fix: --json empty-results contract, accent-insensitive company matching

applyr/commands/analytics.py

@@ -96,7 +96,10 @@ def cmd_pipeline(min_score: int = 0, as_json: bool = False) -> None:
         conn.close()
 
     if not rows:
-        print("No offers in the database.")
+        if as_json:
+            print(json.dumps({status: [] for status in _STATUS_ORDER}, indent=2, ensure_ascii=False))
+        else:
+            print("No offers in the database.")
         return
 
     # Group by status
@@ -294,7 +297,10 @@ def cmd_gaps(limit: int = 10, as_json: bool = False) -> None:
     rows = _live_skill_gaps(limit)
 
     if not rows:
-        print("No skill gaps recorded yet.")
+        if as_json:
+            print(json.dumps([], indent=2, ensure_ascii=False))
+        else:
+            print("No skill gaps recorded yet.")
         return
 
     worst_gap = max(r["total_gap"] for r in rows)
@@ -438,7 +444,10 @@ def cmd_trends(period: str = "week", as_json: bool = False) -> None:
         conn.close()
 
     if not rows:
-        print("No dated offers found.")
+        if as_json:
+            print(json.dumps([], indent=2, ensure_ascii=False))
+        else:
+            print("No dated offers found.")
         return
 
     if as_json:
@@ -660,7 +669,10 @@ def cmd_plan(limit: int = 10, as_json: bool = False) -> None:
     rows = _live_skill_gaps()
 
     if not rows:
-        print("No skill gaps recorded yet.")
+        if as_json:
+            print(json.dumps([], indent=2, ensure_ascii=False))
+        else:
+            print("No skill gaps recorded yet.")
         return
 
     # `total_gap` is already frequency times average gap — the points a topic
@@ -724,7 +736,10 @@ def cmd_salary(seniority: str | None = None, category: str | None = None, as_jso
         conn.close()
 
     if not rows:
-        print("No salary data available.")
+        if as_json:
+            print(json.dumps({"by_seniority": [], "by_category": []}, indent=2, ensure_ascii=False))
+        else:
+            print("No salary data available.")
         return
 
     # Group by seniority

applyr/commands/core.py

@@ -785,10 +785,13 @@ def cmd_list(status_filter: str | None = None, sort_by: str = "date_applied", li
         conn.close()
 
     if not rows:
-        msg = f"No offers found"
-        if status_filter:
-            msg += f" with status '{status_filter}'"
-        print(msg + ".")
+        if as_json:
+            print(json.dumps([], indent=2, ensure_ascii=False))
+        else:
+            msg = f"No offers found"
+            if status_filter:
+                msg += f" with status '{status_filter}'"
+            print(msg + ".")
         return
 
     if as_json:
@@ -1125,7 +1128,10 @@ def cmd_search(keyword: str, status_filter: str | None = None, company: str | No
     target = f"company '{company}'" if company else f"'{keyword}'"
 
     if not rows:
-        print(f"No offers found matching {target}.")
+        if as_json:
+            print(json.dumps([], indent=2, ensure_ascii=False))
+        else:
+            print(f"No offers found matching {target}.")
         return
 
     if as_json:

applyr/duplicates.py

@@ -13,6 +13,7 @@
 
 import re
 import sqlite3
+import unicodedata
 from difflib import SequenceMatcher
 
 from applyr.constants import DUPLICATE_SIMILARITY_THRESHOLD
@@ -58,13 +59,37 @@ def title_similarity(a: str, b: str) -> float:
     return SequenceMatcher(None, norm_a, norm_b).ratio()
 
