Standardisation 9

Readme.org

@@ -171,6 +171,12 @@ scripts migrés vers cette convention sont actuellement :
 
 - =copies_tools.py=, =grouping.py= et =verify_groups.py= ;
 - =post-correction.py= et =resolve_manual.py= ;
+- =page_splitter.py=, =cutleft.py=, =plotting.py= et
+  =splitting_int.py= ;
+- =gemini_for_labels.py= ;
+- =gemini_for_enonce.py= et =enonce_info.py= ;
+- =correction.py=, =submit_batches.py=, =batch_status.py= et
+  =fetch_batched_results.py= ;
 - =annotating.py=, =annotating_with_checks.py= et
   =annotating_by_label.py= ;
 - =reading_annotations.py= et =reading_grouped_annotations.py= ;
@@ -230,6 +236,12 @@ Dans le dossier de l'évaluation, mettre les fichiers suivants de l'évaluation
 
  - Alternative personnelle : `python enonce_info.py Interro`
 
+   Ces deux commandes suivent la convention des scripts standardisés.
+   Leur import ne lance aucun traitement et les erreurs partielles sont
+   distinguées des échecs. Les réponses d'extraction mises en cache par
+   =gemini_for_enonce.py= et le fichier =labels= sont publiés
+   atomiquement.
+
 ** Prétraitement des copies
 
 Mettre les copies scannées au format pdf dans =Interro=.
@@ -251,6 +263,12 @@ Mettre les copies scannées au format pdf dans =Interro=.
     + de déplacer la délimitation à droite/gauche
 
     Fix issues with =python page_splitter.py Interro14/Copies/Copie01.pdf=
+
+    Le PDF transformé est construit dans un dossier temporaire. La
+    copie produite et la sauvegarde dans =Copies Originales= sont
+    ensuite installées avec rollback : une erreur conserve les deux
+    versions précédentes. Une relance ciblée lit directement la
+    sauvegarde originale sans la déplacer au préalable.
  4. =python cutleft.py Interro=
 
     Découpe la partie gauche des copies, là où il devrait y avoir les
@@ -260,6 +278,10 @@ Mettre les copies scannées au format pdf dans =Interro=.
 
     Rerun on a single file with =python cutleft.py Interro/Copies/Copie01.pdf=
 
+    Les images et le fichier =_schema.json= d'une copie sont remplacés
+    ensemble. Fermer l'outil juste après la dernière validation ne peut
+    donc plus interrompre un thread de sauvegarde en arrière-plan.
+
 ** Labelisation et regroupement
 
 Optional : Set proxy with ~export HTTPS_PROXY="http://10.0.0.1:3128"~
@@ -268,6 +290,14 @@ Optional : Set proxy with ~export HTTPS_PROXY="http://10.0.0.1:3128"~
 
     Fait des requêtes à Gemini pour identifier les labels des
     questions dans images générées à partir des parties gauches des copies.
+
+    Une copie PDF ou une image précise de =Cutleft= peut également être
+    ciblée. Plusieurs cibles de la même évaluation sont acceptées. Les
+    parties d'une copie restent traitées séquentiellement afin de
+    conserver les labels précédents comme contexte, tandis que les
+    copies différentes sont traitées en parallèle. Chaque réponse JSON
+    validée est écrite atomiquement. Une cible sans image correspondante
+    produit le code de sortie 4.
  2. =python plotting.py Interro=
 
     Permet de vérifier visuellement les labels trouvés.
@@ -282,6 +312,10 @@ Optional : Set proxy with ~export HTTPS_PROXY="http://10.0.0.1:3128"~
     Pour modifier une seule copie :
     =python plotting.py Interro/Copies/Copie01.pdf=
 
+    Les coordonnées agrégées sont écrites atomiquement dans le JSON de
+    la copie. Fermer la fenêtre avant la fin d'une copie ne remplace pas
+    son JSON par un résultat incomplet.
+
     It also generates les =Copie01.json=, à partir des =Copie01_01.json=
     En cas de soucis, (par exemple les pages ne sont pas dans le bon ordre)
      - Réordonner les pages du fichier pdf
@@ -291,6 +325,11 @@ Optional : Set proxy with ~export HTTPS_PROXY="http://10.0.0.1:3128"~
 
     Découpe les copies suivant les exercices
     Peut-être appelé avec une seule copie.
+
+    Les réponses d'une copie sont préparées dans un dossier temporaire,
+    puis remplacent ensemble le dossier précédent. En cas d'erreur,
+    l'ancienne version est conservée. Les réponses devenues obsolètes
+    restent archivées dans le sous-dossier =Missing=.
  4. =python grouping.py Interro=
 
     Regroupe les mêmes questions de différentes copies en groupes de
@@ -313,6 +352,13 @@ Optional : Set proxy with ~export HTTPS_PROXY="http://10.0.0.1:3128"~
 
     Fais les requêtes de correction à Gemini.
 
+    =correction.py= peut être relancé sans supprimer son état. Les
+    fichiers =correction.json= et =correction_progress.json= sont mis à
+    jour atomiquement. Avec =--overwrite=, leur version précédente reste
+    en place jusqu'à la première écriture réussie de la nouvelle
+    exécution. =--reset= est la seule option qui supprime explicitement
+    cet état et restaure les fichiers =*_old.pdf=.
+
     L'argument =limit= limite le nombre de requêtes à Gemini Pro
     (chères), pour une version low cost, passer =--limit 0=, toutes
     les requêtes seront sur Gemini Flash.
@@ -327,6 +373,16 @@ Optional : Set proxy with ~export HTTPS_PROXY="http://10.0.0.1:3128"~
     + =python batch_status.py=
     + =python fetch_batched_results.py Interro=
     + =python correction.py Interro --deal-with-batched=
+
+    Les quatre commandes de ce flux suivent la convention des scripts
+    standardisés. Les fichiers de requêtes et le résultat JSONL combiné
+    sont publiés atomiquement : une interruption ne laisse pas de fichier
+    final partiellement écrit. =submit_batches.py= conserve aussi les
+    identifiants distants dans =batch_jobs.json= ; la récupération les
+    utilise en priorité et garde la recherche par nom pour les anciens
+    batchs. =batch_status.py --download JOB --output
+    resultat.jsonl= permet aussi de télécharger atomiquement le résultat
+    d'un job particulier.
  3. =python post-correction.py Interro=
 
     - Essaye de corriger des erreurs d'encodage/d'accents dans

batch_status.py

@@ -1,88 +1,93 @@
-import os
+from __future__ import annotations
-import sys
+
 import argparse
+from collections.abc import Sequence
+from pathlib import Path
+
 from google import genai
 
-if "GEMINI_API_KEY" not in os.environ:
+import config
-    sys.exit("Error: GEMINI_API_KEY environment variable not set.")
+from copienator import (
-
+    CliError,
-client = genai.Client()
+    ExitCode,
-
+    atomic_write_bytes,
-def list_jobs():
+    execute,
-    print("Fetching recent batch jobs...\n")
+    standard_parser,
-    try:
+)
-        batch_jobs = client.batches.list()
-        jobs_found = False
-
-        for job in batch_jobs:
-            jobs_found = True
-            state = job.state.name if hasattr(job.state, 'name') else job.state
-
-            print("-" * 60)
-            print(f"Job Name:      {job.name}")
-
-            if hasattr(job, 'display_name') and job.display_name:
-                print(f"Display Name:  {job.display_name}")
-
-            print(f"State:         {state}")
-
-            if state == 'JOB_STATE_FAILED' and hasattr(job, 'error'):
-                print(f"Error:         {job.error}")
-
-            if state == 'JOB_STATE_SUCCEEDED' and hasattr(job, 'dest') and job.dest:
-                if hasattr(job.dest, 'file_name') and job.dest.file_name:
-                    print(f"Output File:   {job.dest.file_name}")
-
-        if not jobs_found:
-            print("No batch jobs found.")
-        else:
-            print("-" * 60)
-            print("\nTo download a completed job, run:")
-            print("python batch_status.py --download batches/<YOUR_BATCH_ID>")
-
-    except Exception as e:
-        sys.exit(f"An error occurred while listing jobs: {e}")
 
 
-def download_job(job_name):
+def _client():
-    print(f"Checking status for {job_name}...\n")
+    if not config.API_KEY:
-    try:
+        raise CliError("GEMINI_API_KEY is not configured")
-        job = client.batches.get(name=job_name)
+    return genai.Client(api_key=config.API_KEY)
-        state = job.state.name if hasattr(job.state, 'name') else job.state
 
-        print(f"State: {state}")
 
-        if state != 'JOB_STATE_SUCCEEDED':
+def list_jobs(*, client=None) -> ExitCode:
-            print("Job is not ready yet or has failed.")
+    client = client or _client()
-            if state == 'JOB_STATE_FAILED' and hasattr(job, 'error'):
+    print("Fetching recent batch jobs...")
-                print(f"Error: {job.error}")
+    jobs = list(client.batches.list())
-            return
+    for job in jobs:
+        state = job.state.name if hasattr(job.state, "name") else job.state
+        print(f"{job.name}: {state}")
+        if getattr(job, "display_name", None):
+            print(f"  Display name: {job.display_name}")
+        if state == "JOB_STATE_FAILED" and getattr(job, "error", None):
+            print(f"  Error: {job.error}")
+        destination = getattr(job, "dest", None)
+        if state == "JOB_STATE_SUCCEEDED" and getattr(
+            destination, "file_name", None
+        ):
+            print(f"  Output file: {destination.file_name}")
+    if not jobs:
+        print("No batch jobs found.")
+    return ExitCode.SUCCESS
 
-        if hasattr(job, 'dest') and job.dest and hasattr(job.dest, 'file_name') and job.dest.file_name:
-            result_file_name = job.dest.file_name
-            print(f"Downloading results from {result_file_name}...")
 
-            file_content_bytes = client.files.download(file=result_file_name)
+def download_job(
-            output_path = f"results_{job_name.replace('/', '_')}.jsonl"
+    job_name: str,
+    *,
+    output: Path | None = None,
+    client=None,
+) -> ExitCode:
+    client = client or _client()
+    job = client.batches.get(name=job_name)
+    state = job.state.name if hasattr(job.state, "name") else job.state
+    print(f"State: {state}")
+    if state != "JOB_STATE_SUCCEEDED":
+        if state == "JOB_STATE_FAILED" and getattr(job, "error", None):
+            print(f"Error: {job.error}")
+        return ExitCode.PARTIAL
+    destination = getattr(job, "dest", None)
+    file_name = getattr(destination, "file_name", None)
+    if not file_name:
+        print("Job succeeded but no output file was found.")
+        return ExitCode.PARTIAL
+    payload = client.files.download(file=file_name)
+    output_path = output or Path(f"results_{job_name.replace('/', '_')}.jsonl")
+    atomic_write_bytes(output_path, payload)
+    print(f"Saved batch results to {output_path}")
+    return ExitCode.SUCCESS
 
-            with open(output_path, "wb") as f:
-                f.write(file_content_bytes)
 
-            print(f"Success! Saved to {output_path}")
+def build_parser() -> argparse.ArgumentParser:
-            print(f"You can now feed this to your correction script using: --deal-with-batched {output_path}")
+    parser = standard_parser("List or download Gemini correction batch jobs")
-        else:
+    parser.add_argument("--download", metavar="JOB_NAME")
-            print("Job succeeded but no output file was found.")
+    parser.add_argument("--output", type=Path, help="Downloaded JSONL destination")
+    return parser
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+
+    def handle(args: argparse.Namespace) -> ExitCode:
+        if args.output is not None and not args.download:
+            raise CliError("--output requires --download", ExitCode.INVALID_ARGUMENTS)
+        if args.download:
+            return download_job(args.download, output=args.output)
+        return list_jobs()
+
+    return execute(parser, argv, handle)
 
-    except Exception as e:
-        sys.exit(f"An error occurred while fetching the job: {e}")
 
 if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Manage Gemini Batch Jobs")
+    raise SystemExit(main())
-    parser.add_argument("--download", type=str, metavar="JOB_NAME",
-                        help="Download the results for a specific batch job (e.g. batches/123456)")
-    args = parser.parse_args()
-
-    if args.download:
-        download_job(args.download)
-    else:
-        list_jobs()

copienator/__init__.py

@@ -14,6 +14,7 @@ from .cli import (
 from .json_io import (
     JsonLockTimeout,
     atomic_update_json,
+    atomic_write_bytes,
     atomic_write_json,
     atomic_write_text,
     read_json,
@@ -32,6 +33,7 @@ __all__ = [
     "WorkspaceNotFoundError",
     "WorkspaceValidationError",
     "atomic_update_json",
+    "atomic_write_bytes",
     "atomic_write_json",
     "atomic_write_text",
     "evaluation_parser",

copienator/filesystem.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import shutil
 import uuid
+from collections.abc import Iterable
 from contextlib import contextmanager
 from pathlib import Path
 
@@ -44,7 +45,11 @@ def staged_directory(destination: str | Path):
 
 
 @contextmanager
-def staged_files(destination: str | Path):
+def staged_files(
+    destination: str | Path,
+    *,
+    remove: Iterable[str] = (),
+):
     """Stage a set of files and merge them into a directory with rollback."""
     target = Path(destination)
     target.parent.mkdir(parents=True, exist_ok=True)
@@ -56,9 +61,15 @@ def staged_files(destination: str | Path):
     try:
         yield staging
         staged = sorted(path for path in staging.iterdir() if path.is_file())
+        staged_names = {path.name for path in staged}
+        removed_names = set(remove) - staged_names
         target.mkdir(parents=True, exist_ok=True)
         backup.mkdir()
         try:
+            for name in sorted(removed_names):
+                destination_path = target / name
+                if destination_path.is_file() or destination_path.is_symlink():
+                    destination_path.replace(backup / name)
             for source in staged:
                 destination_path = target / source.name
                 if destination_path.exists() or destination_path.is_symlink():

copienator/json_io.py

@@ -78,6 +78,10 @@ def atomic_write_text(
     _atomic_write(Path(path), text.encode(encoding))
 
