Update static/schema_help.md

static/schema_help.md

@@ -1,31 +1,26 @@
-# Schema Help
+MindsEye Dataset Configs (Important)
 
-This dataset ships as two configs:
+If you have:
 
-## 1) nodes
-`data/raw/nodes.csv`
+nodes.csv (Prompt Evolution Nodes — immutable cognitive states.)
 
-Represents immutable Prompt Evolution Tree (PET) states:
-- node identity
-- prompt/template content
-- parent lineage
-- metadata tags
-- timestamps
-- provenance links
+runs.csv (Execution Records — each reasoning run.)
 
-## 2) runs
-`data/raw/runs.csv`
+They SHOULD NOT be in the same dataset builder config unless they share identical columns.
 
-Represents execution records (Gemini runs):
-- run identity
-- node_id link
-- model + token usage
-- outputs + result_json
-- score + status
-- timestamps + triggering source
+Correct options:
+Option A (recommended): Two configs
 
-## Why configs matter
-Hugging Face’s CSV dataset builder requires:
-**all CSV files in the same config/split must share identical columns**.
+config: nodes -> data/raw/nodes.csv
 
-Nodes and runs intentionally have different schemas, so they MUST be split into separate configs.
+config: runs -> data/raw/runs.csv
+
+Option B: Two splits
+
+split: nodes -> nodes.csv
+
+split: runs -> runs.csv
+
+If HF says:
+“All the data files must have the same columns…”
+it means it thinks multiple CSVs belong to the same table.