Neural data cache

.gitignore

@@ -37,9 +37,9 @@ Makefile
 # Machine-local runtime files
 /runtime/
 
-# Search checkpoints, archive deltas, manifests, and per-run metrics under
-# results/search are intentionally tracked so disjoint ranges merge through Git.
-# Only shared derived views, logs, and interrupted temporary files are ignored.
+# Geometry checkpoints, MAP-Elites deltas, manifests, and per-run metrics under
+# results/search remain trackable so disjoint topology ranges can merge through Git.
+# Shared derived views, logs, and interrupted temporary files stay local.
 /results/search/leaderboard.tsv
 /results/search/run.log
 /results/search/runs/**/run.log
@@ -47,6 +47,17 @@ Makefile
 /results/search/**/*.tmp.*
 /results/search/**/*.partial
 
+# Local ML dataset and derived neural cache. These files can grow to 200 GB,
+# are machine-local, and must never be committed. Geometry checkpoints (*.szcp)
+# and mergeable search deltas (*.szar) are intentionally not ignored here.
+/results/search/**/training/
+/results/search/**/neural/
+/results/search/runs/**/TRAINING_DATA_LIMIT_REACHED_REWRITE_REQUIRED.tsv
+*.sztd
+*.sztd.*
+*.szonn
+*.szonn.*
+
 # Python cache and virtual environments
 __pycache__/
 *.py[cod]