 
+def atomic_write_bytes(path: str | Path, payload: bytes) -> None:
+    _atomic_write(Path(path), payload)
+
+
 def atomic_write_json(
     path: str | Path,
     value: JsonValue,

copienator/workspace.py

@@ -101,6 +101,14 @@ class EvaluationWorkspace:
     def correction_progress_file(self) -> Path:
         return self.root / "correction_progress.json"
 
+    @property
+    def batch_jobs_file(self) -> Path:
+        return self.root / "batch_jobs.json"
+
+    @property
+    def batched_correction_result_file(self) -> Path:
+        return self.root / "batched_correction_result.jsonl"
+
     @property
     def manual_resolutions_file(self) -> Path:
         return self.root / "manual_resolutions.txt"

correction.py

@@ -1,91 +1,59 @@
-import sys
+from __future__ import annotations
-import os
+
-import time
-from pathlib import Path
 import argparse
-import prompting
-import signal
-from google import genai
 import base64
-import shlex
-import json
-import threading
 import concurrent.futures
+import json
+import os
+import shlex
+import shutil
+import sys
+import threading
+import time
+from collections.abc import Sequence
+from pathlib import Path
 
-from copienator import atomic_write_json
+from google import genai
-
-if len(sys.argv) < 2:
-    sys.exit("Usage: python script.py 'InterroTest/Ex 2/Group_1.jpg' OR <InputDir> OR 'file1' 'file2'")
-
-# Parse Arguments
-parser = argparse.ArgumentParser()
-parser.add_argument("paths", nargs="+", help="List of images or directories")
-parser.add_argument("--overwrite", action="store_true",
-                    help="Force redo requests even if output exists")
-parser.add_argument("--limit", type=int, help="limit calls to gemini rpo integer")
-parser.add_argument("--refaire", action="store_true",
-                    help="Redo specific copies/labels defined in refaire.json")
-parser.add_argument("--batch", action="store_true",
-                    help="Generate a JSONL file of requests to send to the Gemini Batch API")
-parser.add_argument("--batch-from", type=str, metavar="LABEL",
-                    help="Do live requests before LABEL, and batch requests from LABEL onwards")
-parser.add_argument("--deal-with-batched", action="store_true",
-                    help="Process a JSONL file containing completed batch results")
-parser.add_argument("--reset", action="store_true",
-                    help="Remove correction.json, revert _old.pdf, delete _new.pdf, then exit")
-args, _ = parser.parse_known_args()
-
-tasks = [] # List of tuples: (filepath_str, label_str)
-results = {}
-
-
-for path_str in args.paths:
-    arg_path = Path(path_str)
-
-    if not arg_path.exists():
-        print(f"Warning: {path_str} not found. Skipping.")
-        continue
-
-    if arg_path.is_file() and arg_path.suffix.lower() == ".jpg":
-        # Handle individual file
-        # Note: assumes structure InterroTest/Ex 2/Group_1.jpg
-        label = arg_path.parent.name
-        INPUT_DIR = arg_path.parent.parent.parent
-        COPIES_DIR = INPUT_DIR / "Copies"
-        GROUPS_DIR = INPUT_DIR / "Par label"
-        tasks.append((str(arg_path), label))
-        if label not in results:
-            results[label] = []
-
-    elif arg_path.is_dir():
-        INPUT_DIR = arg_path
-        COPIES_DIR = INPUT_DIR / "Copies"
-        GROUPS_DIR = INPUT_DIR / "Par label"
-        # Handle directory (original behavior)
-        for sub in GROUPS_DIR.iterdir():
-            if sub.is_dir():
-                label = sub.name
-                if label not in results:
-                    results[label] = []
-                for img in sub.glob("*.jpg"):
-                    tasks.append((str(img), label))
 
+import config
+import grouping
+import prompting
+from copienator import (
+    CliError,
+    EvaluationWorkspace,
+    ExitCode,
+    atomic_write_json,
+    atomic_write_text,
+    execute,
+    read_json,
+    target_parser,
+    workspace_from_target,
+)
+from utils import enonce_total, read_all_labels
 
 NB_THREADS = 12
 
 # PROXY_URL = "http://192.168.241.1:3128"
-PROXY_URL = None
-
-if PROXY_URL:
-    os.environ["http_proxy"] = PROXY_URL
-    os.environ["https_proxy"] = PROXY_URL
-
-import config    
-
 MODEL_ID_pro = config.MODEL_PRO_ID
 MODEL_ID_flash = config.MODEL_FLASH_ID
 api_key = config.API_KEY
 
+# Runtime globals retained while the processing helpers are migrated incrementally.
+INPUT_DIR = Path()
+COPIES_DIR = Path()
+GROUPS_DIR = Path()
+output_path = Path()
+progress_path = Path()
+tasks: list[tuple] = []
+tasks_to_process: list[tuple] = []
+results: dict = {}
+completed_tasks: list = []
+errors_summary: list = []
+overwrite = False
+limit = None
+client = None
+start_time = 0.0
+
 # --- Thread-safe Logging ---
 log_lock = threading.Lock()
 thread_logs = {}
@@ -113,57 +81,6 @@ def flush_thread_log(tid=None):
                 f.write("\n".join(thread_logs[tid]) + "\n\n")
             thread_logs[tid].clear()
 
-def handle_interrupt(sig, frame):
-    """Flush all partial/unfinished logs if program is interrupted."""
-    print("\nInterrupt received. Flushing partial logs...", file=sys.stderr)
-    for tid in list(thread_logs.keys()):
-        flush_thread_log(tid)
-    sys.exit(1)
-
-
-signal.signal(signal.SIGINT, handle_interrupt)
-signal.signal(signal.SIGTERM, handle_interrupt)
-# ---------------------------
-
-client = genai.Client(api_key=api_key)
-output_path = INPUT_DIR / "correction.json"
-progress_path = INPUT_DIR / "correction_progress.json"
-
-if args.reset:
-    print("--- Running Reset ---")
-    if output_path.exists():
-        output_path.unlink()
-        print(f"Deleted: {output_path}")
-    if progress_path.exists():
-        progress_path.unlink()
-        print(f"Deleted: {progress_path}")
-
-    if COPIES_DIR.exists():
-        for copie_dir in COPIES_DIR.iterdir():
-            if not copie_dir.is_dir():
-                continue
-
-            # Revert _old.pdf files
-            for old_pdf in copie_dir.glob("*_old.pdf"):
-                orig_pdf = copie_dir / old_pdf.name.replace("_old.pdf", ".pdf")
-                if orig_pdf.exists():
-                    orig_pdf.unlink() # Prevent FileExistsError on Windows
-                old_pdf.rename(orig_pdf)
-                print(f"Moved: {copie_dir.name}/{old_pdf.name} -> {orig_pdf.name}")
-
-            # Delete _new.pdf files
-            for new_pdf in copie_dir.glob("*_new.pdf"):
-                new_pdf.unlink()
-                print(f"Deleted: {copie_dir.name}/{new_pdf.name}")
-
-    sys.exit("Reset almost complete. For each deleted `_new`, you should manually delete the group in `Par label`")
-
-start_time = time.time()
-overwrite = args.overwrite
-limit = args.limit
-completed_tasks = []
-errors_summary = []
-
 # --- Lock for thread-safe file writing ---
 io_lock = threading.Lock()
 pro_lock = threading.Lock()
@@ -171,21 +88,119 @@ pro_count = 0
 flash_count = 0
 pro_quota_exhausted = False
 
-if overwrite:
-    if output_path.exists():
-        output_path.unlink()
-    if progress_path.exists():
-        progress_path.unlink()
-else:
-    if progress_path.exists():
-        with open(progress_path, "r", encoding="utf-8") as f:
-            completed_tasks = json.load(f)
-    if output_path.exists():
-        with open(output_path, "r", encoding="utf-8") as f:
-            results = json.load(f)
 
-completed_set = set((str(f), l) for f, l in completed_tasks)
+def discover_tasks(
-tasks_to_process = [t for t in tasks if (str(t[0]), t[1]) not in completed_set]
+    workspace: EvaluationWorkspace,
+    targets: list[Path],
+) -> tuple[list[tuple[str, str]], list[str]]:
+    workspace.require_directories("Copies", "Par label")
+    discovered: list[tuple[str, str]] = []
+    warnings: list[str] = []
+    for target in targets:
+        if target.is_file():
+            if target.suffix.casefold() != ".jpg":
+                raise CliError(
+                    f"Correction target is not a group JPG: {target}",
+                    ExitCode.INVALID_ARGUMENTS,
+                )
+            try:
+                target.relative_to(workspace.groups_dir)
+            except ValueError as exc:
+                raise CliError(
+                    f"Group image is not inside {workspace.groups_dir}: {target}",
+                    ExitCode.INVALID_ARGUMENTS,
+                ) from exc
+            discovered.append((str(target), target.parent.name))
+            continue
+
+        group_directories = sorted(
+            (path for path in workspace.groups_dir.iterdir() if path.is_dir()),
+            key=lambda path: path.name.casefold(),
+        )
+        for group_directory in group_directories:
+            images = sorted(
+                group_directory.glob("*.jpg"), key=lambda path: path.name.casefold()
+            )
+            discovered.extend(
+                (str(image), group_directory.name) for image in images
+            )
+        if not group_directories:
+            warnings.append(f"No label groups found in {workspace.groups_dir}")
+    return list(dict.fromkeys(discovered)), warnings
+
+
+def configure_runtime(
+    workspace: EvaluationWorkspace,
+    discovered_tasks: list[tuple[str, str]],
+    args: argparse.Namespace,
+    *,
+    api_client=None,
+) -> None:
+    global INPUT_DIR, COPIES_DIR, GROUPS_DIR, output_path, progress_path
+    global tasks, tasks_to_process, results, completed_tasks, errors_summary
+    global overwrite, limit, client, start_time
+    global pro_count, flash_count, pro_quota_exhausted
+
+    INPUT_DIR = workspace.root
+    COPIES_DIR = workspace.copies_dir
+    GROUPS_DIR = workspace.groups_dir
+    output_path = workspace.correction_file
+    progress_path = workspace.correction_progress_file
+    tasks = list(discovered_tasks)
+    overwrite = bool(args.overwrite)
+    limit = args.limit
+    start_time = time.time()
+    errors_summary = []
+    completed_tasks = []
+    results = {label: [] for _file, label in tasks}
+    thread_logs.clear()
+    pro_count = 0
+    flash_count = 0
+    pro_quota_exhausted = False
+
+    if not overwrite:
+        if progress_path.is_file():
+            loaded_progress = read_json(progress_path)
+            if not isinstance(loaded_progress, list):
+                raise TypeError("correction_progress.json must contain a JSON array")
+            completed_tasks = loaded_progress
+        if output_path.is_file():
+            loaded_results = read_json(output_path)
+            if not isinstance(loaded_results, dict):
+                raise TypeError("correction.json must contain a JSON object")
+            results = loaded_results
+
+    completed_set = {(str(file_path), label) for file_path, label in completed_tasks}
+    tasks_to_process = [
+        task for task in tasks if (str(task[0]), task[1]) not in completed_set
+    ]
+    client = api_client
+
+
+def reset_workspace(workspace: EvaluationWorkspace) -> None:
+    """Apply the explicitly requested correction reset."""
+    print("--- Running Reset ---")
+    for path in (workspace.correction_file, workspace.correction_progress_file):
+        if path.exists():
+            path.unlink()
+            print(f"Deleted: {path}")
+    if workspace.copies_dir.is_dir():
+        for copy_directory in workspace.copies_dir.iterdir():
+            if not copy_directory.is_dir():
+                continue
+            for old_pdf in copy_directory.glob("*_old.pdf"):
+                original = old_pdf.with_name(old_pdf.name.replace("_old.pdf", ".pdf"))
+                if original.exists():
+                    original.unlink()
+                old_pdf.replace(original)
+                print(f"Moved: {copy_directory.name}/{old_pdf.name} -> {original.name}")
+            for new_pdf in copy_directory.glob("*_new.pdf"):
+                new_pdf.unlink()
+                print(f"Deleted: {copy_directory.name}/{new_pdf.name}")
+    print(
+        "Reset almost complete. Manually remove groups associated with deleted "
+        "_new PDFs from 'Par label'."
+    )
 
 def call_gemini_with_retries(model_id, contents, config,
                              fallback_model_id=MODEL_ID_flash):
@@ -267,9 +282,6 @@ def correct_boxes_with_gemini(pid, label, pdf_path, original_feedbacks,
 
     return global_feedbacks + corrected_feedbacks
 
-import shutil
-import grouping
-
 def get_next_group_idx(label):
     """Finds the next available Group index for a given label."""
     target_folder = GROUPS_DIR / label
@@ -278,8 +290,6 @@ def get_next_group_idx(label):
     if not existing: return 0
     return max([int(f.stem.split("_")[1]) for f in existing])
 
