summaryrefslogtreecommitdiff
path: root/THEORY.md
diff options
context:
space:
mode:
Diffstat (limited to 'THEORY.md')
-rw-r--r--THEORY.md6
1 files changed, 5 insertions, 1 deletions
diff --git a/THEORY.md b/THEORY.md
index 1c4f5d0..14c585f 100644
--- a/THEORY.md
+++ b/THEORY.md
@@ -247,7 +247,11 @@ The executable audit checks the complete BatchNorm-coupled antithetic JVP to
predicted Q/R moments equal the exact causal delta-rule targets to `2.3e-16`
relative error. Actual training never performs that copy. These identities
establish mechanics only; a frozen-forward causal-capture gate must pass before
-any task-scale endpoint is opened.
+any task-scale endpoint is opened. That empirical V4 gate subsequently failed:
+the stable rate left early alignment near zero, while larger rates caused a
+58.5x feedback norm ratio or nonfinite state. The unbiased identity therefore
+does not imply usable finite-budget signal in the 267,904-dimensional joint
+parameter space.
## 2. What is sufficient for a descent step