@meshxdata/fops 0.1.32 → 0.1.35

package/src/plugins/bundled/fops-plugin-file/lib/embed-index.js

@@ -387,7 +387,14 @@ export async function runEmbedIndex(api, opts) {
           store.setMeta("embedding_dim", "384");
         } catch {
           // Embeddings failed — index still usable without them
+          spinner.stop();
+          console.log(WARN("  ⚠ Column embeddings skipped — MiniLM model not ready."));
+          console.log(DIM("    Run `fops embed index` first to download the model, then re-run `fops embed file index`."));
+          console.log(DIM("    Without embeddings, matching falls back to Jaccard (exact column name overlap only)."));
         }
+      } else {
+        console.log(WARN("  ⚠ Column embeddings skipped — embeddings plugin not available."));
+        console.log(DIM("    Run `fops embed index` first to download the MiniLM model, then re-run `fops embed file index`."));
       }
 
       store.setMeta("version", "3");

package/src/plugins/bundled/fops-plugin-file/lib/match.js

@@ -155,13 +155,20 @@ export async function embedSchemaEntries(entries, embeddingClient, opts = {}) {
 
 /**
  * Normalize a column name for embedding: split on separators, lowercase.
+ * For compound names (e.g. order_status), the suffix is repeated to boost its
+ * weight in the embedding so "order_status" stays close to "status" and doesn't
+ * drift toward "order_id" due to a shared prefix.
  */
 function normalizeColumnName(name) {
-  return (name || "")
-    .replace(/[_\-./]/g, " ")
+  const parts = (name || "")
     .replace(/([a-z])([A-Z])/g, "$1 $2") // camelCase split
-    .toLowerCase()
-    .trim() || name;
+    .split(/[_\-./\s]+/)
+    .filter(Boolean)
+    .map((p) => p.toLowerCase());
+  if (parts.length === 0) return name || "";
+  if (parts.length === 1) return parts[0];
+  // Repeat the last token to emphasise the semantic role over the entity prefix
+  return parts.join(" ") + " " + parts[parts.length - 1];
 }
 
 /**