Upload experiments/phase6_qualitative_latents/results/autointerp/summary.md with huggingface_hub

experiments/phase6_qualitative_latents/results/autointerp/summary.md

@@ -6,17 +6,20 @@ N_TOP_FEATURES = 32  (from 8).  Multi-judge = 2 Haiku prompts.
 
 | arch | seed | concat | /N sem | cov k/P | cov entropy | judge disagree | pdvar ∩var |
 |---|---|---|---|---|---|---|---|
-| agentic_txc_10_bare | 1 | A | 24/32 | 3/3 | 0.96 | 0.03 | 17 |
-| agentic_txc_10_bare | 1 | B | 21/32 | 4/4 | 1.27 | 0.06 | 22 |
-| agentic_txc_10_bare | 1 | random | 2/32 | 6/7 | 1.63 | 0.16 | 19 |
-| agentic_txc_10_bare | 2 | A | 19/32 | 3/3 | 0.96 | 0.03 | 19 |
-| agentic_txc_10_bare | 2 | B | 13/32 | 4/4 | 1.36 | 0.03 | 25 |
-| agentic_txc_10_bare | 2 | random | 2/32 | 7/7 | 1.78 | 0.25 | 22 |
+| agentic_txc_10_bare | 42 | A | 21/32 | 3/3 | 1.01 | 0.00 | 20 |
+| agentic_txc_10_bare | 42 | B | 19/32 | 4/4 | 1.34 | 0.06 | 24 |
+| agentic_txc_10_bare | 42 | random | 6/32 | 7/7 | 1.55 | 0.25 | 19 |
+| agentic_txc_12_bare_batchtopk | 42 | A | 23/32 | 3/3 | 1.02 | 0.03 | 15 |
+| agentic_txc_12_bare_batchtopk | 42 | B | 17/32 | 4/4 | 1.33 | 0.06 | 24 |
+| agentic_txc_12_bare_batchtopk | 42 | random | 2/32 | 7/7 | 1.65 | 0.16 | 19 |
 
 ## Means ± stderr across seeds (per arch × concat)
 
 | arch | concat | n_seeds | mean /N sem | mean cov k/P | mean cov entropy |
 |---|---|---|---|---|---|
-| agentic_txc_10_bare | A | 2 | 21.5 ± 2.50 | 3.0 ± 0.00 | 0.96 ± 0.00 |
-| agentic_txc_10_bare | B | 2 | 17.0 ± 4.00 | 4.0 ± 0.00 | 1.32 ± 0.05 |
-| agentic_txc_10_bare | random | 2 | 2.0 ± 0.00 | 6.5 ± 0.50 | 1.71 ± 0.08 |
+| agentic_txc_10_bare | A | 1 | 21.0 ± 0.00 | 3.0 ± 0.00 | 1.01 ± 0.00 |
+| agentic_txc_10_bare | B | 1 | 19.0 ± 0.00 | 4.0 ± 0.00 | 1.34 ± 0.00 |
+| agentic_txc_10_bare | random | 1 | 6.0 ± 0.00 | 7.0 ± 0.00 | 1.55 ± 0.00 |
+| agentic_txc_12_bare_batchtopk | A | 1 | 23.0 ± 0.00 | 3.0 ± 0.00 | 1.02 ± 0.00 |
+| agentic_txc_12_bare_batchtopk | B | 1 | 17.0 ± 0.00 | 4.0 ± 0.00 | 1.33 ± 0.00 |
+| agentic_txc_12_bare_batchtopk | random | 1 | 2.0 ± 0.00 | 7.0 ± 0.00 | 1.65 ± 0.00 |