refactor: delete legacy leftovers with no callers

CHANGELOG.md

@@ -9,6 +9,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 - **`first_speaker` replaces `persona_speaks_first` and `--provider-speaks-first`**—Who opens a conversation is now one setting with one name: `"first_speaker": "user"` or `"chatbot"` in a config file, and `--first-speaker {user,chatbot}` on the CLI (default `user`). Update configs: `"persona_speaks_first": true` becomes `"first_speaker": "user"`, and `false` becomes `"chatbot"`. A config with the old field stops with an error that says so. The files in `configs/` are updated.
 - **Code moved under `src/vera_mh/`**—`generate`, `judge`, `score`, `utils`, `llm_clients`, and `vera_cli` are now subpackages of one package, so imports gain a prefix: `from judge.runner import X` becomes `from vera_mh.judge.runner import X`, and the same goes for `patch()` targets. `python -m score.compare` is now `python -m vera_mh.score.compare`. `uv sync` installs the package and a `vera` command (`vera = "vera_mh.cli:main"`), which replaces `python vera.py`: the root `vera.py` is gone, so run `uv run vera <command>`. Design: [docs/design/src-layout.md](docs/design/src-layout.md) (#285).
 - **Legacy scripts removed**—`legacy/` is gone (`legacy.generate`, `legacy.judge`, `legacy.score`, `legacy.run_pipeline`, `legacy.pool_vera_scores`, `run_recommended_vera_pipeline.sh`), as announced in v2.0.4, along with `docs/legacy-scripts.md`, the tests that drove the scripts, and the `data/SI/rubric_manifest.json` and `data/HFO/rubric_manifest.json` symlinks only they read. Use the `vera` commands; the flag mapping is in [v2.0.4's `docs/legacy-scripts.md`](https://github.com/SpringCare/VERA-MH/blob/v2.0.4/docs/legacy-scripts.md). `distribute_files.py`, `scripts/run_combinations.sh`, and `scripts/score_publication_evaluations.sh` are removed too. To run several chatbots, loop over `vera generate --chatbot`; `--user` already takes several models.
+- **Legacy leftovers removed from `judge/`**—`judge_single_conversation` (also exported from `vera_mh.judge`), `build_single_conversation_judge_run_key`, and `get_judge_logs_root` are gone; nothing called them once the legacy scripts left. `build_judge_task_log_path` now requires `output_folder` and drops `run_key` and `logs_root`: judge logs always go to `<output_folder>/logs/`, and `VERA_JUDGE_LOGS_ROOT` is no longer read. To judge one conversation, put it in a folder of its own and pass that to `vera judge --conversations`.
 
 ### Runtime, CLI, and pipeline
 

src/vera_mh/generate/runner.py

@@ -262,36 +262,6 @@ def _job_result(
             out["skip_reason"] = skip_reason
         return out
 
-    def _create_conversation_jobs(
-        self, persona_names: Optional[List[str]] = None
-    ) -> List[Tuple[dict, int, int, int]]:
-        """Create job tuples for all persona/run combinations."""
-        personas = load_prompts_from_csv(
-            persona_names,
-            prompt_path=self.persona_prompt_path,
-            persona_context_template_path=self.persona_context_template_path,
-            user_language=self.user_language,
-            max_personas=self.max_personas,
-        )
-        jobs: List[Tuple[dict, int, int, int]] = []
-        conversation_index = 1
-        for persona in personas:
-            for run in range(1, self.runs_per_prompt + 1):
-                jobs.append(
-                    (
-                        {
-                            "prompt": persona["prompt"],
-                            "name": persona["Name"],
-                            "run": run,
-                        },
-                        self.max_turns,
-                        conversation_index,
-                        run,
-                    )
-                )
-                conversation_index += 1
-        return jobs
-
     async def _worker(
         self,
         worker_id: int,

src/vera_mh/judge/__init__.py

@@ -2,14 +2,10 @@
 
 from .llm_judge import LLMJudge
 from .run import run_judging
-from .runner import (
-    judge_conversations,
-    judge_single_conversation,
-)
+from .runner import judge_conversations
 
 __all__ = [
     "LLMJudge",
     "judge_conversations",
-    "judge_single_conversation",
     "run_judging",
 ]

src/vera_mh/judge/llm_judge.py

@@ -43,7 +43,7 @@ def __init__(
             verbose: Whether to print verbose output during initialization
         """
 
