fix: logging overhaul — Alembic fileConfig disable_existing_loggers=False, propagate=False on loggers, ADR docs, catch-up migration for missing columns, sqlparse dep, build archive naming

backend/alembic/env.py

@@ -20,6 +20,18 @@ if "DATABASE_URL" in os.environ:
 
 # Interpret the config file for Python logging.
 # This line sets up loggers basically.
+# @ADR [LOG-001] disable_existing_loggers=False
+# @RATIONALE Python 3.13's logging.config.fileConfig() defaults to
+#   disable_existing_loggers=True, which sets logger.disabled = True on ALL
+#   existing loggers not listed in alembic.ini's [loggers] section. This
+#   includes our 'superset_tools_app' and 'cot' loggers. With disabled=True,
+#   Logger.isEnabledFor() immediately returns False (see Python 3.13 source),
+#   silently dropping every logger.info()/warning()/error() call across the
+#   entire application. The SS-TOOLS logging system manages its own handlers
+#   via configure_logger() — Alembic's INI-based config is irrelevant here.
+# @REJECTED Adding superset_tools_app to alembic.ini [loggers] was rejected
+#   because it couples app logging config to Alembic's ini format, which is
+#   fragile and non-obvious for future maintainers.
 if config.config_file_name is not None:
     fileConfig(config.config_file_name, disable_existing_loggers=False)
 

backend/alembic/versions/2df63b7ce038_add_missing_columns_policy_id_provider_.py

@@ -0,0 +1,52 @@
+"""add missing columns policy_id provider_id is_multimodal
+
+@ADR [LOG-003] Catch-up migration for columns added in inserted revisions.
+@RATIONALE Migrations 9f8e7d6c5b4a (is_multimodal), a7b1c2d3e4f5 (provider_id),
+  and b1c2d3e4f5a6 (policy_id) were inserted into the chain before the
+  existing head 86c7b1d6a710. Databases already at 86c7b1d6a710 skip them
+  because Alembic sees no gap between current and target revision.
+  This migration adds the same columns with 86c7b1d6a710 as parent, so
+  it runs on both fresh and pre-stamped databases.
+
+Revision ID: 2df63b7ce038
+Revises: 86c7b1d6a710
+Create Date: 2026-05-27 11:17:26.789634
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = '2df63b7ce038'
+down_revision: Union[str, Sequence[str], None] = '86c7b1d6a710'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    """Upgrade schema."""
+    # Add is_multimodal to llm_providers (from 9f8e7d6c5b4a)
+    op.add_column("llm_providers",
+        sa.Column("is_multimodal", sa.Boolean(), nullable=False, server_default="false")
+    )
+    op.alter_column("llm_providers", "is_multimodal", server_default=None)
+
+    # Add provider_id to validation_policies (from a7b1c2d3e4f5)
+    op.add_column("validation_policies",
+        sa.Column("provider_id", sa.String(), nullable=True)
+    )
+
+    # Add policy_id to llm_validation_results (from b1c2d3e4f5a6)
+    op.add_column("llm_validation_results",
+        sa.Column("policy_id", sa.String(), nullable=True, index=True)
+    )
+
+
+def downgrade() -> None:
+    """Downgrade schema."""
+    op.drop_column("llm_validation_results", "policy_id")
+    op.drop_column("validation_policies", "provider_id")
+    op.drop_column("llm_providers", "is_multimodal")

backend/requirements.txt

@@ -57,5 +57,6 @@ playwright
 tenacity
 Pillow
 ruff>=0.11.0
+sqlparse>=0.5.0
  lingua-language-detector==2.1.1
 testcontainers[postgres]>=4.0

