figure: visualize endpoint tracking failure

3 files changed, 9 insertions, 0 deletions

diff --git a/results/figs/figureS_tracking_failure.pdf b/results/figs/figureS_tracking_failure.pdf
new file mode 100644
index 0000000..6943f67
--- /dev/null
+++ b/results/figs/figureS_tracking_failure.pdf
diff --git a/results/figs/figureS_tracking_failure.png b/results/figs/figureS_tracking_failure.png
new file mode 100644
index 0000000..f5344e7
--- /dev/null
+++ b/results/figs/figureS_tracking_failure.png
diff --git a/results/figs/figureS_tracking_failure_caption.md b/results/figs/figureS_tracking_failure_caption.md
new file mode 100644
index 0000000..8a9a500
--- /dev/null
+++ b/results/figs/figureS_tracking_failure_caption.md
@@ -0,0 +1,9 @@
+**Supplementary Figure — Endpoint alignment is not a trajectory certificate.**
+**a,** The sole frozen RRM-3 run becomes unstable before the first learning-rate
+drop: mean training loss peaks at `1.5808e16` in epoch 92. Lower rates reduce
+the loss to `8.9503e13` by epoch 200 but do not restore the classifier.
+**b,** The final snapshot is deceptively reassuring: early/all-layer teaching
+alignment is `0.877/0.849`, feedback/forward cosine is `0.999998`, and feedback
+norms are controlled, yet validation accuracy is 10% and validation loss is
+NaN. BP accuracy (`91.62%`) and unit ideal references are shown in gray. This
+is a failed inherited residual-mirror baseline, not positive SDIL evidence.