feat: add account_id filter for sources in search_beliefs

.github/agents/architecture-domain-specialist.md

@@ -131,6 +131,7 @@ Marshmallow schemas define the canonical format for parameter dictionaries. All
 - [ ] **Dictionary access**: Verify code uses dict keys from `data_key`, not Python attributes
 - [ ] **Parameter modification**: Check `pop()`, `del`, assignment operations use correct keys
 - [ ] **Storage consistency**: Ensure DataSource.attributes, job.meta use schema format
+- [ ] **Schema parity**: When adding a filter/parameter to `Sensor.search_beliefs`, verify it is added to BOTH `Input` (io.py) AND `BeliefsSearchConfigSchema` (reporting/__init__.py). These two schemas serve overlapping purposes but are distinct classes — omitting one creates a silent gap where documented features silently fail at schema validation time.
 
 **Domain Pattern: Schema Format Migrations**
 
@@ -588,3 +589,18 @@ After each assignment:
    Change:
    - Added guidance on <specific topic>
    ```
+
+### Lessons Learned
+
+**Session 2026-03-24 (PR #2058 — add account_id to DataSource)**:
+
+- **New domain invariant**: User-type DataSources now have `account_id` populated. Document new FK relationships and invariants in the Domain Knowledge section immediately.
+- **Migration checklist**: Added an explicit Alembic migration checklist after reviewing the migration for this PR. Key patterns: correlated subquery for bulk backfill, SQLAlchemy Core stubs (no ORM imports), `batch_alter_table` for all ALTER operations, exact constraint name matching.
+- **Missed API Specialist coordination**: The PR changed endpoint behavior (POST sensor data sets account_id on the created data source). The API Specialist should have been engaged to verify backward compatibility. When domain model changes affect how endpoints behave or what they return, flag for API Specialist review.
+- **Self-improvement failure**: Despite having explicit self-improvement requirements, no agent updated its instructions during this PR session. This was caught by the Coordinator post-hoc. The agent must update its own instructions as the LAST step of every assignment, not skip it.
+
+**Session 2026-04 (PR #2065 — add account_id filter to search_beliefs)**:
+
+- **Schema parity gap**: The PR added `account_id` to `BeliefsSearchConfigSchema` but not to `Input` (io.py). These two schemas both expose `Sensor.search_beliefs` parameters; omitting a parameter from one creates a silent gap. The architecture agent must check both schemas on any search_beliefs parameter addition.
+- **Documentation vs. implementation mismatch**: The `reporting.rst` docs stated reporters can filter by `account_id`, but this only works if `Input` also has the field. Docs that outrun schema support mislead users. Always verify the full schema chain before documenting a feature.
+- **DataSource account_id=None for non-user sources**: The existing invariant (reporters/schedulers/forecasters have `account_id=None`) limits the usefulness of `account_id` filtering: it only matches user-type sources. PRs adding `account_id` filters should either document this limitation explicitly or reconsider the invariant.

.github/agents/test-specialist.md

@@ -118,11 +118,11 @@ If ANY test fails during full suite execution:
 **Click context errors**:
 - Check IdField decorators (`@with_appcontext` vs `@with_appcontext_if_needed()`)
 - Compare against SensorIdField pattern
-- See Review Lead's Click context error pattern
+- See Lead's Click context error pattern
 
-### Integration with Review Lead
+### Integration with Lead
 
-The Test Specialist MUST provide evidence of full test suite execution to Review Lead.
+The Test Specialist MUST provide evidence of full test suite execution to Lead.
 
 **Required evidence format:**
 ```
@@ -134,14 +134,14 @@ Full test suite execution:
 - Coverage: 87.2% (unchanged)
 ```
 
-**Review Lead verification:**
-Review Lead's session close checklist includes:
+**Lead verification:**
+Lead's session close checklist includes:
 - [ ] Test Specialist confirmed full test suite execution
 - [ ] All tests pass (100%)
 - [ ] Test output captured and reviewed
 
 **Enforcement:**
-Review Lead cannot close session until Test Specialist provides evidence of full test suite execution with 100% pass rate.
+Lead cannot close session until Test Specialist provides evidence of full test suite execution with 100% pass rate.
 
