Update app.py

app.py

@@ -1,7 +1,7 @@
 # ================================================================
 # Self-Sensing Concrete Assistant — Predictor (XGB) + Hybrid RAG
-# - Easier conditions: no completion gate; always attempt prediction
-# - Stable categoricals "NA"; feature alignment to model
+# - Predictor tab: required fields marked with *
+# - Prediction fixed: NA is accepted for required categoricals
 # - RAG page-extraction regex fixed for [[PAGE=...]]
 # ================================================================
 
@@ -106,7 +106,14 @@ CATEGORICAL_COLS = {
     "Current Type"
 }
 
-# (Labels with * remain; UI unchanged)
+OPTIONAL_FIELDS = {
+    "Filler 2 Type",
+    "Filler 2 Diameter (µm)",
+    "Filler 2 Length (mm)",
+    "Filler 2 Dimensionality",
+}
+
+# Required fields (as you specified earlier)
 REQUIRED_FIELDS = {
     "Filler 1 Type",
     "Filler 1 Diameter (µm)",
@@ -155,21 +162,18 @@ def _canon_cat(v: Any) -> str:
         return CANON_NA
     return s
 
-def _to_float_or_nan(v):
-    if v in ("", None):
-        return np.nan
-    try:
-        # allow "1,234.5" by stripping commas
-        return float(str(v).replace(",", ""))
-    except Exception:
-        return np.nan
-
 def _coerce_to_row(form_dict: dict) -> pd.DataFrame:
     row = {}
     for col in MAIN_VARIABLES:
         v = form_dict.get(col, None)
         if col in NUMERIC_COLS:
-            row[col] = _to_float_or_nan(v)
+            if v in ("", None):
+                row[col] = np.nan
+            else:
+                try:
+                    row[col] = float(v)
+                except Exception:
+                    row[col] = np.nan
         elif col in CATEGORICAL_COLS:
             row[col] = _canon_cat(v)
         else:
@@ -177,12 +181,33 @@ def _coerce_to_row(form_dict: dict) -> pd.DataFrame:
             row[col] = s if s else CANON_NA
     return pd.DataFrame([row], columns=MAIN_VARIABLES)
 
+def _is_complete(form_dict: dict) -> bool:
+    """
+    FIX: For required *categoricals*, NA counts as 'provided' (acceptable),
+    so users aren't blocked when NA is a legitimate choice.
+    Numeric required fields must be non-NaN.
+    """
+    for col in REQUIRED_FIELDS:
+        v = form_dict.get(col, None)
+        if col in NUMERIC_COLS:
+            try:
+                if v in ("", None) or (isinstance(v, float) and np.isnan(v)):
+                    return False
+            except Exception:
+                return False
+        else:
+            # Required categoricals/text: accept any non-empty after canonicalization,
+            # and accept CANON_NA as "provided".
+            s = _canon_cat(v)
+            if s == "" or s is None:
+                return False
+    return True
+
 def _align_columns_to_model(df: pd.DataFrame, mdl) -> pd.DataFrame:
     """Align incoming dataframe columns to the model's expected feature order."""
     try:
         feat = getattr(mdl, "feature_names_in_", None)
         if feat is not None and len(feat) > 0:
-            # add any missing columns as NaN, keep extras (model will ignore via transformer)
             for c in feat:
                 if c not in df.columns:
                     df[c] = np.nan
@@ -193,13 +218,9 @@ def _align_columns_to_model(df: pd.DataFrame, mdl) -> pd.DataFrame:
     return df
 
 def predict_fn(**kwargs):
-    """
-    SUPER-LENIENT: Always attempt prediction.
-    - No completeness checks
-    - Missing numerics -> NaN
-    - Categoricals -> canonical 'NA'
-    - If model missing or inference error -> return 0.0 (keeps UI stable)
-    """
+    # Keep your contract: 0.0 if incomplete or on error
+    if not _is_complete(kwargs):
+        return 0.0
     mdl = _load_model_or_error()
     if mdl is None:
         return 0.0
@@ -212,7 +233,9 @@ def predict_fn(**kwargs):
         else:
             y = y_raw
         y = float(np.asarray(y).ravel()[0])
-        return 0.0 if y < 0 else y
+        if y < 0:
+            y = 0.0
+        return y
     except Exception as e:
         print(f"[Predict] {e}")
         traceback.print_exc()