-        # Setup logger: batch judging and `judge.py` pass an explicit path from
+        # Setup logger: batch judging passes an explicit path from
         # build_judge_task_log_path (scoped to the run). Omitted log_file is only
         # for direct construction, tests, etc.—then judge_unscoped/ under
         # default_adhoc_parent(); UUID stem avoids concurrent overwrites.
@@ -63,7 +63,7 @@ def __init__(
 
         # One log path per batch job: stem matches the evaluation TSV (see
         # build_judge_task_log_path). Different evaluation runs use different
-        # output folders, so run_key differs and logs do not collide across runs.
+        # output folders, so logs do not collide across runs.
         # mode="w" truncates on each job start; re-running the same job in the
         # same folder replaces the log (and TSV), which is intentional.
         file_handler = logging.FileHandler(log_file, mode="w", encoding="utf-8")

src/vera_mh/judge/runner.py

@@ -747,30 +747,3 @@ async def judge_conversations(
             print(f"  ({skipped_error_n} skipped due to errors)")
 
     return results, output_folder
-
-
-async def judge_single_conversation(
-    judge: LLMJudge,
-    conversation: ConversationData,
-    output_folder: str,
-) -> Optional[Dict[str, Any]]:
-    """
-    Judge a single conversation.
-
-    Args:
-        judge: LLMJudge instance
-        conversation: ConversationData with content and metadata
-        output_folder: Output folder for results
-
-    Returns:
-        Evaluation results or None if failed
-    """
-    conversation_filename = conversation.metadata.get("filename", "unknown.txt")
-    print(f"📄 Judging: {conversation_filename}")
-
-    result = await judge.evaluate_conversation_question_flow(
-        conversation, output_folder=output_folder, auto_save=True
-    )
-
-    print(f"🟢 Done: {conversation_filename}")
-    return result

src/vera_mh/judge/utils.py

@@ -1,7 +1,6 @@
 """Utility functions for the judge module."""
 
 import os
-from datetime import datetime
 from pathlib import Path
 from typing import Optional, Sequence
 
@@ -78,65 +77,22 @@ def judge_evaluation_tsv_filename(
     return f"{conversation_name}_{judge_suffix}.tsv"
 
 
-def build_single_conversation_judge_run_key(
-    conversation_path: str | Path,
-    *,
-    now: Optional[datetime] = None,
-) -> str:
-    """
-    Human-readable folder basename for ``judge.py`` single-file (``-c``) mode.
-
-    Format: ``single_<timestamp_ms>__<conversation_stem>``. Batch judging uses the
-    evaluation folder basename instead; this gives a stable, unique folder name
-    per CLI run. Per-task logs use :func:`build_judge_task_log_path` with
-    ``output_folder`` set, so no ``run_key`` is involved.
-    """
-    dt = datetime.now() if now is None else now
-    ts = dt.strftime("%Y%m%d_%H%M%S_%f")[:-3]
-    stem = Path(conversation_path).stem
-    return f"single_{ts}__{stem}"
-
-
-def get_judge_logs_root() -> str:
-    """
-    Root directory for legacy per-task judge LLM logs (when ``output_folder`` is
-    omitted from :func:`build_judge_task_log_path`).
-
-    Override with env ``VERA_JUDGE_LOGS_ROOT`` (e.g. set by pytest to a temp dir).
-    Default: ``judge_logs`` in the current working directory.
-    """
-    return os.environ.get("VERA_JUDGE_LOGS_ROOT", "judge_logs")
-
-
 def build_judge_task_log_path(
     conversation_filename: str,
     judge_model: str,
     judge_instance: Optional[int] = None,
     *,
-    run_key: Optional[str] = None,
-    logs_root: Optional[str] = None,
-    output_folder: Optional[str] = None,
+    output_folder: str,
 ) -> str:
     """
-    Path to the per-task judge LLM log file, parallel to judge_evaluation_tsv_filename.
-
-    With ``output_folder``: ``{output_folder}/logs/{stem}.log`` (``run_key`` unused).
-
-    Legacy layout (no ``output_folder``): ``{logs_root}/{run_key}/{stem}.log``
-    (``logs_root`` defaults to :func:`get_judge_logs_root`). ``run_key`` is required
-    for this branch.
+    Path to the per-task judge LLM log file, parallel to judge_evaluation_tsv_filename:
+    ``{output_folder}/logs/{stem}.log``.
     """
     tsv_name = judge_evaluation_tsv_filename(
         conversation_filename, judge_model, judge_instance
     )
     stem = Path(tsv_name).stem
-    if output_folder is not None:
-        return str(Path(output_folder) / "logs" / f"{stem}.log")
-    if not run_key:
-        raise ValueError("run_key is required when output_folder is not set")
-    # Legacy layout: {logs_root}/{run_key}/{stem}.log
-    root = logs_root if logs_root is not None else get_judge_logs_root()
-    return str(Path(root) / run_key / f"{stem}.log")
+    return str(Path(output_folder) / "logs" / f"{stem}.log")
 
