Fix autofill: unify endpoint with filtered _autofill_values, empty-answer fallback, checkbox detection

app.py

@@ -77,26 +77,35 @@ def _autofill_values(uid, form_id):
     profile = store.get_profile(uid) if uid else {}
     about = profile.get("about_me", "")
     display = profile.get("display_name", "")
-    field_desc = ", ".join(f"{f['key']} ({f.get('label', '')})" for f in fm)
+    # filter to a manageable set: all text fields + checkboxes with a real label (capped)
+    text_fields = [f for f in fm if f.get("type") != "checkbox"]
+    cb_fields = [f for f in fm if f.get("type") == "checkbox" and f.get("label", "").strip()]
+    cb_fields = cb_fields[:40]
+    sel = text_fields + cb_fields
+    field_desc = ", ".join(
+        f"{f['key']} ({f.get('label', '')})" + (" [CB]" if f.get('type') == 'checkbox' else "")
+        for f in sel
+    )
     prompt = (
         "You are filling a Washington State court form. Map the user's facts to the form fields.\n"
         f"Form: {spec['name']}\n"
         f"Fields: {field_desc}\n\n"
         f"User's display name: {display or '(not set)'}\n"
         f"User's information (About Me):\n{about or '(not provided)'}\n\n"
-        "Output ONLY valid JSON like {\"county\": \"...\", \"petitioner\": \"...\"}. "
-        "For any field you cannot determine, output an empty string \"\". "
+        "Output ONLY valid JSON. For text fields, output the value (or empty string if unknown). "
+        "For CHECKBOX fields, output 'X' ONLY when the user's facts clearly and unambiguously "
+        "match that option; otherwise output ''. Be conservative — do not guess. "
         "Petitioner is the user themself (use the display name); respondent is their spouse."
     )
     try:
         resp = _ollama_chat(RAG_MODEL, [{"role": "user", "content": prompt}],
-                            temperature=0.1, num_predict=800)
+                            temperature=0.1, num_predict=1600)
         content = resp.get("message", {}).get("content", "")
         m = re.search(r"\{.*\}", content, re.DOTALL)
         vals = json.loads(m.group(0)) if m else {}
     except Exception:
         vals = {}
-    return {f["key"]: str(vals.get(f["key"], "") or "") for f in fm}
+    return {f["key"]: str(vals.get(f["key"], "") or "") for f in sel}
 
 
 def _save_pdf_to_vault(uid, path, name):
@@ -338,6 +347,13 @@ def _build_answer(agent, message, history, about_me, settings, uid, display_name
 
     final = resp.get("message", {})
     answer = (final.get("content") or final.get("thinking") or "").strip()
+    if not answer:
+        # transient empty response — fall back to the reliable model (no tools)
+        try:
+            resp = _ollama_chat(GENERAL_MODEL, messages)
+            answer = (resp.get("message", {}).get("content") or "").strip()
+        except Exception:
+            answer = ""
     answer = _strip_sources(answer)
     return {"answer": answer, "citations": citations, "grounded": bool(context_blocks)}
 
@@ -547,34 +563,10 @@ def api_forms_autofill():
     uid = _uid()
     d = request.get_json(force=True, silent=True) or {}
     form_id = d.get("form_id", "")
-    spec = forms.FORM_NAMES.get(form_id)
-    fm = forms.FIELDMAP.get(form_id, [])
-    if not spec:
+    if not forms.FORM_NAMES.get(form_id):
         return jsonify({"error": "unknown form"}), 400
-    profile = store.get_profile(uid) if uid else {}
-    about = profile.get("about_me", "")
-    display = profile.get("display_name", "")
-    field_desc = ", ".join(f"{f['key']} ({f.get('label','')})" for f in fm)
-    prompt = (
-        "You are filling a Washington State court form. Map the user's facts to the form fields.\n"
-        f"Form: {spec['name']}\n"
-        f"Fields: {field_desc}\n\n"
-        f"User's display name: {display or '(not set)'}\n"
-        f"User's information (About Me):\n{about or '(not provided)'}\n\n"
-        "Output ONLY valid JSON like {\"county\": \"...\", \"petitioner\": \"...\"}. "
-        "For any field you cannot determine, output an empty string. "
-        "Petitioner is the user themself (use the display name); respondent is their spouse."
-    )
-    try:
-        resp = _ollama_chat(RAG_MODEL, [{"role": "user", "content": prompt}],
-                            temperature=0.1, num_predict=800)
-        content = resp.get("message", {}).get("content", "")
-        m = re.search(r"\{.*\}", content, re.DOTALL)
-        vals = json.loads(m.group(0)) if m else {}
-    except Exception:
-        vals = {}
-    known = {f["key"]: str(vals.get(f["key"], "") or "") for f in fm}
-    return jsonify({"form_id": form_id, "values": known})
+    values = _autofill_values(uid, form_id)
+    return jsonify({"form_id": form_id, "values": values})
 