-from utils import read_all_labels, enonce_total
-
 def handle_label_errors(pid, label, res, pdf_path):
     """Handles Gemini labeling errors, moves/copies files, and returns new tasks."""
     new_tasks = []
@@ -332,7 +342,7 @@ def handle_label_errors(pid, label, res, pdf_path):
         tprint(f"\tHandling additional-answer for {pid} {label}")
         try:
             add_labels = json.loads(call_gemini_with_retries(MODEL_ID_flash, contents, config))
-        except Exception:
+        except Exception:  # noqa: BLE001 - invalid auxiliary model response
             add_labels = []
 
         keep_error = False
@@ -374,7 +384,7 @@ def handle_label_errors(pid, label, res, pdf_path):
 
 def process_single_task(task_tuple, precomputed_response=None):
     try:
-        global pro_count, flash_count, pro_quota_exhausted
+        global pro_count, flash_count
         file_path = task_tuple[0]
         label = task_tuple[1]
         can_spawn_tasks = task_tuple[2] if len(task_tuple) > 2 else True
@@ -383,8 +393,7 @@ def process_single_task(task_tuple, precomputed_response=None):
         json_path = group_name + '.json'
         new_tasks = []
 
-        with open(json_path, 'r') as f:
+        group_data = read_json(json_path)
-            group_data = json.load(f)
 
         n = len(group_data)
         d_data = {l[0]: (l[1], l[2], l[3]) for l in group_data}
@@ -466,7 +475,7 @@ def process_single_task(task_tuple, precomputed_response=None):
                 for (i,f) in enumerate(res["feedback"]):
                     b = f.get("box_2d")
                     if b:
-                        ymin, xmin, ymax, xmax = b
+                        ymin, _xmin, ymax, xmax = b
                         ymin = ymin * total_height // 1000
                         ymax = ymax * total_height // 1000
 
@@ -493,7 +502,7 @@ def process_single_task(task_tuple, precomputed_response=None):
                         res["feedback"] = correct_boxes_with_gemini(
                             pid, label, pdf_path, res["feedback"],
                             yming, ymaxg, width_r, total_height)
-                    except Exception as e:
+                    except Exception as e:  # noqa: BLE001 - correction fallback
                         tprint(f"\tCorrection failed for Copie {pid}, {group_name} : {e}\n\tRemoving the boxes")
                         # Fallback if the second request fails entirely
                         for (i, f) in enumerate(res["feedback"]):
@@ -516,7 +525,7 @@ def process_single_task(task_tuple, precomputed_response=None):
             tprint(f"Error decoding JSON for {file_path}", file=sys.stderr)
             with io_lock:
                 errors_summary.append(("Error decoding JSON response", file_path))
-        except Exception as e:
+        except Exception as e:  # noqa: BLE001 - per-task processing boundary
             error_msg = f"Exception processing {file_path}: {e}"
             print(error_msg, file=sys.stderr)
             with io_lock:
@@ -602,19 +611,20 @@ def resolve_delayed_moves():
 
     return new_tasks
 
-if __name__ == "__main__":
+def run_configured(args: argparse.Namespace) -> ExitCode:
+    global client, tasks_to_process
+    if client is None:
+        client = genai.Client(api_key=api_key)
     if args.refaire:
         refaire_path = INPUT_DIR / "refaire.json"
         overwritten_path = INPUT_DIR / "overwritten_correction.json"
 
         if refaire_path.exists():
-            with open(refaire_path, "r", encoding="utf-8") as f:
+            refaire_list = read_json(refaire_path)
-                refaire_list = json.load(f)
 
             overwritten_data = []
             if overwritten_path.exists():
-                with open(overwritten_path, "r", encoding="utf-8") as f:
+                overwritten_data = read_json(overwritten_path)
-                    overwritten_data = json.load(f)
 
             dirty_results = False
 
@@ -659,10 +669,12 @@ if __name__ == "__main__":
                     pdf_path = copie_dir / f"{label}.pdf"
                     is_new = False
 
-                    if not pdf_path.exists():
+                    if (
-                        if (copie_dir / f"{label}_new.pdf").exists():
+                        not pdf_path.exists()
-                            pdf_path = copie_dir / f"{label}_new.pdf"
+                        and (copie_dir / f"{label}_new.pdf").exists()
-                            is_new = True
+                    ):
+                        pdf_path = copie_dir / f"{label}_new.pdf"
+                        is_new = True
                         # elif (copie_dir / f"{label}_old.pdf").exists():
                             # pdf_path = copie_dir / f"{label}_old.pdf"
 
@@ -690,7 +702,11 @@ if __name__ == "__main__":
                     input(f"About to batch from: {args.batch_from}. Press Enter to confirm...")
                     break
             if args.batch_from not in all_labels:
-                sys.exit(f"Error: Label '{args.batch_from}' not found. Available labels: {all_labels}")
+                raise CliError(
+                    f"Label '{args.batch_from}' not found. Available labels: "
+                    f"{all_labels}",
+                    ExitCode.INVALID_ARGUMENTS,
+                )
 
             target_idx = all_labels.index(args.batch_from)
             live_tasks = []
@@ -714,58 +730,66 @@ if __name__ == "__main__":
 
             count_flash = 0
             count_pro = 0
-
+            flash_lines = []
-            with open(batch_flash_file, "w", encoding="utf-8") as f_flash, \
+            pro_lines = []
-                 open(batch_pro_file, "w", encoding="utf-8") as f_pro:
+            for task in batch_tasks:
-
+                file_path, label = task[0], task[1]
-                for task in batch_tasks:
+                json_path = Path(file_path).with_suffix(".json")
-                    file_path, label = task[0], task[1]
+                group_data = read_json(json_path)
-                    group_name = os.path.splitext(file_path)[0]
+                use_flash = len(group_data) >= 4 or group_data[-1][2] <= 500
-                    json_path = group_name + '.json'
+                b64_img = base64.b64encode(Path(file_path).read_bytes()).decode(
-
+                    "utf-8"
-                    with open(json_path, 'r') as jf:
+                )
-                        group_data = json.load(jf)
+                request = {
-                    use_flash = len(group_data) >= 4 or group_data[-1][2] <= 500
+                    "key": file_path,
-
+                    "request": {
-                    image_data = Path(file_path).read_bytes()
+                        "contents": [
-                    b64_img = base64.b64encode(image_data).decode("utf-8")
+                            {
-
-                    # Format payload matching Gemini Batch API file requirements
-                    req = {
-                        "key": file_path,  # The ID returned in the output file
-                        "request": {
-                            "contents": [{
                                 "role": "user",
                                 "parts": [
-                                    {"inlineData": {"mimeType": "image/jpeg", "data": b64_img}},
+                                    {
-                                    {"text": prompting.make_prompt(INPUT_DIR,label)}
+                                        "inlineData": {
-                                ]
+                                            "mimeType": "image/jpeg",
-                            }],
+                                            "data": b64_img,
-                            "generation_config": {
+                                        }
-                                "temperature": 1.0,
+                                    },
-                                "topP": 0.95,
+                                    {"text": prompting.make_prompt(INPUT_DIR, label)},
-                                "maxOutputTokens": 65535,
+                                ],
-                                "responseMimeType": "application/json",
-                                "responseSchema": prompting.UNROLLED_SCHEMA
                             }
-                        }
+                        ],
-                    }
+                        "generation_config": {
+                            "temperature": 1.0,
+                            "topP": 0.95,
+                            "maxOutputTokens": 65535,
+                            "responseMimeType": "application/json",
+                            "responseSchema": prompting.UNROLLED_SCHEMA,
+                        },
+                    },
+                }
+                line = json.dumps(request)
+                if use_flash:
+                    flash_lines.append(line)
+                    count_flash += 1
+                else:
+                    pro_lines.append(line)
+                    count_pro += 1
+            atomic_write_text(
+                batch_flash_file,
+                "\n".join(flash_lines) + ("\n" if flash_lines else ""),
+            )
+            atomic_write_text(
+                batch_pro_file,
+                "\n".join(pro_lines) + ("\n" if pro_lines else ""),
+            )
 
-                    if use_flash:
+            print("Batch generation complete.")
-                        f_flash.write(json.dumps(req) + "\n")
-                        count_flash += 1
-                    else:
-                        f_pro.write(json.dumps(req) + "\n")
-                        count_pro += 1
-
-            print(f"Batch generation complete.")
             print(f" - {count_flash} requests saved to {batch_flash_file} (for {MODEL_ID_flash})")
             print(f" - {count_pro} requests saved to {batch_pro_file} (for {MODEL_ID_pro})")
             print("Upload these files via the File API and create two separate batch jobs.")
 
         # If there's no live tasks to do, and we aren't doing a batched ingestion, exit right away
         if not tasks_to_process and not args.deal_with_batched:
-            sys.exit(0)
+            return ExitCode.SUCCESS
 
     batched_responses = {}
     if args.deal_with_batched:
@@ -807,8 +831,11 @@ if __name__ == "__main__":
                         if new_generated_tasks:
                             for new_task in new_generated_tasks:
                                 futures[executor.submit(process_single_task, new_task)] = new_task
-                    except Exception as e:
+                    except Exception as e:  # noqa: BLE001 - future boundary
                         print(f"Exception during task execution: {e}", file=sys.stderr)
+                        failed_task = futures[future]
+                        with io_lock:
+                            errors_summary.append((str(e), failed_task[0]))
 
             tasks_to_process = [] # Vider la liste une fois traitée
 
@@ -840,9 +867,12 @@ if __name__ == "__main__":
 
     if unresolved_delayed:
         manual_path = INPUT_DIR / "manual_resolutions.txt"
-        with open(manual_path, "w", encoding="utf-8") as f:
+        atomic_write_text(
-            f.write("###  Use -> x>, -x, ss, sx, xx, xs\n")
+            manual_path,
-            f.write("\n".join(unresolved_delayed) + "\n")
+            "###  Use -> x>, -x, ss, sx, xx, xs\n"
+            + "\n".join(unresolved_delayed)
+            + "\n",
+        )
         print(f"\n[!] Unresolved delayed tasks found! Wrote to {manual_path}.")
         print("    Please edit it manually, then run `python resolve_manual.py <InputDir>`")
 
@@ -855,3 +885,101 @@ if __name__ == "__main__":
             print(err, file=sys.stderr)
             escaped_path = shlex.quote(str(file))
             print(f"Run :    python correction.py {escaped_path}")
+    return ExitCode.PARTIAL if errors_summary else ExitCode.SUCCESS
+
+
+def run(
+    workspace: EvaluationWorkspace,
+    targets: list[Path],
+    args: argparse.Namespace,
+    *,
+    api_client=None,
+) -> ExitCode:
+    if args.reset:
+        workspace.require_directories("Copies")
+        reset_workspace(workspace)
+        return ExitCode.SUCCESS
+    workspace.require_directories("Copies", "Par label")
+    workspace.require_files("labels")
+    if args.refaire:
+        workspace.require_files("refaire.json")
+    discovered, warnings = discover_tasks(workspace, targets)
+    for warning in warnings:
+        print(f"Warning: {warning}")
+    configure_runtime(workspace, discovered, args, api_client=api_client)
+    if not discovered and not args.refaire:
+        return ExitCode.PARTIAL
+    try:
+        status = run_configured(args)
+    finally:
+        for thread_id in list(thread_logs):
+            flush_thread_log(thread_id)
+    if warnings and status == ExitCode.SUCCESS:
+        return ExitCode.PARTIAL
+    return status
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = target_parser("Correct grouped answers with Gemini")
+    parser.add_argument(
+        "additional_targets",
+        nargs="*",
+        type=Path,
+        help="Additional group JPG files from the same evaluation",
+    )
+    parser.add_argument("--overwrite", action="store_true", help="Redo requests")
+    parser.add_argument("--limit", type=int, help="Maximum Gemini Pro calls")
+    parser.add_argument(
+        "--refaire",
+        action="store_true",
+        help="Redo copies and labels listed in refaire.json",
+    )
+    parser.add_argument(
+        "--batch",
+        action="store_true",
+        help="Generate Gemini batch request JSONL files",
+    )
+    parser.add_argument(
+        "--batch-from",
+        metavar="LABEL",
+        help="Process earlier labels live and batch from LABEL onward",
+    )
+    parser.add_argument(
+        "--deal-with-batched",
+        action="store_true",
+        help="Consume batched_correction_result.jsonl",
+    )
+    parser.add_argument(
+        "--reset",
+        action="store_true",
+        help="Delete correction state, restore _old PDFs, and delete _new PDFs",
+    )
+    return parser
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+
+    def handle(args: argparse.Namespace) -> ExitCode:
+        workspace, target = workspace_from_target(args)
+        targets = [target]
+        for additional in args.additional_targets:
+            resolved = additional.expanduser().resolve()
+            if not resolved.exists():
+                raise CliError(
+                    f"Target does not exist: {resolved}",
+                    ExitCode.INVALID_WORKSPACE,
+                )
+            if EvaluationWorkspace.discover(resolved).root != workspace.root:
+                raise CliError(
+                    "All targets must belong to the same evaluation",
+                    ExitCode.INVALID_ARGUMENTS,
+                )
+            targets.append(resolved)
+        return run(workspace, targets, args)
+
+    return execute(parser, argv, handle)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

cutleft.py

@@ -1,366 +1,343 @@
-import sys
+from __future__ import annotations
-from functools import lru_cache
+
-import os
-import time
-import json  # Added for schema output
 import argparse
+import threading
+import time
 import tkinter as tk
+from collections.abc import Sequence
+from functools import lru_cache
+from pathlib import Path
+from queue import Empty, Queue
 from threading import Thread
-from queue import Queue, Empty
+
 from pdf2image import convert_from_path
 from PIL import Image, ImageTk
 