 
+def normalize_company(text: str | None) -> str:
+    """Lowercase and strip diacritics, so "Mática" and "Matica" compare equal.
+
+    Company names get typed inconsistently far more often than two
+    genuinely different companies happen to share a name — a posting
+    pasted from LinkedIn, OCR, or a rushed manual entry drops accents far
+    more often than it invents a real collision. Unlike substring or fuzzy
+    matching (rejected for company names — see module docstring), stripping
+    diacritics never merges two different companies together, so it carries
+    none of that false-positive risk.
+    """
+    if not text:
+        return ""
+    decomposed = unicodedata.normalize("NFKD", text)
+    return "".join(c for c in decomposed if not unicodedata.combining(c)).lower()
+
+
+def _register_company_normalizer(conn: sqlite3.Connection) -> None:
+    """Expose normalize_company() to SQL as unaccent_lower(). Idempotent —
+    re-registering the same function on a connection is safe and cheap."""
+    conn.create_function("unaccent_lower", 1, normalize_company)
+
+
 def find_exact(conn: sqlite3.Connection, title: str, company: str | None) -> sqlite3.Row | None:
-    """Find an offer with the same title and company, case-insensitively."""
+    """Find an offer with the same title and company, case/accent-insensitively."""
+    _register_company_normalizer(conn)
     return conn.execute(
         """SELECT id, title, status, date_received, compatibility_pct
            FROM offers
            WHERE LOWER(title) = LOWER(?)
-             AND LOWER(COALESCE(company,'')) = LOWER(COALESCE(?,''))""",
+             AND unaccent_lower(COALESCE(company,'')) = unaccent_lower(COALESCE(?,''))""",
         (title, company),
     ).fetchone()
 
@@ -73,10 +98,11 @@ def find_company_offers(conn: sqlite3.Connection, company: str | None) -> list[s
     """All offers already recorded for a company, newest first."""
     if not company:
         return []
+    _register_company_normalizer(conn)
     return conn.execute(
         """SELECT id, title, status, date_received, compatibility_pct
            FROM offers
-           WHERE LOWER(COALESCE(company,'')) = LOWER(?)
+           WHERE unaccent_lower(COALESCE(company,'')) = unaccent_lower(?)
            ORDER BY id DESC""",
         (company,),
     ).fetchall()

tests/test_cli_routing.py

@@ -438,13 +438,9 @@ def test_json_no_color_combined(self, run_cli, capsys, tmp_db):
     def test_json_flag_gaps(self, run_cli, capsys, tmp_db):
         out, err, code = _run(run_cli, capsys, ["--json", "gaps"])
         assert code == 0
-        # With empty DB, gaps may print a message or empty JSON
-        if out.strip():
-            try:
-                data = json.loads(out)
-                assert isinstance(data, dict)
-            except json.JSONDecodeError:
-                pass  # Some commands print messages instead of JSON when empty
+        # cmd_gaps's JSON payload is a list of skill-gap entries — empty DB is []
+        data = json.loads(out)
+        assert isinstance(data, list)
 
     def test_json_flag_followups(self, run_cli, capsys, tmp_db):
         out, err, code = _run(run_cli, capsys, ["--json", "followups"])
@@ -479,12 +475,9 @@ def test_json_flag_summary(self, run_cli, capsys, tmp_db):
     def test_json_flag_trends(self, run_cli, capsys, tmp_db):
         out, err, code = _run(run_cli, capsys, ["--json", "trends"])
         assert code == 0
-        if out.strip():
-            try:
-                data = json.loads(out)
-                assert isinstance(data, dict)
-            except json.JSONDecodeError:
-                pass
+        # cmd_trends's JSON payload is a list of period entries — empty DB is []
+        data = json.loads(out)
+        assert isinstance(data, list)
 
     def test_json_flag_doctor(self, run_cli, capsys, tmp_db):
         out, err, code = _run(run_cli, capsys, ["--json", "doctor"])

tests/test_duplicates.py

@@ -3,7 +3,14 @@
 import pytest
 
 from applyr.constants import DUPLICATE_SIMILARITY_THRESHOLD
-from applyr.duplicates import find_similar, normalize_title, title_similarity
+from applyr.duplicates import (
+    find_company_offers,
+    find_exact,
+    find_similar,
+    normalize_company,
+    normalize_title,
+    title_similarity,
+)
 
