the encoder, and it is published rather than cropped out. The qualification it
+deserves is that neither build followed that prompt — it asked for an astronaut
+and got neither — so the model had no confident answer there for a small
+conditioning change to disturb.
+
+`samples/README.md` carries the prompts, the per-clip conditioning drift, and
+what these clips do and do not establish.
 
 ## Memory