Add EXTRACT_IGNORE list for labels that aren't food or events

Grade ranges like K-8 are always dropped; other labels (such as BHS) are now listed in EXTRACT_IGNORE instead of being built in. Matches whole entries in any case, in menu items and event titles.

phase0/extract.py

@@ -9,6 +9,7 @@ Settings (environment variables):
   EXTRACT_THINK   1 to let an Ollama model reason before answering           (default 0)
   EXTRACT_TEMPERATURE  Ollama sampling temperature                           (default 0)
   EXTRACT_DOUBLE_READ  1 to read every page twice and flag where the reads disagree (default 0)
+  EXTRACT_IGNORE  comma-separated labels to drop from menu items and event titles, e.g. "BHS,Milk"   (default none)
   ANTHROPIC_API_KEY  only for claude-* models
 """
 import base64
@@ -38,6 +39,7 @@ OLLAMA_URL = os.environ.get("OLLAMA_URL", "http://localhost:11434").rstrip("/")
 THINK = os.environ.get("EXTRACT_THINK") == "1"
 TEMPERATURE = float(os.environ.get("EXTRACT_TEMPERATURE", "0"))
 DOUBLE_READ = os.environ.get("EXTRACT_DOUBLE_READ") == "1"
+IGNORE = {w.strip().lower() for w in os.environ.get("EXTRACT_IGNORE", "").split(",") if w.strip()}
 EFFORT = "high"  # Claude only
 LONG_EDGE = 2000  # ponytail: rotate + downscale only; add deskew/crop when angled photos score worse than flat ones
 LOW_CONFIDENCE = 0.7
@@ -153,7 +155,13 @@ Menus
 `confidence` is 0 to 1. Go below 0.7 whenever you had to guess."""
 
 GRID_LAYOUTS = ("month_grid", "menu")
-LABEL = re.compile(r"(k|\d{1,2})\s*-\s*\d{1,2}|bhs", re.I)  # ponytail: school-specific labels belong in the source's prompt notes
+GRADE_RANGE = re.compile(r"(k|\d{1,2})\s*-\s*\d{1,2}", re.I)  # "K-8", "6-12": never food, never an event
+
+
+def ignored(text, words):
+    """A grade range, or one of the labels the user listed (whole text, any case)."""
+    text = (text or "").strip()
+    return bool(GRADE_RANGE.fullmatch(text)) or text.lower() in words
 
 
 def load_image(path):
@@ -184,15 +192,18 @@ def iso(s):
         return s
 
 
-def cleanup(result):
-    """Deterministic tidy-up before validation: pad dates, drop events with no real title."""
+def cleanup(result, ignore=None):
+    """Deterministic tidy-up before validation: pad dates, drop events with no real title,
+    and drop labels (grade ranges and the ignore list) from event titles and menu items."""
+    ignore = IGNORE if ignore is None else ignore
     for item in [result["date_range"]] + result["events"] + result["menu_items"]:
         for key in ("date", "end_date", "start", "end"):
             if item.get(key):
                 item[key] = iso(item[key])
     result["events"] = [e for e in result["events"] if re.sub(r"[^a-z]", "", e["title"].lower()) not in ("", "event")]
+    result["events"] = [e for e in result["events"] if not ignored(e["title"], ignore)]
     for m in result["menu_items"]:
-        m["items"] = [i for i in m["items"] if not LABEL.fullmatch(i.strip())]
+        m["items"] = [i for i in m["items"] if not ignored(i, ignore)]
     return result
 
 
@@ -490,7 +501,14 @@ def selftest():
     # Year boundary: December row ending in January
     row = [col(wd, d) for wd, d in zip(sun_first, [27, 28, 29, 30, 31, 1, 2])]
     assert first_column_date(row, 2026, 12) == (date(2026, 12, 27), 0)
-    assert [i for i in ["K-8", "Grapes", "BHS", "6-12", "Corn"] if not LABEL.fullmatch(i)] == ["Grapes", "Corn"]
+    # Labels: grade ranges always go; other words only when listed, matched whole and in any case
+    items = ["K-8", "Grapes", "BHS", "6-12", "Corn", "bhs ", "BHS Pizza"]
+    assert [i for i in items if not ignored(i, set())] == ["Grapes", "BHS", "Corn", "bhs ", "BHS Pizza"]
+    assert [i for i in items if not ignored(i, {"bhs"})] == ["Grapes", "Corn", "BHS Pizza"]
+    r = cleanup({"date_range": {}, "menu_items": [{"date": "2026-11-04", "items": ["K-8", "Grapes", "BHS"]}],
+                 "events": [{"date": "2026-11-04", "title": "BHS", "kind": "event"},
+                            {"date": "2026-11-04", "title": "No School", "kind": "no_school"}]}, {"bhs"})
+    assert r["menu_items"][0]["items"] == ["Grapes"] and [e["title"] for e in r["events"]] == ["No School"], r
     # Double read: agreement adds nothing, a moved date and a one-sided item are flagged
     ev = lambda d, t, **kw: {"date": d, "end_date": None, "all_day": True, "start_time": None,
                              "end_time": None, "kind": "event", "title": t, **kw}