 
 ## Testing Patterns for flexmeasures
@@ -880,6 +880,30 @@ flask db current
 - **Assertions**: Use descriptive assertion messages for failures
 - **Mocking**: Use pytest fixtures and mocking when testing external dependencies
 
+### Testing DataSource Properties After API Calls
+
+When writing tests that verify data source properties (e.g. `account_id`, `user`, `type`) after an API call:
+
+1. **Use `fresh_db` fixture** — tests that POST data and then query the resulting data source are modifying the DB and must use the function-scoped `fresh_db` fixture. Place these tests in a `_fresh_db` module.
+
+2. **Query by user, not just name** — data sources created by the same user across test runs may collide; use `filter_by(user=user)` or `filter_by(user_id=user.id)` for precision.
+
+3. **Pattern** (from `test_post_sensor_data_sets_account_id_on_data_source`):
+   ```python
+   # Fetch the user that made the request
+   user = db.session.execute(
+       select(User).filter_by(email="test_supplier_user_4@seita.nl")
+   ).scalar_one()
+   # Fetch the data source created for that user
+   data_source = db.session.execute(
+       select(Source).filter_by(user=user)
+   ).scalar_one_or_none()
+   assert data_source is not None
+   assert data_source.account_id == user.account_id
+   ```
+
+4. **Check both existence and value** — don't just assert `data_source is not None`; also assert the specific field value you're testing.
+
 ## Understanding Test Design Intent (CRITICAL)
 
 **Before changing a test, understand WHY it's designed that way.**
@@ -1060,14 +1084,14 @@ After each assignment:
    - Added guidance on <specific topic>
    ```
 
-Example:
+### Lessons Learned
 
-```
-agents/test-specialist: learned to verify claims with actual test runs
-Context:
-- Session #456 claimed tests passed but they were never actually run
-- Led to bug slipping through to production
-Change:
-- Added "Actually Run Tests" section with verification steps
-- Emphasized checking test output and coverage
-```
+**Session 2026-03-24 (PR #2058 — add account_id to DataSource)**:
+
+- **Self-improvement failure**: Despite having explicit instructions to update this agent file after each assignment, no update was made during this PR session. This was caught by the Coordinator post-hoc. The agent must treat instruction updates as the LAST mandatory step of any assignment.
+- **DataSource property testing**: Added guidance in "Testing DataSource Properties After API Calls" above. When testing properties set by the API on a data source (like `account_id`), use `fresh_db`, query by user to avoid ambiguity, and assert both existence and the specific field value.
+
+**Session 2026-04 (PR #2065 — add account_id filter to search_beliefs)**:
+
+- **Model layer coverage gap**: The PR added `account_id` parameter to three model methods: `Sensor.search_beliefs`, `TimedBelief.search`, and `GenericAsset.search_beliefs`. Tests covered only the first two. When a parameter is added to multiple model-layer methods, each method must be tested independently — especially when they use different code paths (e.g., `GenericAsset.search_beliefs` delegates through `Sensor.search_beliefs` but may not if the delegation chain changes). Add a checklist item: "When a filter/param is added to `search_beliefs` across multiple models, ensure at least one test exercises each model class."
+- **Empty-list edge case**: `account_id=[]` is schema-valid but semantically means "match nothing" (SQL `IN ()` returns zero rows). Tests should include this edge case and assert the expected empty result rather than letting it pass as a valid no-op.

.github/agents/tooling-ci-specialist.md

@@ -163,12 +163,12 @@ Code that bypasses pre-commit:
 
 **Who runs pre-commit:**
 - **During code changes**: Agent making changes runs pre-commit before committing
-- **Before PR close**: Review Lead verifies pre-commit execution
+- **Before PR close**: Lead verifies pre-commit execution
 - **In PR review**: Tooling & CI Specialist validates config matches CI
 
 **Enforcement:**
-- Review Lead's session close checklist includes pre-commit verification
-- Review Lead cannot close session without pre-commit evidence
+- Lead's session close checklist includes pre-commit verification
+- Lead cannot close session without pre-commit evidence
 - If pre-commit fails, agent must fix all issues before proceeding
 
 #### Common Failures and Fixes
@@ -206,9 +206,9 @@ black .
 ci/run_mypy.sh
 ```
 
-#### Integration with Review Lead
+#### Integration with Lead
 
-**Review Lead checklist items:**
+**Lead checklist items:**
 - [ ] Pre-commit hooks installed
 - [ ] All hooks pass: `pre-commit run --all-files`
 - [ ] Zero failures from flake8, black, mypy
@@ -219,7 +219,7 @@ ci/run_mypy.sh
 - Or confirm: "Pre-commit verified: all hooks passed"
 
 **Enforcement:**
-Review Lead MUST verify pre-commit execution before closing session.
+Lead MUST verify pre-commit execution before closing session.
 
 ### Agent Environment Setup