-# --- Configuration ---
+from copienator import (
+    CliError,
+    EvaluationWorkspace,
+    ExitCode,
+    atomic_write_json,
+    execute,
+    target_parser,
+    workspace_from_target,
+)
+from copienator.filesystem import staged_files
+
 DELIMITER_WIDTH = 5
 DELIMITER_COLOR = (0, 0, 0)
 OUTPUT_SIZE = (1800, 1000)
-
-parser = argparse.ArgumentParser(description="PDF Cropper")
-parser.add_argument("path", help="Directory path or PDF file path")
-parser.add_argument("--fullpage", action="store_true", help="Process all files in full page mode (1 page per output file)")
-args = parser.parse_args()
-
-path_arg = args.path
-fullpage_mode = args.fullpage
-files = []
-INPUT_DIR = ""
-COPIES_DIR = ""
-
-if os.path.isfile(path_arg) and path_arg.lower().endswith('.pdf'):
-    COPIES_DIR = os.path.abspath(os.path.dirname(path_arg))
-    # If the file is inside a "Copies" folder, set INPUT_DIR to the parent
-    if os.path.basename(COPIES_DIR).lower() == 'copies':
-        INPUT_DIR = os.path.dirname(COPIES_DIR)
-    else:
-        INPUT_DIR = COPIES_DIR
-    files = [os.path.basename(path_arg)]
-elif os.path.isdir(path_arg):
-    # Support passing either the base dir or the Copies dir directly
-    abs_path = os.path.abspath(path_arg)
-    if os.path.basename(abs_path).lower() == 'copies':
-        COPIES_DIR = abs_path
-        INPUT_DIR = os.path.dirname(abs_path)
-    else:
-        INPUT_DIR = abs_path
-        COPIES_DIR = os.path.join(INPUT_DIR, 'Copies')
-
-    if os.path.exists(COPIES_DIR):
-        files = sorted([f for f in os.listdir(COPIES_DIR) if f.lower().endswith('.pdf') and
-                        "nonc" not in f.lower()])
-    else:
-        sys.exit(f"Error: Could not find 'Copies' directory inside {INPUT_DIR}")
-else:
-    sys.exit("Error: Input must be a directory or a PDF file.")
-
-
-OUTPUT_DIR = os.path.join(INPUT_DIR, 'Cutleft')
-
-if not os.path.exists(OUTPUT_DIR):
-    os.makedirs(OUTPUT_DIR)
-
-# --- Processing Logic ---
-
-def distribute_pages(total_pages, max_per_file=5):
-    """
-    Calculates how to split pages into chunks <= max_per_file,
-    balancing the number of columns per file.
-    Example: 12 pages, max 5 -> [4, 4, 4]
-    """
-    if total_pages == 0:
-        return []
-
-    # Calculate minimum number of files needed
-    num_files = (total_pages + max_per_file - 1) // max_per_file
-
-    # Calculate base size and remainder
-    base_count = total_pages // num_files
-    remainder = total_pages % num_files
-
-    distribution = []
-    for i in range(num_files):
-        # Distribute remainder to the first few files
-        count = base_count + (1 if i < remainder else 0)
-        distribution.append(count)
-
-    return distribution
-
-def stitch_images(image_list):
-    """Helper to stitch a list of images horizontally with delimiters."""
-    if not image_list:
-        return None
-
-    num_images = len(image_list)
-    total_width = sum(img.width for img in image_list) + (num_images - 1) * DELIMITER_WIDTH
-    max_height = max(img.height for img in image_list)
-
-    combined = Image.new('RGB', (total_width, max_height), color=(255, 255, 255))
-
-    x_offset = 0
-    for idx, img in enumerate(image_list):
-        combined.paste(img, (x_offset, 0))
-        x_offset += img.width
-        if idx < num_images - 1:
-            delimiter = Image.new('RGB', (DELIMITER_WIDTH, max_height), color=DELIMITER_COLOR)
-            combined.paste(delimiter, (x_offset, 0))
-            x_offset += DELIMITER_WIDTH
-
-    return combined
-
-import threading
 pdf_cache_lock = threading.Lock()
 
 
+def distribute_pages(total_pages: int, max_per_file: int = 5) -> list[int]:
+    """Distribute pages into balanced chunks no larger than max_per_file."""
+    if total_pages == 0:
+        return []
+    number_of_files = (total_pages + max_per_file - 1) // max_per_file
+    base_count, remainder = divmod(total_pages, number_of_files)
+    return [
+        base_count + (1 if index < remainder else 0)
+        for index in range(number_of_files)
+    ]
+
+
+def stitch_images(image_list: list[Image.Image]) -> Image.Image | None:
+    if not image_list:
+        return None
+    total_width = sum(image.width for image in image_list)
+    total_width += (len(image_list) - 1) * DELIMITER_WIDTH
+    max_height = max(image.height for image in image_list)
+    combined = Image.new("RGB", (total_width, max_height), color="white")
+    x_offset = 0
+    for index, image in enumerate(image_list):
+        combined.paste(image, (x_offset, 0))
+        x_offset += image.width
+        if index < len(image_list) - 1:
+            delimiter = Image.new(
+                "RGB", (DELIMITER_WIDTH, max_height), color=DELIMITER_COLOR
+            )
+            combined.paste(delimiter, (x_offset, 0))
+            x_offset += DELIMITER_WIDTH
+    return combined
+
+
 @lru_cache(maxsize=3)
-def _get_pdf_pages_cached(filename):
+def _get_pdf_pages_cached(pdf_path: Path) -> list[Image.Image]:
-    pdf_path = os.path.join(COPIES_DIR, filename)
     return convert_from_path(pdf_path)
 
-def get_pdf_pages(filename):
+
-    """Thread-safe wrapper for the cached PDF conversion."""
+def get_pdf_pages(pdf_path: Path) -> list[Image.Image]:
+    """Thread-safe wrapper around the small PDF conversion cache."""
     with pdf_cache_lock:
-        return _get_pdf_pages_cached(filename)
+        return _get_pdf_pages_cached(pdf_path)
 
-def process_single_pdf(filename, shift_offset=0, max_per_file=5):
+
-    """
+def process_single_pdf(
-    Converts PDF to stitched images.
+    pdf_path: Path,
-    Returns a tuple: (preview_image_resized, list_of_split_images, schema_dict)
+    shift_offset: int = 0,
-    """
+    max_per_file: int = 5,
+) -> tuple[Image.Image, list[Image.Image], dict[str, object]] | None:
+    """Convert one PDF into a preview, full-resolution splits and metadata."""
     try:
-        pages = get_pdf_pages(filename)
         cropped_images = []
-
+        for image in get_pdf_pages(pdf_path):
-        for img in pages:
+            width, height = image.size
-            width, height = img.size
-
             if max_per_file == 1:
-                # If Single Page mode, take the full width (ignore shift/crop)
+                left, right = 0, width
-                left = 0
-                right = width
             else:
-                # Original "Cutleft" logic (approx 1/3 width)
+                left = max(0, 100 + shift_offset)
-                left = 100 + shift_offset
+                right = min(width, width // 3 + 100 + shift_offset)
-                right = (width // 3) + 100 + shift_offset
-
-            # Ensure crop box is valid
-            left = max(0, left)
-            right = min(width, right)
-
             if right > left:
-                crop_box = (left, 0, right, height)
+                cropped_images.append(image.crop((left, 0, right, height)))
-                cropped = img.crop(crop_box)
-                cropped_images.append(cropped)
-
         if not cropped_images:
             return None
 
-        # 1. Generate Schema / Distribution
+        distribution = distribute_pages(len(cropped_images), max_per_file)
-        col_distribution = distribute_pages(len(cropped_images), max_per_file=max_per_file)
-
-        # 2. Generate Split Images (Full Resolution)
         split_images = []
-        current_idx = 0
+        current_index = 0
-        for count in col_distribution:
+        for count in distribution:
-            chunk = cropped_images[current_idx : current_idx + count]
+            stitched = stitch_images(cropped_images[current_index : current_index + count])
-            stitched_chunk = stitch_images(chunk)
+            if stitched is not None:
-            split_images.append(stitched_chunk)
+                split_images.append(stitched)
-            current_idx += count
+            current_index += count
-
-        # 3. Generate Preview (All stitched together, Resized)
         full_stitch = stitch_images(cropped_images)
-        preview_resized = full_stitch.resize(OUTPUT_SIZE, Image.BILINEAR)
+        if full_stitch is None:
-
+            return None
-        schema = {
+        preview = full_stitch.resize(OUTPUT_SIZE, Image.Resampling.BILINEAR)
-            "original_filename": filename,
+        schema: dict[str, object] = {
+            "original_filename": pdf_path.name,
             "total_pages": len(cropped_images),
             "number_of_files": len(split_images),
-            "columns_per_file": col_distribution
+            "columns_per_file": distribution,
         }
-
+        return preview, split_images, schema
-        return (preview_resized, split_images, schema)
+    except Exception as exc:  # noqa: BLE001 - interactive item failure
-
+        print(f"Error processing {pdf_path.name}: {exc}")
-    except Exception as e:
-        print(f"Error processing {filename}: {e}")
         return None
 
-def save_results(result_tuple, filename):
-    """
-    Saves the split images and the schema JSON.
-    """
-    _, splits, schema = result_tuple
-    base_name = os.path.splitext(filename)[0]
 
-    # --- Cleanup: Delete existing files for this PDF ---
+def _previous_cutleft_outputs(output_dir: Path, base_name: str) -> set[str]:
-    for f in os.listdir(OUTPUT_DIR):
+    if not output_dir.is_dir():
-        file_path = os.path.join(OUTPUT_DIR, f)
+        return set()
-        if f == f"{base_name}_schema.json":
+    result = {f"{base_name}_schema.json"}
-            os.remove(file_path)
+    for path in output_dir.glob(f"{base_name}_*.jpg"):
-        elif f.startswith(f"{base_name}_") and f.endswith(".jpg"):
+        suffix = path.stem.removeprefix(f"{base_name}_")
-            suffix = f[len(base_name)+1:-4]
+        if suffix.isdigit():
-            if suffix.isdigit():
+            result.add(path.name)
-                os.remove(file_path)
+    return result
-
-    # Save Images
-    for i, img in enumerate(splits):
-        suffix = f"_{i+1:02d}"
-        output_filename = f"{base_name}{suffix}.jpg"
-        output_path = os.path.join(OUTPUT_DIR, output_filename)
-        img.save(output_path, "JPEG", quality=95)
-        print(f"Saved: {output_filename}")
-
-    # Save Schema
-    json_filename = f"{base_name}_schema.json"
-    json_path = os.path.join(OUTPUT_DIR, json_filename)
-    with open(json_path, 'w') as f:
-        json.dump(schema, f, indent=4)
-    print(f"Saved schema: {json_filename}")
 
 
-# --- GUI Application ---
+def save_results(
+    result: tuple[Image.Image, list[Image.Image], dict[str, object]],
+    pdf_path: Path,
+    output_dir: Path,
+) -> None:
+    """Atomically replace every Cutleft output associated with one copy."""
+    _, splits, schema = result
+    base_name = pdf_path.stem
+    previous = _previous_cutleft_outputs(output_dir, base_name)
+    with staged_files(output_dir, remove=previous) as staging:
+        for index, image in enumerate(splits, start=1):
+            filename = f"{base_name}_{index:02d}.jpg"
+            image.save(staging / filename, "JPEG", quality=95)
+        atomic_write_json(staging / f"{base_name}_schema.json", schema)
+    for index in range(1, len(splits) + 1):
+        print(f"Saved: {base_name}_{index:02d}.jpg")
+    print(f"Saved schema: {base_name}_schema.json")
+
 
 class ImageReviewer:
-    def __init__(self, file_list, default_max_per_file=5):
+    def __init__(
-        self.files = file_list
+        self,
+        files: list[Path],
+        output_dir: Path,
+        default_max_per_file: int = 5,
+    ) -> None:
+        self.files = files
+        self.output_dir = output_dir
         self.index = 0
         self.current_shift = 0
         self.default_max_per_file = default_max_per_file
         self.current_max_per_file = default_max_per_file
-        self.current_preview = None # Only stores the resized preview for GUI
+        self.current_preview: Image.Image | None = None
         self.is_processing = False
+        self.manual_queue: Queue[
+            tuple[Image.Image, list[Image.Image], dict[str, object]] | None
+        ] = Queue()
 
-        # Queue for manual re-processing results
-        self.manual_queue = Queue()
-
-        # Setup GUI
         self.root = tk.Tk()
         self.root.title("PDF Cropper")
         self.root.geometry("+100+100")
-
         self.label_img = tk.Label(self.root)
         self.label_img.pack()
-
         self.label_info = tk.Label(self.root, text="", font=("Arial", 12, "bold"))
         self.label_info.pack(pady=5)
+        self.root.bind("<Return>", self.on_next)
+        self.root.bind("n", lambda _event: self.on_shift(50))
+        self.root.bind("N", lambda _event: self.on_shift(100))
+        self.root.bind("t", lambda _event: self.on_shift(-50))
+        self.root.bind("1", lambda _event: self.on_set_max_pages(1))
 
-        # Bindings
+        Thread(target=self.prefetch_worker, daemon=True).start()
-        self.root.bind('<Return>', self.on_next)
-        self.root.bind('n', lambda e: self.on_shift(50))
-        self.root.bind('N', lambda e: self.on_shift(100))
-        self.root.bind('t', lambda e: self.on_shift(-50))
-        self.root.bind('1', lambda e: self.on_set_max_pages(1))
-
-        # Start background pre-fetcher
-        self.bg_thread = Thread(target=self.prefetch_worker, daemon=True)
-        self.bg_thread.start()
-
-        # Load first image
         self.load_current_image()
-
         self.root.lift()
         self.root.focus_force()
         self.root.mainloop()
 
