ablation.md: add pre-LIBERO checkpoint comparison (stage2/stage3/pretrain_A, all 0% on LIBERO closed-loop)

ablation.md

@@ -221,6 +221,7 @@ actually look like - should help.
 | pixel + proprio age | Success | 95% CI | delta | paired 95% CI | flips | McNemar p |
 |---|---|---|---|---|---|---|
 | 0.25 s | 21/150 = 14.0% | 9.3–20.5 | -70.0 | -78.0…-61.3 | +4/-109 | 0.0000*** |
+| 0.5 s | 0/150 = 0.0% | 0.0–2.5 | -84.0 | -89.3…-78.0 | +0/-126 | 0.0000*** |
 
 At 0.25 s, the internally-consistent version scores **14.0%** against **23.3%** for stale pixels with fresh proprioception — significantly *worse* (-9.3 points, 20 episodes lost against 6 won, McNemar p = 0.0094), not better. So the collapse is not explained by the observation being self-inconsistent — staling proprioception on top of the pixels makes things worse, not better. Holding proprioception back to make a cached vision tower's input "agree with itself" is not a viable mitigation: geometric staleness costs, and adding more of it costs more.
 
@@ -286,6 +287,24 @@ done: 12 cells plus two controls, 500 + 12x150 + 2x150 = 2,300 closed-loop episo
 are being evaluated closed-loop on the same suite, to size how much of the 85.2%
 full-rate number comes from the LIBERO fine-tune itself.
 
+## 9. Other checkpoints in the same repo, same harness
+
+Three checkpoints that predate the LIBERO fine-tune, evaluated closed-loop on
+`libero_spatial` with the exact same conventions and no ablation (full-rate,
+100 episodes = init states 0-9). None of them were fine-tuned on LIBERO, so this
+sizes how much of the 85.2% full-rate number is the fine-tune rather than the
+shared pretrain.
+
+| Checkpoint | What it is | Success | 95% CI |
+|---|---|---|---|
+| `checkpoints/stage2_final` | SO-100/101 + Bridge mixture, frozen backbone | 0/100 = 0.0% | 0.0-3.7 |
+| `checkpoints/stage3_final` | + spatial distillation | 0/100 = 0.0% | 0.0-3.7 |
+| `tinyvla2/pretrain_A` | TinyVLA-2 cross-embodiment ablation, variant A | 0/100 = 0.0% | 0.0-3.7 |
+
+**`tinyvla2/pretrain_A` scores 0.0% on LIBERO** (0/100), consistent with stage2/stage3: none of the three pre-fine-tune checkpoints do anything useful on this suite. This is a different model from the one the rest of this document ablates — it is the *control* arm of a separate cross-embodiment ablation (variants A/B/C), not a LIBERO checkpoint. Its own model card already reports it failing cross-embodiment transfer on held-out `jaco` (endpoint error 1392 mm vs. a 128 mm zero-prediction floor, n=8) — a result from the author, not independently reproduced here since the held-out-embodiment harness this document uses is LIBERO/MetaWorld/SO-101, not the jaco eval that number came from. `conditioning`, `morph_tokens`, and `action_space` in its config are not implemented in `modeling_tinyvla.py` as of this checkpoint family's release — they were added to `configuration_tinyvla.py` here as pass-through fields so the checkpoint would load, since variant A's own documentation states it is architecturally identical to the embodiment-ID scheme already implemented (`EmbodimentEmbedding`). Whether that also holds for variants B and C is unverified.
+
+---
+
 ## Reproduction
 
 ```bash