Extend perturbation audit to vanilla early-epoch checkpoints

Cross-metric disambiguation confirmation. Vanilla DFA at ep 1
(meaningful regime, ||g||~6e-7) deep rho across 3 seeds:

  s42:  deep rho -0.008
  s123: deep rho +0.000
  s456: deep rho -0.000
  mean: -0.003 ± 0.005

Compare to penalized DFA 3-seed: deep rho +0.080 ± 0.011.

The disambiguation (penalty CREATES alignment, not just reveals it) is
now confirmed by TWO independent metrics:
  - cos: vanilla -0.008 ± 0.013, penalized +0.155 ± 0.025
  - rho: vanilla -0.003 ± 0.005, penalized +0.080 ± 0.011

Both metrics agree on the vanilla→penalized transition. The l0 (embedding)
rho is high (~0.25-0.29) at every vanilla checkpoint, mirroring the cos
l0 +0.42 — the embedding layer is genuinely useful while the deep blocks
are not, by BOTH metrics. The penalty restores some deep usefulness to
~+0.08 rho / +0.16 cos.

Cross-metric agreement rules out single-metric artifacts on either side.

1 files changed, 6 insertions, 0 deletions

diff --git a/experiments/perturbation_correlation_audit.py b/experiments/perturbation_correlation_audit.py
index cba84ea..163d3a8 100644
--- a/experiments/perturbation_correlation_audit.py
+++ b/experiments/perturbation_correlation_audit.py
@@ -127,6 +127,12 @@ def main():
          "results/dfa_pen_short/dfa_pen_lam0.01_s123.pt", 123),
         ("penalized DFA s456 lam=1e-2 30ep",
          "results/dfa_pen_short/dfa_pen_lam0.01_s456.pt", 456),
+        ("vanilla DFA s42 ep1 (meaningful regime)",
+         "results/vanilla_dfa_early_ckpts/vanilla_dfa_s42_ep1.pt", 42),
+        ("vanilla DFA s123 ep1 (meaningful regime)",
+         "results/vanilla_dfa_early_ckpts/vanilla_dfa_s123_ep1.pt", 123),
+        ("vanilla DFA s456 ep1 (meaningful regime)",
+         "results/vanilla_dfa_early_ckpts/vanilla_dfa_s456_ep1.pt", 456),
     ]
 
     print("=" * 76)