-    def on_set_max_pages(self, count):
+    def on_set_max_pages(self, count: int) -> None:
         if self.is_processing:
             return
         self.current_max_per_file = count
         print(f"Setting max pages per file: {count}")
         self.trigger_processing(self.files[self.index], self.current_shift)
 
-    def prefetch_worker(self):
+    def prefetch_worker(self) -> None:
-        """Background thread to load the NEXT file's PDF pages into RAM."""
+        processed_index = -1
-        idx_to_process = -1
         while True:
             target = self.index + 1
-            if target < len(self.files) and target != idx_to_process:
+            if target < len(self.files) and target != processed_index:
-                fname = self.files[target]
+                get_pdf_pages(self.files[target])
-                get_pdf_pages(fname)
+                processed_index = target
-                idx_to_process = target
             time.sleep(0.05)
 
-    def load_current_image(self, use_prefetch=False):
+    def load_current_image(self) -> None:
         if self.index >= len(self.files):
             print("All files processed.")
             self.root.destroy()
             return
-
-        filename = self.files[self.index]
         self.is_processing = False
         self.current_shift = 0
+        self.trigger_processing(self.files[self.index], self.current_shift)
 
-        self.trigger_processing(filename, self.current_shift)
+    def trigger_processing(self, pdf_path: Path, shift: int) -> None:
-
-    def trigger_processing(self, filename, shift):
-        """Starts a thread to process image so GUI doesn't freeze."""
         self.is_processing = True
-        self.label_info.configure(text=f"Processing {filename} (Shift {shift})... Please wait.", fg="red")
+        self.label_info.configure(
+            text=f"Processing {pdf_path.name} (Shift {shift})... Please wait.",
+            fg="red",
+        )
 
-        def worker():
+        def worker() -> None:
-            res = process_single_pdf(filename, shift, self.current_max_per_file)
+            self.manual_queue.put(
-            self.manual_queue.put(res)
+                process_single_pdf(pdf_path, shift, self.current_max_per_file)
-
-        Thread(target=worker, daemon=True).start()
-        self.check_manual_queue(filename)
-
-    def check_manual_queue(self, filename):
-        """Polls the manual queue for result."""
-        try:
-            result = self.manual_queue.get_nowait()
-            if result:
-                self.handle_processing_result(result, filename)
-            else:
-                print(f"Failed to process {filename}, skipping.")
-                self.index += 1
-                self.load_current_image(use_prefetch=True)
-            self.is_processing = False
-        except Empty:
-            self.root.after(100, lambda: self.check_manual_queue(filename))
-
-    def handle_processing_result(self, result, filename):
-        """Unpacks result, saves files, and updates display."""
-        preview, splits, schema = result
-        self.current_preview = preview
-
-        Thread(target=save_results, args=(result, filename), daemon=True).start()
-
-        self.update_display(filename, schema)
-
-    def update_display(self, filename, schema=None):
-        if self.current_preview:
-            tk_image = ImageTk.PhotoImage(self.current_preview)
-            self.label_img.configure(image=tk_image)
-            self.label_img.image = tk_image
-
-            schema_info = ""
-            if schema:
-                cols = str(schema['columns_per_file'])
-                schema_info = f"\nFiles: {schema['number_of_files']} | Cols: {cols}"
-
-            self.label_info.configure(
-                text=f"[{self.index+1}/{len(self.files)}] {filename} | Shift: {self.current_shift}px"
-                     f"{schema_info}\n"
-                     f"Enter: Next | n: +50 | N: +100 | t: -50 | 1: use single column",
-                fg="black"
             )
 
-    def on_shift(self, amount):
+        Thread(target=worker, daemon=True).start()
+        self.check_manual_queue(pdf_path)
+
+    def check_manual_queue(self, pdf_path: Path) -> None:
+        try:
+            result = self.manual_queue.get_nowait()
+            if result is None:
+                print(f"Failed to process {pdf_path.name}, skipping.")
+                self.index += 1
+                self.load_current_image()
+            else:
+                self.handle_processing_result(result, pdf_path)
+            self.is_processing = False
+        except Empty:
+            self.root.after(100, lambda: self.check_manual_queue(pdf_path))
+
+    def handle_processing_result(
+        self,
+        result: tuple[Image.Image, list[Image.Image], dict[str, object]],
+        pdf_path: Path,
+    ) -> None:
+        self.current_preview = result[0]
+        save_results(result, pdf_path, self.output_dir)
+        self.update_display(pdf_path.name, result[2])
+
+    def update_display(self, filename: str, schema: dict[str, object]) -> None:
+        if self.current_preview is None:
+            return
+        tk_image = ImageTk.PhotoImage(self.current_preview)
+        self.label_img.configure(image=tk_image)
+        self.label_img.image = tk_image
+        self.label_info.configure(
+            text=(
+                f"[{self.index + 1}/{len(self.files)}] {filename} | "
+                f"Shift: {self.current_shift}px\nFiles: {schema['number_of_files']} | "
+                f"Cols: {schema['columns_per_file']}\n"
+                "Enter: Next | n: +50 | N: +100 | t: -50 | "
+                "1: use single column"
+            ),
+            fg="black",
+        )
+
+    def on_shift(self, amount: int) -> None:
         if self.is_processing:
             return
         self.current_shift += amount
         print(f"Applying shift: {self.current_shift}")
         self.trigger_processing(self.files[self.index], self.current_shift)
 
-    def on_next(self, event):
+    def on_next(self, _event: object) -> None:
         if self.is_processing:
             return
         self.index += 1
         self.current_shift = 0
         self.current_max_per_file = self.default_max_per_file
-        self.load_current_image(use_prefetch=True)
+        self.load_current_image()
 
 
-# --- Entry Point ---
+def _selected_files(
-if __name__ == "__main__":
+    workspace: EvaluationWorkspace,
+    target: Path,
+) -> list[Path]:
+    workspace.require_directories("Copies")
+    if target.is_file():
+        if target.suffix.casefold() != ".pdf":
+            raise CliError(f"Target is not a PDF: {target}", ExitCode.INVALID_ARGUMENTS)
+        return [target]
+    return sorted(
+        (
+            path
+            for path in workspace.copies_dir.glob("*.pdf")
+            if "nonc" not in path.name.casefold()
+        ),
+        key=lambda path: path.name.casefold(),
+    )
+
+
+def run(
+    workspace: EvaluationWorkspace,
+    target: Path,
+    *,
+    fullpage: bool = False,
+) -> ExitCode:
+    files = _selected_files(workspace, target)
     if not files:
         print("No PDF files found.")
-    else:
+        return ExitCode.SUCCESS
-        app = ImageReviewer(files, default_max_per_file=1 if fullpage_mode else 5)
+    workspace.cutleft_dir.mkdir(parents=True, exist_ok=True)
+    _get_pdf_pages_cached.cache_clear()
+    ImageReviewer(
+        files,
+        workspace.cutleft_dir,
+        default_max_per_file=1 if fullpage else 5,
+    )
+    return ExitCode.SUCCESS
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = target_parser("Interactively crop the label margin from PDF copies")
+    parser.add_argument(
+        "--fullpage",
+        action="store_true",
+        help="Use each complete page instead of cropping the label margin",
+    )
+    return parser
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+
+    def handle(args: argparse.Namespace) -> ExitCode:
+        workspace, target = workspace_from_target(args)
+        return run(workspace, target, fullpage=args.fullpage)
+
+    return execute(parser, argv, handle)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

enonce_info.py

@@ -1,16 +1,26 @@
-import sys
+from __future__ import annotations
-import os
+
+import argparse
 import glob
 import json
-import urllib.request
+import os
 import re
-import subprocess
+import urllib.request
-import tempfile
+from collections.abc import Sequence
-import shutil
+from uuid import uuid4
 
+from copienator import (
+    CliError,
+    EvaluationWorkspace,
+    ExitCode,
+    evaluation_parser,
+    execute,
+    workspace_from_args,
+)
 from platform_utils import WindowsLabelError, validate_windows_labels
 from utils import compile_to_pdf
 
+
 def fetch_and_save_sub_text(ex_id, indices, label, text_path):
     """Fetches text for a specific sub-question and saves it to Text/{label}.tex"""
     qinds = ",".join(map(str, indices))
@@ -29,6 +39,7 @@ def fetch_and_save_sub_text(ex_id, indices, label, text_path):
             compile_to_pdf(content, pdf_file)
     except Exception as e:
         print(f"Error fetching sub-text from {url}: {e}")
+        raise
 
 def fetch_and_save_sub_sol(ex_id, indices, label, sol_path):
     """Fetches text for a specific sub-question and saves it to Text/{label}.tex"""
@@ -48,6 +59,7 @@ def fetch_and_save_sub_sol(ex_id, indices, label, sol_path):
             compile_to_pdf(content, pdf_file)
     except Exception as e:
         print(f"Error fetching sub-text from {url}: {e}")
+        raise
 
 
 ROMANS_CAP = ["", "I", "II", "III", "IV", "V", "VI", "VII", "VIII", "IX", "X"]
@@ -127,18 +139,21 @@ def save_split_content(text, path, base_fname, problem):
             f.write(chunk)
 
 
-def process_directory(directory):
+def process_directory(workspace: EvaluationWorkspace) -> ExitCode:
+    directory = str(workspace.root)
     # Find the first .tex file in the directory
     tex_files = glob.glob(os.path.join(directory, "*.tex"))
     if not tex_files:
         print(f"No .tex file found in {directory}. Looking in /Staging/Interro/")
-        int_name = directory[:-1] if directory.endswith("/") else directory
+        int_name = directory.removesuffix("/")
         tex_path = os.path.join(os.path.expanduser("~"), "Prépa/Staging/Interro", f"{int_name}.tex")
         if os.path.exists(tex_path):
             tex_file = tex_path
         else:
-            print("Not found in ", tex_path)
+            raise CliError(
-            return
+                f"No .tex input found in {workspace.root}",
+                ExitCode.INVALID_WORKSPACE,
+            )
     else:
         tex_file = tex_files[0]
 
@@ -153,8 +168,10 @@ def process_directory(directory):
     for p in paths.values():
         os.makedirs(p, exist_ok=True)
 
-    labels_file = os.path.join(directory, "labels")
+    labels_file = workspace.labels_file
+    labels_staging = labels_file.with_name(f".{labels_file.name}.{uuid4().hex}.tmp")
     current_ex_num = 1
+    had_errors = False
 
     # Read entirely to allow chunking
     with open(tex_file, 'r', encoding='utf-8') as f_in:
@@ -162,8 +179,11 @@ def process_directory(directory):
 
     # Split by the specific SHEETINFO tag
     blocks = content.split("%%SHEETINFO :")
+    if len(blocks) == 1:
+        print(f"No SHEETINFO blocks found in {tex_file}")
+        return ExitCode.PARTIAL
 
-    with open(labels_file, 'w', encoding='utf-8') as f_labels:
+    with open(labels_staging, 'w', encoding='utf-8') as f_labels:
         # Skip blocks[0] (content before first SHEETINFO)
         for block in blocks[1:]:
             parts_line = block.split("\n", 1)
@@ -177,6 +197,7 @@ def process_directory(directory):
 
             try:
                 data = json.loads(json_str)
+                block_labels = []
                 # Construct 'ids' parameter
                 ex_id = str(data['id'])
                 selection = data.get('select')
@@ -193,7 +214,7 @@ def process_directory(directory):
                 if not indexes:
                     label = f"Ex {current_ex_num}"
                     validate_windows_labels([label])
-                    f_labels.write(f"{label}\n")
+                    block_labels.append(label)
                     fetch_and_save_sub_text(ids, [], label, paths['Text2'])
                     fetch_and_save_sub_sol(ids, [], label, paths['Sol2'])
                 else:
@@ -201,7 +222,7 @@ def process_directory(directory):
                         suffix = format_indices(item['indices'], problem)
                         label = f"Ex {current_ex_num}" + (f" : {suffix}" if suffix else "")
                         validate_windows_labels([label])
-                        f_labels.write(f"{label}\n")
+                        block_labels.append(label)
                         fetch_and_save_sub_text(ids, item['indices'], label, paths['Text2'])
                         fetch_and_save_sub_sol(ids, item['indices'], label, paths['Sol2'])
 
@@ -244,18 +265,31 @@ def process_directory(directory):
                 save_split_content(s_text, paths['Sol'], base_filename, problem)
                 save_split_content(p_text, paths['Persp'], base_filename, problem)
 
+                for label in block_labels:
+                    f_labels.write(f"{label}\n")
                 current_ex_num += 1
 
             except WindowsLabelError:
+                labels_staging.unlink(missing_ok=True)
                 raise
             except json.JSONDecodeError:
                 print(f"Error decoding JSON in block: {json_str}")
-            except Exception as e:
+                had_errors = True
+            except Exception as e:  # noqa: BLE001 - one malformed exercise is partial
                 print(f"Error processing block {ex_id if 'ex_id' in locals() else 'unknown'}: {e}")
+                had_errors = True
+
+    labels_staging.replace(labels_file)
+    return ExitCode.PARTIAL if had_errors else ExitCode.SUCCESS
+
+
+def build_parser() -> argparse.ArgumentParser:
+    return evaluation_parser("Generate statement metadata from SHEETINFO blocks")
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+    return execute(parser, argv, lambda args: process_directory(workspace_from_args(args)))
 
 if __name__ == "__main__":
-    if len(sys.argv) < 2:
+    raise SystemExit(main())
-        print("Usage: python script.py <Dir>")
-        sys.exit(1)
-
-    process_directory(sys.argv[1])

fetch_batched_results.py

@@ -1,63 +1,83 @@
-import os
+from __future__ import annotations
-import sys
+
 import argparse
-from pathlib import Path
+from collections.abc import Sequence
+
 from google import genai
 
