From c1958ef51a15be004ea42e4b293e9c150f183929 Mon Sep 17 00:00:00 2001
From: Adam <adam.gruber@heig-vd.ch>
Date: Tue, 29 Sep 2026 23:24:08 +0200
Subject: [PATCH] fix: count weekly buckets in days so Sunday rows land in
 their week

On Postgres, datediff(week) starts weeks on Sunday while buckets run Monday to Monday, so Sunday rows were counted in the next bucket and the last one got no bucket at all, which crashed the anomaly test result sort.

Fixes elementary-data/elementary#2370
---
 .../tests/test_column_anomalies.py            | 55 +++++++++++++++++++
 .../get_start_bucket_in_data.sql              |  5 ++
 2 files changed, 60 insertions(+)

diff --git a/integration_tests/tests/test_column_anomalies.py b/integration_tests/tests/test_column_anomalies.py
index af93e25c0..f78eb9cd8 100644
--- a/integration_tests/tests/test_column_anomalies.py
+++ b/integration_tests/tests/test_column_anomalies.py
@@ -1,6 +1,7 @@
 from datetime import datetime, timedelta
 from typing import Any, Dict, List
 
+import dateutil.parser
 from data_generator import DATE_FORMAT, generate_dates
 from dbt_project import DbtProject
 from parametrization import Parametrization
@@ -720,3 +721,57 @@ def test_col_excl_detect_train_seven_day_bucket(test_id: str, dbt_project: DbtPr
         "Expected FAIL when exclude_detection_period_from_training=True "
         "(large bucket fix: detection period set to bucket size)"
     )
+
+
+# Regression test for https://github.com/elementary-data/elementary/issues/2370
+def test_column_anomalies_weekly_buckets_with_sunday_rows(
+    test_id: str, dbt_project: DbtProject
+):
+    utc_today = datetime.utcnow().date()
+    last_sunday = utc_today - timedelta(days=utc_today.isoweekday() % 7 or 7)
+    sundays = [last_sunday - timedelta(weeks=week) for week in range(12)]
+
+    # Nulls only on Sundays, with a different count each week, so a Sunday
+    # counted in the wrong week changes the weekly null_count.
+    expected_null_counts = {}
+    data: List[Dict[str, Any]] = []
+    for week, sunday in enumerate(sundays):
+        null_count = 1 + week % 3
+        expected_null_counts[sunday - timedelta(days=6)] = null_count
+        data += [
+            {TIMESTAMP_COLUMN: sunday.strftime(DATE_FORMAT), "superhero": None}
+            for _ in range(null_count)
+        ]
+    data += [
+        {TIMESTAMP_COLUMN: cur_date.strftime(DATE_FORMAT), "superhero": "Superman"}
+        for cur_date in generate_dates(base_date=last_sunday, days_back=12 * 7)
+    ]
+
+    test_args = {
+        **DBT_TEST_ARGS,
+        "time_bucket": {"period": "week", "count": 1},
+        "training_period": {"period": "day", "count": 12 * 7},
+    }
+    test_result = dbt_project.test(
+        test_id, DBT_TEST_NAME, test_args, data=data, test_column="superhero"
+    )
+    assert test_result["status"] == "pass"
+
+    metrics = dbt_project.run_query(
+        f"""
+        select bucket_start, metric_value
+        from {{{{ ref("data_monitoring_metrics") }}}}
+        where metric_name = 'null_count'
+        and lower(full_table_name) like '%{test_id.lower()}'
+        """
+    )
+    null_counts = {
+        dateutil.parser.parse(metric["bucket_start"]).date(): metric["metric_value"]
+        for metric in metrics
+        if metric["bucket_start"] is not None
+    }
+    assert len(null_counts) == len(metrics), "Found metrics without a bucket"
+    assert null_counts
+    for bucket_start, null_count in null_counts.items():
+        assert bucket_start.isoweekday() == 1, f"{bucket_start} is not a Monday"
+        assert null_count == expected_null_counts[bucket_start], bucket_start
diff --git a/macros/edr/data_monitoring/monitors_query/get_start_bucket_in_data.sql b/macros/edr/data_monitoring/monitors_query/get_start_bucket_in_data.sql
index 0c3cb23b6..79c6bd12e 100644
--- a/macros/edr/data_monitoring/monitors_query/get_start_bucket_in_data.sql
+++ b/macros/edr/data_monitoring/monitors_query/get_start_bucket_in_data.sql
@@ -1,4 +1,9 @@
 {% macro get_start_bucket_in_data(timestamp_column, min_bucket_start, time_bucket) %}
+    {#- Count weeks as 7 days: where a week starts in datediff depends on the
+        adapter (Sunday on Postgres), while buckets start on min_bucket_start. -#}
+    {% if time_bucket.period | lower == "week" %}
+        {% set time_bucket = {"period": "day", "count": time_bucket.count * 7} %}
+    {% endif %}
     {% set bucket_start_datediff_expr %}
       floor({{ elementary.edr_datediff(min_bucket_start, elementary.edr_cast_as_timestamp(timestamp_column), time_bucket.period) }} / {{ time_bucket.count }}) * {{ time_bucket.count }}
     {% endset %}