 
 @app.route("/api/forms/fill", methods=["POST"])

detect_fields.py

@@ -85,6 +85,30 @@ def detect_underscores(pno, page):
     return fields
 
 
+def detect_checkboxes(pno, page):
+    """Find checkbox glyphs (U+F0A8) and their labels. Type 'checkbox'."""
+    fields = []
+    words = page.get_text("words")
+    groups = {}
+    for w in words:
+        x0, y0, x1, y1, word = w[0], w[1], w[2], w[3], w[4]
+        key = int(y0 / 5) * 5
+        groups.setdefault(key, []).append((x0, x1, y0, word))
+    for y in sorted(groups):
+        ws = sorted(groups[y], key=lambda t: t[0])
+        for i, (x0, x1, y0, word) in enumerate(ws):
+            if "\uf0a8" in word:
+                label_words = []
+                for wx0, wx1, _, w in ws[i + 1:]:
+                    if "\uf0a8" in w:
+                        break
+                    label_words.append(w)
+                label = " ".join(label_words).strip()
+                fields.append({"page": pno, "x": x0, "y": y0, "type": "checkbox",
+                               "label": label})
+    return fields
+
+
 def main():
     all_fields = {}
     for fn in sorted(os.listdir(FORMS_DIR)):
@@ -99,9 +123,9 @@ def main():
             continue
         fields = []
         seen_keys = set()
-        for pno in range(min(len(d), 3)):  # caption + first 2 pages
-            lines = lines_with_pos(d[pno])
+        for pno in range(min(len(d), 12)):  # caption + all pages
             if pno == 0:
+                lines = lines_with_pos(d[pno])
                 for f in detect_caption(lines):
                     f["page"] = 0
                     if f["key"] not in seen_keys:
@@ -110,6 +134,9 @@ def main():
             for f in detect_underscores(pno, d[pno]):
                 f["key"] = "f_" + str(len(fields))
                 fields.append(f)
+            for f in detect_checkboxes(pno, d[pno]):
+                f["key"] = "cb_" + str(len(fields))
+                fields.append(f)
         d.close()
         all_fields[fid] = fields
     out = "/opt/astraea/forms/fieldmap.json"

forms.py

@@ -51,7 +51,8 @@ def list_forms():
     out = []
     for fid, spec in FORM_NAMES.items():
         fm = FIELDMAP.get(fid, [])
-        fields = [{"key": f["key"], "label": f.get("label", f["key"])} for f in fm]
+        fields = [{"key": f["key"], "label": f.get("label", f["key"]),
+                   "type": f.get("type", "text")} for f in fm]
         out.append({"id": fid, "name": spec["name"], "description": spec["description"],
                     "fields": fields})
     return out
@@ -72,10 +73,19 @@ def fill_form(form_id, values):
         f = fmap.get(key)
         if not f or not text:
             continue
+        page = doc[f.get("page", 0)]
+        if f.get("type") == "checkbox":
+            v = str(text).strip().lower()
+            if v in ("x", "true", "yes", "1", "checked", "check", "on"):
+                x, y = f["x"], f["y"]
+                page.draw_line(pymupdf.Point(x + 1, y + 1), pymupdf.Point(x + 9, y + 9),
+                               color=(0, 0, 0), width=1.3)
+                page.draw_line(pymupdf.Point(x + 9, y + 1), pymupdf.Point(x + 1, y + 9),
+                               color=(0, 0, 0), width=1.3)
+        else:
             text = str(text).strip()
             if not text:
                 continue
-        page = doc[f.get("page", 0)]
             page.insert_text((f["x"], f["y"]), text, fontsize=f.get("size", 11),
                              fontname="helv", color=(0, 0, 0))
     out = os.path.join("/tmp", f"filled_{form_id}_{uuid.uuid4().hex[:8]}.pdf")