-def main():
+import config
-    parser = argparse.ArgumentParser(description="Download and combine completed batch jobs for a directory.")
+from copienator import (
-    parser.add_argument("root_dir", type=str, help="Directory containing the original batches")
+    CliError,
-    args = parser.parse_args()
+    EvaluationWorkspace,
+    ExitCode,
+    atomic_write_bytes,
+    evaluation_parser,
+    execute,
+    read_json,
+    workspace_from_args,
+)
 
-    target_dir = Path(args.root_dir)
-    dir_name = target_dir.name
-    output_path = target_dir / "batched_correction_result.jsonl"
 
-    if "GEMINI_API_KEY" not in os.environ:
+def run(workspace: EvaluationWorkspace, *, client=None) -> ExitCode:
-        sys.exit("Error: GEMINI_API_KEY environment variable not set.")
+    if client is None:
+        if not config.API_KEY:
+            raise CliError("GEMINI_API_KEY is not configured")
+        client = genai.Client(api_key=config.API_KEY)
+    matching = []
+    if workspace.batch_jobs_file.is_file():
+        manifest = read_json(workspace.batch_jobs_file)
+        jobs = manifest.get("jobs") if isinstance(manifest, dict) else None
+        if not isinstance(jobs, dict):
+            raise CliError(f"Invalid batch manifest: {workspace.batch_jobs_file}")
+        matching = [
+            client.batches.get(name=entry["name"])
+            for entry in jobs.values()
+            if isinstance(entry, dict) and isinstance(entry.get("name"), str)
+        ]
+    else:
+        matching = [
+            job
+            for job in client.batches.list()
+            if workspace.name in str(getattr(job, "display_name", ""))
+        ]
+    if not matching:
+        raise CliError(
+            f"No batch jobs found for evaluation {workspace.name!r}"
+        )
+    for job in matching:
+        state = job.state.name if hasattr(job.state, "name") else job.state
+        print(f"{job.display_name}: {state}")
+        if state != "JOB_STATE_SUCCEEDED":
+            print("Not all matching jobs have succeeded yet.")
+            return ExitCode.PARTIAL
 
-    client = genai.Client()
+    chunks = []
+    incomplete = False
+    for job in matching:
+        destination = getattr(job, "dest", None)
+        file_name = getattr(destination, "file_name", None)
+        if not file_name:
+            print(f"Warning: {job.display_name} has no output file.")
+            incomplete = True
+            continue
+        payload = client.files.download(file=file_name)
+        chunks.append(payload.rstrip(b"\n"))
+    if not chunks:
+        return ExitCode.PARTIAL
+    output_path = workspace.batched_correction_result_file
+    atomic_write_bytes(output_path, b"\n".join(chunks) + b"\n")
+    print(f"Saved combined batch results to {output_path}")
+    return ExitCode.PARTIAL if incomplete else ExitCode.SUCCESS
 
-    print(f"Fetching jobs matching '{dir_name}'...")
-    all_jobs = client.batches.list()
-    matching_jobs = []
 
-    # 1. Find jobs associated with this directory
+def build_parser() -> argparse.ArgumentParser:
-    for job in all_jobs:
+    return evaluation_parser("Download and combine correction batch results")
-        if hasattr(job, 'display_name') and job.display_name and dir_name in job.display_name:
-            matching_jobs.append(job)
 
-    if not matching_jobs:
-        sys.exit(f"No batch jobs found containing '{dir_name}' in their display name.")
 
-    # 2. Check that all matching jobs are complete
+def main(argv: Sequence[str] | None = None) -> int:
-    for job in matching_jobs:
+    parser = build_parser()
-        state = job.state.name if hasattr(job.state, 'name') else job.state
+    return execute(parser, argv, lambda args: run(workspace_from_args(args)))
-        print(f"Found Job: {job.display_name} | State: {state}")
-        if state != 'JOB_STATE_SUCCEEDED':
-            sys.exit(f"Error: Job '{job.display_name}' has not succeeded yet. Try again later.")
 
-    # 3. Download and concatenate
-    print("\nAll jobs succeeded. Downloading results...")
-    combined_data = b""
-
-    for job in matching_jobs:
-        if hasattr(job, 'dest') and job.dest and hasattr(job.dest, 'file_name') and job.dest.file_name:
-            print(f"Downloading output for {job.display_name}...")
-            file_content_bytes = client.files.download(file=job.dest.file_name)
-
-            combined_data += file_content_bytes
-            # Ensure proper line separation between files in JSONL
-            if combined_data and not combined_data.endswith(b'\n'):
-                combined_data += b'\n'
-        else:
-            print(f"Warning: Job {job.display_name} succeeded but has no output file.")
-
-    # 4. Save to destination
-    with open(output_path, "wb") as f:
-        f.write(combined_data)
-
-    print(f"\nSuccess! All results concatenated and saved to:\n{output_path}")
 
 if __name__ == "__main__":
-    main()
+    raise SystemExit(main())

gemini_for_enonce.py

@@ -1,18 +1,30 @@
-import re
+from __future__ import annotations
-import os
+
-import utils
-import subprocess
-import sys
 import argparse
+import re
+from collections.abc import Sequence
+from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path
-from pydantic import BaseModel, Field
+
-from typing import List, Union
 from google import genai
 from google.genai import types
-from concurrent.futures import ThreadPoolExecutor
+from pydantic import BaseModel, Field
+
+import config
+import utils
+from copienator import (
+    CliError,
+    EvaluationWorkspace,
+    ExitCode,
+    atomic_write_text,
+    evaluation_parser,
+    execute,
+    workspace_from_args,
+)
 from platform_utils import validate_windows_labels
 from utils import compile_to_pdf
 
+
 def get_lcp(s1: str, s2: str) -> str:
     i = 0
     while i < len(s1) and i < len(s2) and s1[i] == s2[i]:
@@ -25,8 +37,6 @@ def get_lcp(s1: str, s2: str) -> str:
     return lcp
 
 
-import config
-
 MODEL_ID = config.MODEL_LITE_ID
 api_key = config.API_KEY
 
@@ -36,7 +46,7 @@ class QuestionOnlyItem(BaseModel):
     question_content: str = Field(description="The source text of the question, strictly extracted from the enonce file, EXCLUDING the label itself.")
 
 class ExamQuestions(BaseModel):
-    questions: List[QuestionOnlyItem]
+    questions: list[QuestionOnlyItem]
 
 # --- Modèles pour la Requête 2 ---
 class SolutionOnlyItem(BaseModel):
@@ -44,7 +54,7 @@ class SolutionOnlyItem(BaseModel):
     solution_content: str = Field(description="The source text of the solution, strictly extracted from the correction file.")
 
 class ExamSolutions(BaseModel):
-    solutions: List[SolutionOnlyItem]
+    solutions: list[SolutionOnlyItem]
 
 # --- Modèles pour la Requête 3 ---
 class ExtractedContext(BaseModel):
@@ -53,7 +63,7 @@ class ExtractedContext(BaseModel):
     context_content: str = Field(description="The source text of the definitions, notations, or hypotheses, extracted from the enonce.")
 
 class ExamContext(BaseModel):
-    contexts: List[ExtractedContext]
+    contexts: list[ExtractedContext]
 
 # --- Modèles pour la Requête 4 (Barèmes) ---
 class RubricItem(BaseModel):
@@ -61,7 +71,7 @@ class RubricItem(BaseModel):
     rubric_content: str = Field(description="Le barème détaillé en français.")
 
 class GroupRubrics(BaseModel):
-    rubrics: List[RubricItem]
+    rubrics: list[RubricItem]
 
 PROMPT_4 = """Je te fournis les questions, le contexte éventuel, et les corrections pour un groupe de questions d'un examen.
 Ta tâche :
@@ -87,10 +97,10 @@ class ContextItem(BaseModel):
     content: str  # Juste une string encapsulée pour le différencier facilement
 
 class ExamExtraction(BaseModel):
-    items: List[Union[QuestionItem, ContextItem]] # Liste mixte
+    items: list[QuestionItem | ContextItem] # Liste mixte
 
 class GroupedExamExtraction(BaseModel):
-    groups: List[List[Union[QuestionItem, ContextItem]]]
+    groups: list[list[QuestionItem | ContextItem]]
 
 PROMPT_1 = """I am providing:
 1. A PDF of an exam (`enonce.pdf`)
@@ -138,15 +148,20 @@ the `Let N, M be two commutating matrices` part is not a question itself, and is
  Return the result as a JSON list.
 """
 
-def find_file(folder: Path, base_name: str) -> Path:
+def find_file(folder: Path, base_name: str) -> Path | None:
     for ext in [".org", ".tex"]:
         path = folder / f"{base_name}{ext}"
         if path.is_file():
             return path
     return None
 
-def process_exam(folder_path: str, restart: bool = False):
+def process_exam(
-    folder = Path(folder_path)
+    workspace: EvaluationWorkspace,
+    restart: bool = False,
+    *,
+    api_client=None,
+) -> ExitCode:
+    folder = workspace.root
 
     cache_dir = folder / "Cache"
     tmp_dir = folder / "Tmp"
@@ -168,15 +183,21 @@ def process_exam(folder_path: str, restart: bool = False):
     if not correction_path: missing.append("correction.org or correction.tex")
 
     if missing:
-        print(f"Error: Missing files in {folder}: {', '.join(missing)}")
+        raise CliError(
-        sys.exit(1)
+            f"Missing files in {folder}: {', '.join(missing)}",
+            ExitCode.INVALID_WORKSPACE,
+        )
 
     print("Reading files...")
     pdf_bytes = pdf_path.read_bytes()
     enonce_text = enonce_path.read_text(encoding="utf-8")
     correction_text = correction_path.read_text(encoding="utf-8")
 
-    client = genai.Client(api_key=api_key)
+    if api_client is None:
+        if not api_key:
+            raise CliError("GEMINI_API_KEY is not configured")
+        api_client = genai.Client(api_key=api_key)
+    client = api_client
 
     # ==========================================
     # REQUÊTE 1 : Extraction des Énoncés
@@ -210,7 +231,7 @@ def process_exam(folder_path: str, restart: bool = False):
         )
         response_q_text = response_q.text
         print("Saving questions to cache...")
-        cache_q_file.write_text(response_q_text, encoding="utf-8")
+        atomic_write_text(cache_q_file, response_q_text)
 
     questions_data = ExamQuestions.model_validate_json(response_q_text)
 
@@ -248,7 +269,7 @@ def process_exam(folder_path: str, restart: bool = False):
         )
         response_s_text = response_s.text
         print("Saving solutions to cache...")
-        cache_s_file.write_text(response_s_text, encoding="utf-8")
+        atomic_write_text(cache_s_file, response_s_text)
 
     solutions_data = ExamSolutions.model_validate_json(response_s_text)
 
@@ -284,7 +305,7 @@ def process_exam(folder_path: str, restart: bool = False):
         )
         response_c_text = response_c.text
         print("Saving context to cache...")
-        cache_c_file.write_text(response_c_text, encoding="utf-8")
+        atomic_write_text(cache_c_file, response_c_text)
 
     context_data = ExamContext.model_validate_json(response_c_text)
 
@@ -388,7 +409,6 @@ def process_exam(folder_path: str, restart: bool = False):
     for g_indices in q_group_indices:
         group_items = []
         first_q_idx = g_indices[0]
-        last_q_idx = g_indices[-1]
 
         for q_idx in g_indices:
             q_item = questions_only[q_idx]
@@ -497,14 +517,13 @@ def process_exam(folder_path: str, restart: bool = False):
 
         # 2. Actual Parsing
         grouped_items = []
-        current_raw_group = []  # Stores (is_context, label_or_flag, content)
         all_new_q_labels = []
 
         # Pass 1: Read all edited lines and collect question labels in sequence
         for line in edited_lines:
             if line == "---" or " ### " not in line:
                 continue
-            lbl, content_raw = line.split(" ### ", 1)
+            lbl, _content_raw = line.split(" ### ", 1)
             lbl = lbl.strip()
             if lbl != "CONTEXT":
                 all_new_q_labels.append(lbl)
@@ -606,10 +625,6 @@ def process_exam(folder_path: str, restart: bool = False):
     validate_windows_labels(labels_list)
 
     # Save labels and proceed
-    with open(folder / "labels", 'w', encoding='utf-8') as f_labels:
-        for label in labels_list:
-            f_labels.write(f"{label}\n")
-
     grouped_extraction = GroupedExamExtraction(groups=grouped_items)
 
     # 2. Setup output directories
@@ -628,8 +643,7 @@ def process_exam(folder_path: str, restart: bool = False):
         ).strip().lower()
 
         if answer not in ("y", "yes"):
-            print("Aborted.")
+            raise CliError("Output replacement aborted", ExitCode.INVALID_ARGUMENTS)
-            sys.exit(1)
         # Empty each directory
         for d in dirs:
             if d.exists():
@@ -645,6 +659,7 @@ def process_exam(folder_path: str, restart: bool = False):
 
 
     print("Writing grouped question and solution files...")
+    processing_errors = []
 
     for group in grouped_extraction.groups:
         q_items = [item for item in group if isinstance(item, QuestionItem)]
@@ -690,8 +705,9 @@ def process_exam(folder_path: str, restart: bool = False):
             )
             rubrics_data = GroupRubrics.model_validate_json(response_r.text)
             rubrics_map = {r.label: r.rubric_content for r in rubrics_data.rubrics}
-        except Exception as e:
+        except Exception as e:  # noqa: BLE001 - remote API boundary
             print(f"Error generating rubric for group {labels[0]}: {e}")
+            processing_errors.append(str(e))
             rubrics_map = {}
 