 
 def default_adhoc_parent() -> str:

tests/conftest.py

@@ -8,27 +8,20 @@
 from tests.mocks.mock_llm import MockLLM
 from vera_mh.judge.rubric_config import RubricConfig
 
-_judge_logs_temp: Path | None = None
 _adhoc_temp: Path | None = None
 
 
 def pytest_sessionstart(session: pytest.Session) -> None:
-    """Point judge file logs and adhoc outputs at temp dirs (cleaned in finish)."""
-    global _judge_logs_temp, _adhoc_temp
-    _judge_logs_temp = Path(tempfile.mkdtemp(prefix="vera_judge_logs_"))
-    os.environ["VERA_JUDGE_LOGS_ROOT"] = str(_judge_logs_temp)
+    """Point adhoc outputs, including unscoped judge logs, at a temp dir."""
+    global _adhoc_temp
     _adhoc_temp = Path(tempfile.mkdtemp(prefix="vera_adhoc_"))
     os.environ["VERA_ADHOC_PARENT"] = str(_adhoc_temp)
 
 
 def pytest_sessionfinish(session: pytest.Session, exitstatus: int) -> None:
-    """Remove temp dirs; clear env so CLI uses normal defaults."""
-    global _judge_logs_temp, _adhoc_temp
-    os.environ.pop("VERA_JUDGE_LOGS_ROOT", None)
+    """Remove the temp dir; clear env so CLI uses normal defaults."""
+    global _adhoc_temp
     os.environ.pop("VERA_ADHOC_PARENT", None)
-    if _judge_logs_temp is not None and _judge_logs_temp.exists():
-        shutil.rmtree(_judge_logs_temp, ignore_errors=True)
-    _judge_logs_temp = None
     if _adhoc_temp is not None and _adhoc_temp.exists():
         shutil.rmtree(_adhoc_temp, ignore_errors=True)
     _adhoc_temp = None

tests/integration/test_evaluation_runner.py

@@ -32,7 +32,6 @@
 from vera_mh.judge.runner import (
     batch_evaluate_with_individual_judges,
     judge_conversations,
-    judge_single_conversation,
 )
 from vera_mh.utils.judge_results import build_dataframe_from_tsv_files
 
@@ -167,56 +166,6 @@ def create_mock_judge(*args, **kwargs: Dict[str, Any]) -> MockLLM:
         yield
 
 
-@pytest.mark.integration
-@pytest.mark.asyncio
-class TestJudgeSingleConversation:
-    """Tests for judge_single_conversation function.
-
-    Note: judge_single_conversation calls evaluate_conversation method
-    which doesn't exist in LLMJudge (it uses evaluate_conversation_question_flow).
-    This appears to be dead/unused code. Tests focus on actual working paths.
-    """
-
-    async def test_judge_single_conversation_file_not_found(
-        self,
-        mock_rubric_files: str,
-        mock_llm_factory_for_judge,
-        tmp_path: Path,
-    ):
-        """Test handling of missing conversation file.
-
-        Arrange: Create judge with valid rubric, use non-existent file
-        Act: Call judge_single_conversation with missing file
-        Assert: Raises FileNotFoundError when loading conversation
-        """
-        from vera_mh.judge.llm_judge import LLMJudge
-
-        output_folder = str(tmp_path / "evaluations")
-        missing_file = str(tmp_path / "nonexistent.txt")
-
-        # Load rubric config
-        rubric_config = await RubricConfig.load(
-            rubric_folder=mock_rubric_files,
-            rubric_file="rubric.tsv",
-            rubric_prompt_beginning_file="rubric_prompt_beginning.txt",
-            question_prompt_file="question_prompt.txt",
-        )
-
-        judge = LLMJudge(
-            judge_model="mock-judge",
-            rubric_config=rubric_config,
-        )
-
-        # Should raise FileNotFoundError when trying to load missing file
-        with pytest.raises(FileNotFoundError):
-            conversation = await ConversationData.load(missing_file)
-            await judge_single_conversation(
-                judge=judge,
-                conversation=conversation,
-                output_folder=output_folder,
-            )
-
-
 @pytest.mark.integration
 @pytest.mark.asyncio
 class TestBatchEvaluateWithIndividualJudges:

