[Hrushikesh Ramilla (Hrushikesh-ramilla)]

Description

Fixes #3699.

When querying scenario history aggregates (get_scenario_history_aggregates_async in pyrit/memory/memory_interface.py), per-group attempt counts from group_rows are grouped by (scenario_result_id, unit_group_id) in SQL.

For unmatched attempts, raw_gid is formatted as name\x1feval_hash, which lines 7336-7338 convert to the canonical configuration hash canonical_gid = config_hash({"atomic_attack_name": name, "technique_eval_hash": eval_hash}).

Previously, run_groups[canonical_gid] = counts unconditionally assigned the row counts. If an unplanned attempt shared the same atomic attack name and technique configuration as a planned group, its row overwrote the existing canonical group's counts instead of combining the disjoint attempt populations. As a result, the group's planned completed/successful counts were overwritten, while overall run counts remained correct.

Fix

• In MemoryInterface.get_scenario_history_aggregates_async:
  • If canonical_gid already exists in run_groups, combine the counts using combine_execution_counts([run_groups[canonical_gid], counts]).
  • Similarly, if canonical_gid != raw_gid, combine into run_groups[raw_gid] if already present.

Tests

• Added unit regression test test_scenario_history_aggregates_combine_unplanned_attempt_counts_with_canonical_group in tests/unit/memory/memory_interface/test_interface_scenario_history.py using canonical configuration hashes and both planned and unplanned attempt records.
• Verification:
  • pytest tests/unit/memory/memory_interface/test_interface_scenario_history.py: 25 passed.
  • pytest tests/unit/analytics/test_scenario_statistics_parity.py: 37 passed.
  • ruff check pyrit/memory/memory_interface.py tests/unit/memory/memory_interface/test_interface_scenario_history.py: 0 errors.
  • ruff format --check pyrit/memory/memory_interface.py tests/unit/memory/memory_interface/test_interface_scenario_history.py: Clean.