         # 1. Compute the common prefix for the group
@@ -737,7 +753,7 @@ def process_exam(folder_path: str, restart: bool = False):
             elif isinstance(item, ContextItem):
                 raw_ctx = item.content.strip()
                 tabulated_ctx = "\t" + re.sub(r'\n\s*', '\n\t', raw_ctx)
-                text_content_lines.append(f"CONTEXT :")
+                text_content_lines.append("CONTEXT :")
                 text_content_lines.append(tabulated_ctx)
 
                 # --- Save context to Text2 (Concatenating if exists) ---
@@ -762,29 +778,53 @@ def process_exam(folder_path: str, restart: bool = False):
     # ==========================================
     all_tex_files = list(text2_dir.glob("*.tex")) + list(sol2_dir.glob("*.tex"))
 
-    def compile_worker(tex_file: Path):
+    def compile_worker(tex_file: Path) -> str | None:
         """Helper to read content and call the utility function."""
         try:
             content = tex_file.read_text(encoding="utf-8")
             pdf_path = tex_file.with_suffix(".pdf")
             compile_to_pdf(content, pdf_path)
-        except Exception as e:
+        except Exception as e:  # noqa: BLE001 - compiler worker boundary
-            print(f"Error compiling {tex_file.name}: {e}")
+            return f"Error compiling {tex_file.name}: {e}"
+        return None
 
     print(f"Compiling {len(all_tex_files)} files to PDF using 4 threads...")
     with ThreadPoolExecutor(max_workers=4) as executor:
-        executor.map(compile_worker, all_tex_files)
+        compile_errors = [
+            error for error in executor.map(compile_worker, all_tex_files) if error
+        ]
+    for error in compile_errors:
+        print(error)
+    processing_errors.extend(compile_errors)
+    atomic_write_text(
+        workspace.labels_file,
+        "".join(f"{label}\n" for label in labels_list),
+    )
+
+    return ExitCode.PARTIAL if processing_errors else ExitCode.SUCCESS
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = evaluation_parser("Extract exam and solution code via Gemini")
+    parser.add_argument(
+        "--restart",
+        action="store_true",
+        help="Ignore cached Gemini extraction responses",
+    )
+    return parser
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+    return execute(
+        parser,
+        argv,
+        lambda args: process_exam(
+            workspace_from_args(args),
+            restart=args.restart,
+        ),
+    )
 
 
 if __name__ == "__main__":
-    if not api_key:
+    raise SystemExit(main())
-        print("Error: GEMINI_API_KEY environment variable is not set.")
-        sys.exit(1)
-
-    parser = argparse.ArgumentParser(description="Extract exam and solution code via Gemini.")
-    parser.add_argument("folder", help="Directory containing the exam files")
-    parser.add_argument("--restart", action="store_true", help="Ignore cache files and re-run extraction requests.")
-
-
-    args = parser.parse_args()
-    process_exam(args.folder, restart=args.restart)

gemini_for_labels.py

@@ -1,18 +1,30 @@
-from google import genai
+from __future__ import annotations
-from google.genai import types
+
-import base64
-from pathlib import Path
-from pydantic import BaseModel, Field
-from typing import List, Dict
-import sys
-import os
-import time
-import json
 import argparse
 import re
+import time
+import typing
 from collections import defaultdict
+from collections.abc import Callable, Sequence
 from concurrent.futures import ThreadPoolExecutor
+from pathlib import Path
+
+from google import genai
+from google.genai import types
+from pydantic import BaseModel, Field
+
 import config
+from copienator import (
+    CliError,
+    EvaluationWorkspace,
+    ExitCode,
+    atomic_write_json,
+    execute,
+    read_json,
+    target_parser,
+    workspace_from_target,
+)
+from utils import natural_key, read_all_labels
 
 MODEL_ID = config.MODEL_FOR_LABEL_ID
 api_key = config.API_KEY
@@ -125,12 +137,14 @@ Since this copy isn't the first part of a sequence, simply set the
 name to `\"Continued\"`."""
 
 class BoxItem(BaseModel):
-    box_2d: List[int] = Field(description="Bounding box coordinates (e.g., [ymin, xmin, ymax, xmax])")
+    box_2d: list[int] = Field(description="Bounding box coordinates (e.g., [ymin, xmin, ymax, xmax])")
     label: str = Field(description="The label associated with the specific box")
 
 class AnnotationData(BaseModel):
     name: str = Field(description="The name identifier")
-    list: List[BoxItem] = Field(description="List of bounding box items")
+    list: typing.List[BoxItem] = Field(  # noqa: UP006 - field name shadows list
+        description="List of bounding box items"
+    )
 
 
 def generate_request(file, labels, names, context_labels, wrong_labels):
@@ -176,169 +190,261 @@ def generate_request(file, labels, names, context_labels, wrong_labels):
     )
     return (contents, generate_content_config)
 
-# Argument Parsing
-parser = argparse.ArgumentParser(description="Process a directory or specific files using Gemini.")
-parser.add_argument("input_paths", nargs='+', help="The input directory or specific files")
-parser.add_argument("--overwrite", action="store_true", help="Regenerate output even if it exists")
-
-args = parser.parse_args()
-
-# input_arg = Path(args.input_path)
-image_files = []
-
-from utils import natural_key, read_all_labels
-
-for path_str in args.input_paths:
-    input_arg = Path(path_str)
-    target_files = []
-
-    # 1. Determine which files to process
-    if input_arg.is_file():
-        INPUT_DIR = input_arg.parent.parent
-        target_files = [input_arg]
-    elif input_arg.is_dir():
-        INPUT_DIR = input_arg
-        COPIES_DIR = INPUT_DIR / "Copies"
-        target_files = list(COPIES_DIR.glob("Copie*.pdf"))
-        if not target_files:
-            print(f"Warning: No Copie*.pdf files found in {input_arg}")
-    else:
-        print(f"Error: {input_arg} is not a valid file or directory.")
-        continue
-
-    # 2. Run the logic for all collected files
-    for target_file in target_files:
-        # INPUT_DIR = target_file.parent
-        CUTLEFT_DIR = INPUT_DIR / 'Cutleft'
-
-        # Matches stem_01.jpg, stem_02.jpg, etc.
-        found_files = sorted(
-            CUTLEFT_DIR.glob(f"{target_file.stem}_*.jpg"),
-            key=natural_key
-        )
-
-        if found_files:
-            image_files.extend(found_files)
-        else:
-            print(f"Warning: No variants found for {target_file.stem} in {CUTLEFT_DIR}")
-
-all_labels = read_all_labels(INPUT_DIR)
-labels_txt = "\n".join(all_labels) + "\n"
-valid_labels_set = set(all_labels)
-names_path = (INPUT_DIR / "names")
-if not os.path.exists(names_path):
-    names_path = Path("names")
-names_txt = names_path.read_text()
-
-valid_names_set = set(line.strip() for line in names_txt.splitlines() if line.strip())
-valid_names_set.add("Unknown")
-valid_names_set.add("Continued")
-
-client = genai.Client(api_key=api_key)
-
-# Group files by Copy ID (e.g. Copie01_01.jpg -> Copie01)
-# regex: match everything before the last underscore if it ends in digits
-file_groups = defaultdict(list)
-for img in image_files:
-    stem = img.stem
-    # match CopieXX_YY -> Group CopieXX
-    match = re.match(r"(.+)_(\d+)$", stem)
-    if match:
-        group_key = match.group(1)
-        file_groups[group_key].append(img)
-    else:
-        # Fallback for files without underscore numbering
-        file_groups[stem].append(img)
-
-# Sort files within each group to ensure sequential processing
-for key in file_groups:
-    file_groups[key].sort(key=lambda x: x.name)
-
 TARGET_INTERVAL = 3.5
+Sleep = Callable[[float], None]
 
-def process_copy_group(group_key, files):
-    """Processes a list of files belonging to one copy sequentially to maintain context."""
 
-    # Context accumulator for this specific copy
+def selected_images(
-    accumulated_labels = []
+    workspace: EvaluationWorkspace,
+    targets: list[Path],
+) -> tuple[list[Path], list[str]]:
+    """Resolve evaluation, copy-PDF, or Cutleft-image targets."""
+    workspace.require_directories("Copies", "Cutleft")
+    images: list[Path] = []
+    warnings: list[str] = []
+    for target in targets:
+        if target.is_dir():
+            copy_pdfs = sorted(
+                workspace.copies_dir.glob("Copie*.pdf"), key=natural_key
+            )
+            if not copy_pdfs:
+                warnings.append(f"No Copie*.pdf files found in {workspace.copies_dir}")
+            stems = [path.stem for path in copy_pdfs]
+        elif target.suffix.casefold() in {".jpg", ".jpeg"}:
+            if target.parent != workspace.cutleft_dir:
+                raise CliError(
+                    f"Image target is not in {workspace.cutleft_dir}: {target}",
+                    ExitCode.INVALID_ARGUMENTS,
+                )
+            images.append(target)
+            continue
+        elif target.suffix.casefold() == ".pdf":
+            stems = [target.stem]
+        else:
+            raise CliError(
+                f"Unsupported target for label detection: {target}",
+                ExitCode.INVALID_ARGUMENTS,
+            )
+        for stem in stems:
+            found = sorted(
+                workspace.cutleft_dir.glob(f"{stem}_*.jpg"), key=natural_key
+            )
+            if found:
+                images.extend(found)
+            else:
+                warnings.append(
+                    f"No Cutleft image variants found for {stem} in "
+                    f"{workspace.cutleft_dir}"
+                )
+    return list(dict.fromkeys(images)), warnings
 
+
+def group_images(image_files: list[Path]) -> dict[str, list[Path]]:
+    groups: defaultdict[str, list[Path]] = defaultdict(list)
+    for image in image_files:
+        match = re.match(r"(.+)_(\d+)$", image.stem)
+        groups[match.group(1) if match else image.stem].append(image)
+    for files in groups.values():
+        files.sort(key=natural_key)
+    return dict(groups)
+
+
+def _existing_context(output_json: Path) -> list[str]:
+    try:
+        loaded = read_json(output_json)
+        if not isinstance(loaded, dict):
+            return []
+        return [
+            str(item["label"])
+            for item in loaded.get("list", [])
+            if isinstance(item, dict) and "label" in item
+        ]
+    except (OSError, TypeError, ValueError):
+        return []
+
+
+def process_copy_group(
+    workspace: EvaluationWorkspace,
+    group_key: str,
+    files: list[Path],
+    *,
+    client,
+    labels_text: str,
+    names_text: str,
+    valid_labels: set[str],
+    valid_names: set[str],
+    overwrite: bool,
+    sleep: Sleep = time.sleep,
+    target_interval: float = TARGET_INTERVAL,
+) -> int:
+    """Process one student's image parts sequentially to preserve context."""
+    accumulated_labels: list[str] = []
+    generated = 0
     for image_file in files:
-        start_time = time.time()
+        started = time.monotonic()
-        base_name = image_file.stem
+        output_json = workspace.copies_dir / f"{image_file.stem}.json"
-        output_json = INPUT_DIR / "Copies" / f"{base_name}.json"
+        if output_json.exists() and not overwrite:
-
-        # Check existing
-        if output_json.exists() and not args.overwrite:
             print(f"[{group_key}] Skipping {image_file.name}, output exists.")
-            # If skipping, we should try to load existing labels to keep context for next parts
+            accumulated_labels.extend(_existing_context(output_json))
-            try:
-                with open(output_json, 'r') as f:
-                    data = json.load(f)
-                    for item in data.get('list', []):
-                        accumulated_labels.append(item['label'])
-            except:
-                pass # If read fails, next part has no context
             continue
 
-        print(f"[{group_key}] Processing {image_file.name} with {len(accumulated_labels)} accumulated labels...")
+        print(
-
+            f"[{group_key}] Processing {image_file.name} with "
-        attempt = -1
+            f"{len(accumulated_labels)} accumulated labels..."
-        wrong_labels = []
+        )
+        attempt = 0
+        wrong_labels: list[str] = []
         while True:
-            attempt += 1
             if attempt > 0:
-                time.sleep(10 * attempt)
+                sleep(10 * attempt)
             try:
-                contents, config = generate_request(image_file, labels_txt, names_txt, accumulated_labels,
+                contents, request_config = generate_request(
-                                                    wrong_labels)
+                    image_file,
-
+                    labels_text,
+                    names_text,
+                    accumulated_labels,
+                    wrong_labels,
+                )
                 response = client.models.generate_content(
                     model=MODEL_ID,
                     contents=contents,
-                    config=config
+                    config=request_config,
                 )
-
+                annotation = AnnotationData.model_validate_json(response.text)
-                annota = AnnotationData.model_validate_json(response.text)
+                unknown = [
-                unknown = [item.label for item in annota.list if item.label not in valid_labels_set]
+                    item.label
-                name = annota.name
+                    for item in annotation.list
+                    if item.label not in valid_labels
+                ]
                 if unknown:
-                    print(f"Error: {image_file.name} contained unknown labels: {unknown}")
+                    print(
+                        f"Error: {image_file.name} contained unknown labels: "
+                        f"{unknown}"
+                    )
                     wrong_labels.extend(unknown)
-                    print("Retrying request...")
+                    attempt += 1
-                    continue  # Retry immediately
+                    continue
-
+                if annotation.name not in valid_names:
-                if name not in valid_names_set:
+                    print(
-                    print(f"Error: {image_file.name} returned unknown name : {name}")
+                        f"Error: {image_file.name} returned unknown name: "
+                        f"{annotation.name}"
+                    )
                     if attempt == 0:
-                        print("Retrying request...")
+                        attempt += 1
-                        continue  # Retry immediately
+                        continue
-                    else:
+                    annotation.name = "Unknown"
-                        name = "Unknown"
-                annota.name = name
-                # Save result
-                with open(output_json, "w", encoding="utf-8") as f:
-                    json.dump(annota.model_dump(), f, indent=2)
 