tests/unit/judge/test_runner_extra_params.py

@@ -279,23 +279,6 @@ async def test_resume_rebuilds_results_csv_from_all_tsvs_when_batch_empty(
 class TestJudgeTaskLogPath:
     """Per-task log paths align with evaluation TSV basenames."""
 
-    def test_build_judge_task_log_path_matches_tsv_stem(self, tmp_path: Path) -> None:
-        """Log file stem matches TSV basename from judge_evaluation_tsv_filename."""
-        conv = "subdir/abc.txt"
-        model = "gpt-4o"
-        tsv_name = judge_evaluation_tsv_filename(conv, model, 3)
-        expected_stem = Path(tsv_name).stem
-
-        log_path = build_judge_task_log_path(
-            conv,
-            model,
-            3,
-            run_key="j_run__convfolder",
-            logs_root=str(tmp_path),
-        )
-
-        assert log_path == str(tmp_path / "j_run__convfolder" / f"{expected_stem}.log")
-
     def test_build_judge_task_log_path_under_output_folder(
         self, tmp_path: Path
     ) -> None:
@@ -314,13 +297,3 @@ def test_build_judge_task_log_path_under_output_folder(
         )
 
         assert log_path == str(eval_dir / "logs" / f"{expected_stem}.log")
-
-    def test_build_judge_task_log_path_requires_run_key_for_legacy_layout(
-        self, tmp_path: Path
-    ) -> None:
-        with pytest.raises(ValueError, match="run_key"):
-            build_judge_task_log_path(
-                "a.txt",
-                "gpt-4o",
-                logs_root=str(tmp_path),
-            )

tests/unit/judge/test_utils.py

@@ -1,12 +1,10 @@
 """Unit tests for judge utility functions."""
 
 import argparse
-from datetime import datetime
 
 import pytest
 
 from vera_mh.judge.utils import (
-    build_single_conversation_judge_run_key,
     parse_judge_models,
 )
 
@@ -129,17 +127,3 @@ def test_count_alphanumeric_raises(self):
         judge_model = _setup_judge_model_arg(["-j", "gpt-4o:2x"])
         with pytest.raises(ValueError, match="must be an integer"):
             parse_judge_models(judge_model)
-
-
-@pytest.mark.unit
-class TestBuildSingleConversationJudgeRunKey:
-    """Tests for build_single_conversation_judge_run_key."""
-
-    def test_format_matches_cli_single_mode(self):
-        """Stem and truncated microsecond field match prior judge.py inline logic."""
-        fixed = datetime(2026, 4, 20, 15, 30, 45, 123456)
-        key = build_single_conversation_judge_run_key(
-            "/tmp/foo/Amelia_conv.txt",
-            now=fixed,
-        )
-        assert key == "single_20260420_153045_123__Amelia_conv"

tests/unit/utils/test_run_root.py

@@ -44,7 +44,6 @@
     ("judge/runner.py", "_worker"): "log line timestamps",
     ("judge/runner.py", "_run_workers_with_queue"): "log line timestamps",
     ("judge/runner.py", "judge_conversations"): "batch duration",
-    ("judge/utils.py", "build_single_conversation_judge_run_key"): "single mode",
     ("llm_clients/llm_interface.py", "_set_response_metadata"): "response time",
     ("utils/naming.py", "add_timestamp_to_path"): "comparison chart file names",
     ("utils/run_root.py", "run_started_at"): "the run's one clock read",