backend/src/core/logger.py

@@ -6,9 +6,20 @@
 # @RELATION DEPENDS_ON -> [WebSocketLogHandler]
 # @PRE Python 3.7+ with cot_logger ContextVars available.
 # @POST All log output is single-line JSON with ts, level, trace_id, src, marker, intent, payload, error, span_id.
-# @SIDE_EFFECT Configures global logger handlers and formatters; seeds global _enable_belief_state and _task_log_level; writes JSON to console/files.
-# @RATIONALE Replaced BeliefFormatter text-based [Entry]/[Exit]/[Action] logging with CotJsonFormatter JSON output. Old format was not machine-parseable and mixed presentation with intent. CotJsonFormatter produces structured JSON consistent with cot_logger.py MarkerLogger protocol.
-# @REJECTED Keeping both formatters side-by-side was rejected (two inconsistent output formats would confuse log consumers).
+#   Both 'superset_tools_app' and 'cot' loggers have propagate=False — no text duplicates via root logger.
+# @INVARIANT Every logger.info() call MUST produce exactly one JSON line (see [ADR LOG-002] for why).
+# @SIDE_EFFECT Configures global logger handlers and formatters; seeds global _enable_belief_state
+#   and _task_log_level; sets propagate=False on both loggers; writes JSON to console/files.
+# @RELATION DISABLED_BY -> [LOG-001:alembic/env.py:fileConfig]
+#   If alembic/env.py calls fileConfig() without disable_existing_loggers=False, Python 3.13's
+#   logging framework sets logger.disabled = True on 'superset_tools_app', silencing ALL output.
+# @RATIONALE Replaced BeliefFormatter text-based [Entry]/[Exit]/[Action] logging with CotJsonFormatter
+#   JSON output. Old format was not machine-parseable and mixed presentation with intent. CotJsonFormatter
+#   produces structured JSON consistent with cot_logger.py MarkerLogger protocol.
+# @REJECTED Keeping both formatters side-by-side was rejected (two inconsistent output formats would
+#   confuse log consumers).
+# @REJECTED Keeping propagate=True was rejected after uvicorn's dictConfig added a root logger handler
+#   that duplicates every log as plain text (see [ADR LOG-002]).
 # @DATA_CONTRACT All log records conform to molecular CoT JSON schema: {ts, level, trace_id, src, marker, intent, span_id?, payload?, error?}
 # @INVARIANT CotJsonFormatter.format() always returns valid single-line JSON string.
 from contextlib import contextmanager
@@ -162,8 +173,12 @@ def belief_scope(anchor_id: str, message: str = ""):
 # @RELATION CALLS -> [CotJsonFormatter]
 # @RELATION CALLS -> [CotLoggerModule]
 # @PRE config is a valid LoggingConfig instance.
-# @POST Logger levels, handlers, formatters, belief state flag, and task log level are updated.
-# @SIDE_EFFECT Mutates global _enable_belief_state, _task_log_level; adds/removes handlers on both logger and cot_logger; creates file directories.
+# @POST Logger levels, handlers, formatters, belief state flag, task log level, and
+#   propagate flags (False) are updated. Both loggers emit JSON only — no text duplicates.
+# @SIDE_EFFECT Mutates global _enable_belief_state, _task_log_level; adds/removes handlers
+#   on both logger and cot_logger; sets propagate=False on both; creates file directories.
+# @DATA_CONTRACT configure_logger() -> ENABLES both loggers. Every call to logger.info()
+#   MUST produce exactly one JSON line (not zero, not two). See [ADR LOG-001] and [ADR LOG-002].
 def configure_logger(config):
     global _enable_belief_state, _task_log_level
     _enable_belief_state = config.enable_belief_state
@@ -206,9 +221,21 @@ def configure_logger(config):
         cot_handler.setFormatter(CotJsonFormatter())
         cot_logger.addHandler(cot_handler)
 
-    # Disable propagation to prevent duplicate output via uvicorn's root logger handlers.
-    # Both loggers have their own dedicated handlers; propagation would cause each message
-    # to appear twice: once via CotJsonFormatter (JSON) and once via uvicorn's DefaultFormatter (text).
+    # @ADR [LOG-002] Disable logger propagation to prevent duplicate output
+    # @RATIONALE Uvicorn's configure_logging() registers a StreamHandler on the root
+    #   logger via logging.config.dictConfig(). Both 'superset_tools_app' and 'cot'
+    #   loggers have their own dedicated handlers with CotJsonFormatter. With
+    #   propagate=True (default), each log message is handled twice:
+    #   1. Our CotJsonFormatter → structured JSON (desired)
+    #   2. Propagated to root → uvicorn's DefaultFormatter → plain text (duplicate)
+    #   Setting propagate=False prevents the second emission.
+    # @REJECTED Removing uvicorn's root logger handler via logging.getLogger().handlers = []
+    #   was rejected because it couples SS-TOOLS to uvicorn's internal logging setup,
+    #   which could change between uvicorn versions. Setting propagate on our own
+    #   loggers is self-contained and version-safe.
+    # @REJECTED Overriding log level on root logger was rejected (root.setLevel(WARNING))
+    #   because it would suppress legitimate warnings from third-party libraries
+    #   (SQLAlchemy, Alembic, httpx, etc.).
     logger.propagate = False
     cot_logger.propagate = False
 # #endregion configure_logger