 
 class TestNormalizeTitle:
@@ -98,3 +105,50 @@ def test_respects_custom_threshold(self):
     def test_common_posting_variants_are_caught(self, variant):
         rows = [FakeRow(id=1, title=variant)]
         assert find_similar(rows, "Backend Engineer") is not None
+
+
+class TestNormalizeCompany:
+    def test_lowercases(self):
+        assert normalize_company("Acme Corp") == "acme corp"
+
+    def test_strips_diacritics(self):
+        assert normalize_company("Mática Partners") == "matica partners"
+
+    def test_already_unaccented_is_unchanged_besides_case(self):
+        assert normalize_company("Matica Partners") == "matica partners"
+
+    def test_none_becomes_empty_string(self):
+        assert normalize_company(None) == ""
+
+
+class TestFindExactAndFindCompanyOffers:
+    """Found via a real offer: "Mática Partners" and "Matica Partners" were
+    treated as two different companies because the SQL comparison only
+    lowercased, never stripped accents — so add's duplicate warning and
+    search --company silently missed real history at the same company.
+    """
+
+    @pytest.fixture
+    def conn(self, tmp_db):
+        from applyr.db import get_conn
+
+        connection = get_conn(tmp_db)
+        connection.execute(
+            "INSERT INTO offers (title, company) VALUES (?, ?)",
+            ("IA Engineer", "Matica Partners"),
+        )
+        connection.commit()
+        return connection
+
+    def test_find_company_offers_ignores_accent_differences(self, conn):
+        rows = find_company_offers(conn, "Mática Partners")
+        assert len(rows) == 1
+        assert rows[0]["title"] == "IA Engineer"
+
+    def test_find_exact_ignores_accent_differences(self, conn):
+        row = find_exact(conn, "IA Engineer", "Mática Partners")
+        assert row is not None
+
+    def test_find_exact_still_requires_the_same_title(self, conn):
+        row = find_exact(conn, "Backend Developer", "Mática Partners")
+        assert row is None

tests/test_empty_results_json.py

@@ -0,0 +1,61 @@
+"""--json must stay parseable even when there is nothing to report.
+
+Found while auditing applyr end to end: every list-shaped command printed a
+human sentence ("No offers found.", "No skill gaps recorded yet.", ...) on
+an empty result set regardless of --json, breaking the one contract
+AGENT_INSTRUCTIONS.md promises agent callers ("--json en todos los
+comandos"). cmd_gaps_list already did this right (`{"total": 0, "gaps": []}`
+even when empty); the rest are brought in line with it here.
+"""
+
+import json
+
+import pytest
+
+from applyr.commands import (
+    cmd_gaps,
+    cmd_list,
+    cmd_pipeline,
+    cmd_plan,
+    cmd_salary,
+    cmd_search,
+    cmd_trends,
+)
+
+
+@pytest.mark.unit
+class TestEmptyResultsStayJson:
+
+    def test_list(self, tmp_db, tmp_applyr, capsys):
+        cmd_list(as_json=True)
+        assert json.loads(capsys.readouterr().out) == []
+
+    def test_search(self, tmp_db, tmp_applyr, capsys):
+        cmd_search("nothing-matches-this", as_json=True)
+        assert json.loads(capsys.readouterr().out) == []
+
+    def test_search_company(self, tmp_db, tmp_applyr, capsys):
+        cmd_search("", company="Nobody Inc", as_json=True)
+        assert json.loads(capsys.readouterr().out) == []
+
+    def test_pipeline(self, tmp_db, tmp_applyr, capsys):
+        cmd_pipeline(as_json=True)
+        payload = json.loads(capsys.readouterr().out)
+        assert isinstance(payload, dict)
+        assert all(v == [] for v in payload.values())
+
+    def test_gaps(self, tmp_db, tmp_applyr, capsys):
+        cmd_gaps(as_json=True)
+        assert json.loads(capsys.readouterr().out) == []
+
+    def test_trends(self, tmp_db, tmp_applyr, capsys):
+        cmd_trends(as_json=True)
+        assert json.loads(capsys.readouterr().out) == []
+
+    def test_plan(self, tmp_db, tmp_applyr, capsys):
+        cmd_plan(as_json=True)
+        assert json.loads(capsys.readouterr().out) == []
+
+    def test_salary(self, tmp_db, tmp_applyr, capsys):
+        cmd_salary(as_json=True)
+        assert json.loads(capsys.readouterr().out) == {"by_seniority": [], "by_category": []}