[Yogyabhilash]

Description

Fixes #3325.

In the scenario progress GUI and backend read model, scorer evaluation metrics (ScenarioObjectiveScorerMetrics) omitted num_responses (evaluated count) and num_input_responses (total count), even though:

1. ObjectiveScorerMetrics already records both fields during evaluation.
2. The scorer evaluation documentation (doc/code/scoring/4_scorer_metrics.py) explicitly instructs users to compare num_responses against num_input_responses to verify coverage, because responses that do not yield a verdict are excluded from accuracy calculations.

Without these fields, the scenario progress GUI displays evaluated metrics without indicating coverage or incomplete evaluations (e.g., displaying 100% accuracy on a run where only a fraction of prompts had verdicts, with no indication that responses without verdicts were excluded).

Changes:

• pyrit/models/scenario_progress.py: Added optional non-negative integer fields num_responses and num_input_responses to ScenarioObjectiveScorerMetrics (defaults to None for backward compatibility with older/legacy records).
• pyrit/backend/services/scenario_progress_read_model.py: Forwarded num_responses and num_input_responses from official_metrics when building ScenarioObjectiveScorerMetrics.
• frontend/src/types/index.ts: Updated ScenarioObjectiveScorerMetrics interface to include num_responses?: number | null and num_input_responses?: number | null.
• frontend/src/utils/scorerPresentation.ts: Added presentation labels (Evaluated responses, Total responses) and metric descriptions.
• frontend/src/components/AttackResults/ObjectiveScorerDetails.tsx: Rendered coverage context hint when num_responses < num_input_responses explaining that responses without a verdict were excluded from accuracy.

Tests and Documentation

• Added unit tests in tests/unit/backend/test_scenario_progress_read_model.py:
  • test_progress_snapshot_propagates_scorer_coverage_metrics: Verifies that num_responses and num_input_responses are propagated from scorer evaluation files to the progress snapshot.
  • test_load_objective_metrics_reads_coverage_and_legacy_records: Verifies that records with coverage metrics are read correctly and legacy JSONL records lacking num_input_responses continue to load safely without error.
• Verified test suite:
  • pytest tests/unit/backend/test_scenario_progress_read_model.py (60 passed).
  • Full backend test suite tests/unit/backend (3506 passed, 4 skipped).
• Verified code quality:
  • ruff check pyrit tests/unit/backend/test_scenario_progress_read_model.py (all checks passed).
  • ruff format --check pyrit tests/unit/backend/test_scenario_progress_read_model.py (3 files already formatted).
  • uv run --frozen --extra all -m ty check (all checks passed).
• JupyText: Not applicable as no documentation notebooks or tutorial .py pairs were modified.

Checklist

• Targeted fix solving the root cause of Scorer-eval GUI drops the coverage fields the docs say to compare #3325
• Backwards-compatible with existing / legacy evaluation metrics records
• Unit regression tests added and passed locally
• Linting (ruff check) and formatting (ruff format) verified
• Static type checking (ty check) verified

AI assistance disclosure: Prepared with an AI coding agent under direction; verified, tested, and reviewed locally.

[Copilot]

🟢 Approval recommended

No active review issues remain for the change exposing scorer evaluation coverage alongside accuracy.

No open issues 🎉

What changed in this PR

Addresses #3325 by carrying scorer evaluation coverage through the scenario progress API to the GUI, so users can distinguish accuracy from evaluation completeness.

Changes:

• Adds optional evaluated and total response counts to backend models and frontend types.
• Displays coverage metrics and a caveat when responses without verdicts were excluded from accuracy.
• Adds backend tests for count propagation and compatibility with legacy records.

File	Description
tests/​unit/​backend/​test_scenario_progress_read_model.py	Tests coverage propagation and legacy record loading.
pyrit/​models/​scenario_progress.py	Adds optional, non-negative coverage counts.
pyrit/​backend/​services/​scenario_progress_read_model.py	Forwards coverage counts from recorded evaluation metrics.
frontend/​src/​utils/​scorerPresentation.ts	Adds coverage labels and descriptions.
frontend/​src/​types/​index.ts	Adds nullable coverage counts to the metrics interface.
frontend/​src/​components/​AttackResults/​ObjectiveScorerDetails.tsx	Displays a coverage caveat for incomplete evaluations.

_{Review effort: Balanced}