templates/index.html

@@ -344,19 +344,25 @@
     }catch(e){ pv.textContent='Upload failed'; pv.style.color='#f87171'; }
   }
   const INTAKE=[
-    {label:"My name", q:"What's your full name?", p:"Jane Marie Doe"},
-    {label:"Spouse's name", q:"What's your spouse's full name?", p:"John Robert Doe"},
-    {label:"Married", q:"When and where did you get married?", p:"June 15, 2011, Seattle, WA"},
-    {label:"Separated", q:"When did you separate?", p:"June 2025"},
-    {label:"Children", q:"How many children do you share, and their names and ages?", p:"Two — Emma (12) and Liam (9)"},
+    {label:"My full legal name", q:"What's your full legal name (first, middle, last)?", p:"Jane Marie Doe"},
+    {label:"My mailing address", q:"What's your current mailing address (street, city, state, ZIP)?", p:"123 Main St, Everett, WA 98201"},
+    {label:"Spouse's full legal name", q:"What's your spouse's full legal name?", p:"John Robert Doe"},
+    {label:"Spouse's mailing address", q:"What's your spouse's mailing address?", p:"456 Oak Ave, Everett, WA 98201"},
+    {label:"Marriage date & place", q:"When and where did you get married (exact date + city/county/state)?", p:"June 15, 2011, Seattle, WA"},
+    {label:"Separation date", q:"When did you separate (exact date)?", p:"June 1, 2025"},
+    {label:"Case number", q:"Has a case been filed? What's the case number (if you know it)?", p:"Not yet / 24-3-01234-31"},
+    {label:"Children (names, ages, DOBs)", q:"List each child's full name, age, and date of birth.", p:"Emma Grace Doe, 12, born 01/15/2013; Liam James Doe, 9, born 03/22/2016"},
     {label:"Current custody", q:"Where do the children live now, and what's the current schedule?", p:"They live with me; dad has them every other weekend"},
-    {label:"Home", q:"Do you own a home or other real estate? Address, when bought, whose name is on the title/mortgage?", p:"House at 123 Main St, bought 2012, both names on title"},
+    {label:"Home / real estate", q:"Do you own a home or other real estate? Address, when bought, whose name is on the title/mortgage?", p:"House at 123 Main St, bought 2012, both names on title"},
     {label:"Assets", q:"What are your main assets? (bank accounts, retirement, vehicles, business, investments)", p:"Joint savings ~$20k, my 401k ~$85k, two cars"},
     {label:"Debts", q:"What are your main debts? (mortgage, credit cards, loans)", p:"Mortgage $280k remaining, $9k credit card, $12k car loan"},
-    {label:"Income", q:"What's your employment and income? And your spouse's?", p:"I make $72k/yr as a nurse; spouse makes $95k as an engineer"},
-    {label:"Case status", q:"Has either of you filed for divorce yet, and what does each side want?", p:"Spouse filed; wants full custody and to keep the house"},
-    {label:"Safety", q:"Is there any domestic violence, abuse, or safety concern? Any protection orders?", p:"No / Yes — details"},
-    {label:"Goals", q:"What outcome matters most to you? What are you most worried about?", p:"Keeping the kids with me and keeping the house"},
+    {label:"My income & employer", q:"What's your job, employer, and annual gross income?", p:"Registered nurse, Providence, $72k/yr"},
+    {label:"Spouse's income & employer", q:"What's your spouse's job, employer, and annual gross income?", p:"Software engineer, Amazon, $95k/yr"},
+    {label:"Children's health insurance", q:"Who provides health insurance for the children, and through which employer?", p:"My employer (Providence) covers the kids"},
+    {label:"Case status", q:"Has either of you filed for divorce yet, and what does each side want (custody, support, property)?", p:"Spouse filed; wants full custody and to keep the house"},
+    {label:"Safety / DV", q:"Is there any domestic violence, abuse, or safety concern? Any protection orders in place?", p:"No / Yes — details"},
+    {label:"Other cases", q:"Are there any other court cases involving you, your spouse, or the children (criminal, CPS, other family law)?", p:"None"},
+    {label:"Goals & worries", q:"What outcome matters most to you? What are you most worried about?", p:"Keeping the kids with me and keeping the house"},
   ];
   let intakeIdx=0; const intakeAnswers={};
   function openIntake(){ intakeIdx=0; for(const k in intakeAnswers) delete intakeAnswers[k]; closeModal('profileModal'); renderIntake(); document.getElementById('intakeModal').classList.add('open'); }