test(data): silver business-rule data tests — non-negative counts (#1321)

src/ingestion/dbt/tests/collaboration/assert_collab_document_counts_non_negative.sql

@@ -0,0 +1,36 @@
+{{ config(
+    tags=['data_quality'],
+    severity='warn',
+    store_failures=true,
+    meta={
+        'title': 'Collab document activity counts are non-negative',
+        'domain': 'collab',
+        'category': 'physical_bound',
+        'tier': 'error',
+        'remediation': 'A negative activity count is physically impossible and points to a broken transform or a bad source row; any metric that sums the column is corrupted. Inspect the stored failing rows by unique_key, trace back to the m365 sharepoint/onedrive feeder mapping, and fix the source. A NULL count (a column a product never emits, e.g. visited_page_count on OneDrive) is not a violation.'
+    }
+) }}
+-- Business-rule data test (#1321 silver-layer integrity).
+-- Activity counts can never be negative; a negative value means a broken
+-- transform or bad source row and would corrupt any metric that sums them.
+-- NULL (not-ingested, e.g. visited_page_count on OneDrive) is intentionally
+-- NOT flagged — NULL < 0 is NULL, not a violation; honest NULLs are handled
+-- separately. Read FINAL so transient ReplacingMergeTree duplicates can't
+-- surface as repeated violations of the same row. `date` is selected purely
+-- as triage context in the stored-failure rows (the activity day of the bad
+-- count) — it is not a filter: a negative count is corruption on any partition,
+-- so the check stays unbounded over the full table.
+SELECT
+    unique_key,
+    date,
+    viewed_or_edited_count,
+    synced_count,
+    shared_internally_count,
+    shared_externally_count,
+    visited_page_count
+FROM {{ ref('class_collab_document_activity') }} FINAL
+WHERE viewed_or_edited_count   < 0
+   OR synced_count             < 0
+   OR shared_internally_count  < 0
+   OR shared_externally_count  < 0
+   OR visited_page_count       < 0