results: close shared-feedback feasibility screen

1 files changed, 19 insertions, 0 deletions

diff --git a/RESULTS.md b/RESULTS.md
index 1fd7b0b..6da95a0 100644
--- a/RESULTS.md
+++ b/RESULTS.md
@@ -291,6 +291,25 @@ soma-conditioned R2 and positive gradient alignment do not guarantee that the re
 innovation is purely instructional. This narrows the algorithmic claim and motivates an explicit
 model of context, rather than treating every somatically unpredictable apical component as error.
 
+## Endogenous shared-feedback feasibility (validation only)
+
+To test whether necessary ordinary feedback could make innovation responsible for scaling rather
+than merely robust to injected traffic, a frozen contextual task allowed the task context to enter
+the network only through the same apical pathway that later carried reciprocal KP instruction.
+There was no nuisance, traffic ratio, measurement offset, or generated bias. The four conditions
+shared their complete forward computation; oracle KP alone separately observed the instructional
+component. Source revision `b345784` was committed before the sole endpoint.
+
+The architecture check succeeded but the proposed learning failure did not occur. Oracle reached
+`99.316%` validation accuracy and lost `22.998` points when context was removed, proving that the
+ordinary apical component was required for the task. Innovation's neutral predictor explained
+`95.06%` of mean per-cell context variance, left at most `1.553%` context RMS, and saw zero task
+instructions. Nevertheless, raw shared KP reached `99.023%`, norm-matched raw reached `99.170%`,
+and innovation reached `99.219%`. The raw--oracle gap was only `0.293` points and the
+innovation--raw gain only `0.195` points, rather than the frozen five-point margins. **The S0 gate
+fails and opens no depth experiment.** Necessary, predictable shared activity is therefore not by
+itself sufficient to create a scaling obstruction; the realization is closed without tuning.
+
 ## Legacy accuracy (pre-audit implementation)
 | task                         | BP        | DFA   | SDIL          |
 |------------------------------|-----------|-------|---------------|