-                # Update context for the next part in this group
+                atomic_write_json(output_json, annotation.model_dump())
-                for box in annota.list:
+                accumulated_labels.extend(box.label for box in annotation.list)
-                    accumulated_labels.append(box.label)
+                generated += 1
-                break # exit retry loop
+                break
-            except Exception as e:
+            except KeyboardInterrupt:
-                print(f"Error processing {image_file.name}: {e}\n\tIt will be retried.")
+                raise
+            except Exception as exc:  # noqa: BLE001 - remote API retry boundary
+                print(
+                    f"Error processing {image_file.name}: {exc}\n"
+                    "\tIt will be retried."
+                )
+                attempt += 1
+        sleep(max(0.0, target_interval - (time.monotonic() - started)))
+    return generated
 
-        # Rate Limiting
-        elapsed = time.time() - start_time
-        time.sleep(max(0, TARGET_INTERVAL - elapsed))
 
-# Run ThreadPool on GROUPS (Copies), not individual files
+def run(
-# Each thread handles one student's full exam copy sequentially
+    workspace: EvaluationWorkspace,
-with ThreadPoolExecutor(max_workers=12) as executor:
+    targets: list[Path],
-    # Convert dict items to arguments for map
+    *,
-    # executor.map expects a function and an iterable.
+    overwrite: bool = False,
-    # We use a lambda or separate function to unpack the tuple if needed,
+    client=None,
-    # but here we'll just submit futures.
+    sleep: Sleep = time.sleep,
-    futures = [executor.submit(process_copy_group, k, v) for k, v in file_groups.items()]
+    max_workers: int = 12,
+) -> ExitCode:
+    workspace.require_files("labels")
+    images, warnings = selected_images(workspace, targets)
+    for warning in warnings:
+        print(f"Warning: {warning}")
+    if not images:
+        return ExitCode.PARTIAL
 
-    # Wait for all to complete
+    all_labels = read_all_labels(workspace.root)
-    for future in futures:
+    labels_text = "\n".join(all_labels) + "\n"
-        future.result()
+    names_path = workspace.names_file()
+    if not names_path.is_file():
+        raise CliError(f"Names file not found: {names_path}", ExitCode.INVALID_WORKSPACE)
+    names_text = names_path.read_text(encoding="utf-8")
+    valid_names = {
+        line.strip() for line in names_text.splitlines() if line.strip()
+    } | {"Unknown", "Continued"}
+    if client is None:
+        client = genai.Client(api_key=api_key)
+
+    groups = group_images(images)
+    with ThreadPoolExecutor(max_workers=max_workers) as executor:
+        futures = [
+            executor.submit(
+                process_copy_group,
+                workspace,
+                group_key,
+                files,
+                client=client,
+                labels_text=labels_text,
+                names_text=names_text,
+                valid_labels=set(all_labels),
+                valid_names=valid_names,
+                overwrite=overwrite,
+                sleep=sleep,
+            )
+            for group_key, files in groups.items()
+        ]
+        for future in futures:
+            future.result()
+    return ExitCode.PARTIAL if warnings else ExitCode.SUCCESS
+
+
+def build_parser() -> argparse.ArgumentParser:
+    parser = target_parser("Detect handwritten question labels with Gemini")
+    parser.add_argument(
+        "additional_targets",
+        nargs="*",
+        type=Path,
+        help="Additional copy PDFs or Cutleft images from the same evaluation",
+    )
+    parser.add_argument(
+        "--overwrite",
+        action="store_true",
+        help="Regenerate JSON outputs that already exist",
+    )
+    return parser
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+
+    def handle(args: argparse.Namespace) -> ExitCode:
+        workspace, target = workspace_from_target(
+            args, repository=Path(__file__).resolve().parent
+        )
+        targets = [target]
+        for additional in args.additional_targets:
+            resolved = additional.expanduser().resolve()
+            if not resolved.exists():
+                raise CliError(
+                    f"Target does not exist: {resolved}",
+                    ExitCode.INVALID_WORKSPACE,
+                )
+            additional_workspace = EvaluationWorkspace.discover(
+                resolved, repository=workspace.repository
+            )
+            if additional_workspace.root != workspace.root:
+                raise CliError(
+                    "All targets must belong to the same evaluation",
+                    ExitCode.INVALID_ARGUMENTS,
+                )
+            targets.append(resolved)
+        return run(workspace, targets, overwrite=args.overwrite)
+
+    return execute(parser, argv, handle)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

page_splitter.py

@@ -1,9 +1,15 @@
+from __future__ import annotations
+
+import argparse
 import glob
 import os
 import re
 import shutil
-import sys
+import tempfile
 import tkinter as tk
+import uuid
+from collections.abc import Sequence
+from pathlib import Path
 from tkinter import messagebox
 
 import fitz  # PyMuPDF
@@ -11,16 +17,78 @@ from PIL import Image, ImageDraw, ImageTk
 from pypdf import PdfReader, PdfWriter
 
 from config import PAGE_SPLITTER_KB
+from copienator import (
+    CliError,
+    EvaluationWorkspace,
+    ExitCode,
+    execute,
+    target_parser,
+    workspace_from_target,
+)
 from platform_utils import launch_pdf_arranger
 
-
 # --- Constants ---
 # Conversion factor: 1 cm to points (1 inch = 2.54 cm, 72 points = 1 inch)
 CM_TO_POINTS = (1 / 2.54) * 72
 
-def list_pdf_files(directory):
+def list_pdf_files(directory: str | Path) -> list[Path]:
-    l = list(reversed(sorted(glob.glob(os.path.join(directory, "*.pdf")))))
+    paths = sorted(Path(directory).glob("*.pdf"), key=lambda path: path.name.casefold())
-    return [u for u in l if "enonce" not in u]
+    return [path for path in paths if "enonce" not in path.name.casefold()]
+
+
+def _temporary_sibling(path: Path, purpose: str) -> Path:
+    return path.with_name(f".{path.name}.{purpose}.{uuid.uuid4().hex}.tmp")
+
+
+def commit_processed_pdf(
+    workspace: EvaluationWorkspace,
+    original_path: Path,
+    generated_path: Path,
+) -> Path:
+    """Commit a processed copy and its original backup with rollback."""
+    backup_path = workspace.original_copies_dir / original_path.name
+    output_path = workspace.copies_dir / original_path.name
+    workspace.original_copies_dir.mkdir(parents=True, exist_ok=True)
+    workspace.copies_dir.mkdir(parents=True, exist_ok=True)
+
+    staged_backup = None
+    if original_path.resolve() != backup_path.resolve():
+        staged_backup = _temporary_sibling(backup_path, "new-original")
+        shutil.copy2(original_path, staged_backup)
+    saved_backup = _temporary_sibling(backup_path, "old-original")
+    saved_output = _temporary_sibling(output_path, "old-output")
+    backup_replaced = False
+    output_replaced = False
+    try:
+        if staged_backup is not None:
+            if backup_path.exists():
+                backup_path.replace(saved_backup)
+            staged_backup.replace(backup_path)
+            backup_replaced = True
+        if output_path.exists():
+            output_path.replace(saved_output)
+        generated_path.replace(output_path)
+        output_replaced = True
+        if original_path.resolve() not in {
+            backup_path.resolve(),
+            output_path.resolve(),
+        }:
+            original_path.unlink()
+    except Exception:
+        if output_replaced and output_path.exists():
+            output_path.unlink()
+        if saved_output.exists():
+            saved_output.replace(output_path)
+        if backup_replaced and backup_path.exists():
+            backup_path.unlink()
+        if saved_backup.exists():
+            saved_backup.replace(backup_path)
+        raise
+    finally:
+        for temporary in (staged_backup, saved_backup, saved_output):
+            if temporary is not None and temporary.exists():
+                temporary.unlink()
+    return output_path
 
 class PDFPreviewer:
 
@@ -30,26 +98,36 @@ class PDFPreviewer:
             return False
         self.pdf_path = self.inputs.pop()
         self.file_rotation = 0
-        self.base_name = os.path.splitext(os.path.basename(self.pdf_path))[0]
+        self.base_name = self.pdf_path.stem
-        self.split_dir = f"{self.base_name}_split"
+        self._temporary_directory = tempfile.TemporaryDirectory(
-        self.reorder_dir = f"{self.base_name}_reorder"
+            prefix=f".{self.base_name}.page-splitter.",
-
+            dir=self.workspace.root,
-        # Create a temporary output file
+        )
-        self.final_file = f"{self.base_name}_temp.pdf"
+        working_dir = Path(self._temporary_directory.name)
+        self.split_dir = working_dir / "split"
+        self.reorder_dir = working_dir / "reorder"
+        self.final_file = working_dir / f"{self.base_name}.pdf"
 
         self.current_page_index = 0
         self.page_settings = []
         self.processing = False  # Flag to prevent multiple finish calls
         try:
             self.doc = fitz.open(self.pdf_path)
-        except Exception as e:
+        except (OSError, RuntimeError, ValueError) as e:
+            self.failed = True
+            self._temporary_directory.cleanup()
             messagebox.showerror("Error", f"Failed to open PDF file: {e}")
             self.master.destroy()
             return
-        self.master.title(f"PDF Splitter - {os.path.basename(self.pdf_path)}")
+        self.master.title(f"PDF Splitter - {self.pdf_path.name}")
         return True
 
-    def __init__(self, master, path):
+    def __init__(
+        self,
+        master: tk.Tk,
+        workspace: EvaluationWorkspace,
+        inputs: list[Path],
+    ) -> None:
         """
         Initializes the application.
 
@@ -57,40 +135,16 @@ class PDFPreviewer:
             master (tk.Tk): The root Tkinter window.
             pdf_path (str): The path to the input PDF file.
         """
-        if not os.path.exists(path):
+        self.workspace = workspace
-            messagebox.showerror("Error", f"File not found: {path}")
+        self.inputs = inputs
-            master.destroy()
-            return
-
-        if os.path.isdir(path):
-            self.inputs = list_pdf_files(path)
-        else:
-            # Check for existing original in backup and restore if found
-            dir_name = os.path.dirname(os.path.abspath(path))
-            file_name = os.path.basename(path)
-            if os.path.basename(dir_name) == "Copies":
-                dir_name = os.path.dirname(dir_name)
-                path = os.path.join(dir_name, file_name)
-            backup_path = os.path.join(dir_name, "Copies Originales", file_name)
-
-            if os.path.exists(backup_path):
-                try:
-                    shutil.move(backup_path, path)
-                    print(f"Restored original file from: {backup_path}")
-                except Exception as e:
-                    messagebox.showerror("Error", f"Failed to restore original file: {e}")
-                    master.destroy()
-                    return
-
-            self.inputs = [path]
-
         self.output_dir = None
         self.master = master
         self.num = 0
         self.global_rotation = 0 # Rotation appliquée à tous les fichiers
         self.history = []
+        self.failed = False
         if not self.setup_next_file():
-            print(f"Aucun fichier PDF valide trouvé dans : {path}")
+            print(f"No PDF files found in {workspace.root}")
             master.destroy()
             return
 
@@ -249,7 +303,7 @@ class PDFPreviewer:
             # Re-open the file from disk to reset changes (like moved pages)
             try:
                 self.doc = fitz.open(self.pdf_path)
-            except Exception as e:
+            except (OSError, RuntimeError, ValueError) as e:
                 messagebox.showerror("Error", f"Failed to reopen PDF file: {e}")
                 self.master.destroy()
                 return
@@ -328,7 +382,8 @@ class PDFPreviewer:
             self._initialize_current_page_settings()
             self.load_page()
         else:
-            self.finish_and_process()
+            if not self.finish_and_process():
+                return
             self.history.append(self.pdf_path)
             if self.setup_next_file():
                 self._initialize_current_page_settings()
@@ -336,59 +391,24 @@ class PDFPreviewer:
             else:
                 self.master.destroy()
 
-    def finish_and_process(self):
+    def finish_and_process(self) -> bool:
-        """Starts the PDF splitting process and moves files."""
+        """Render and transactionally install the processed PDF."""
-        self.split_pdf()
-        # print("Debug : ", self.page_settings)
-        # input("Splitting done. Continue ?")
-        self.reorder_pdfs()
-        # input("Reorder done. Continue ?")
-        self.concate_files()
-
-        # Logic to move original to backup and replace with new file
         try:
-            abs_path = os.path.abspath(self.pdf_path)
+            self.split_pdf()
-            dir_name = os.path.dirname(abs_path)
+            self.reorder_pdfs()
-            file_name = os.path.basename(abs_path)
+            self.concate_files()
-
+            commit_processed_pdf(self.workspace, self.pdf_path, self.final_file)
-            backup_dir = os.path.join(dir_name, "Copies Originales")
+        except Exception as exc:  # noqa: BLE001 - interactive boundary
-            copies_dir = os.path.join(dir_name, "Copies")
+            self.failed = True
-            os.makedirs(backup_dir, exist_ok=True)
+            self.processing = False
-            os.makedirs(copies_dir, exist_ok=True)
+            print(f"Failed to process {self.pdf_path}: {exc}")
-
+            messagebox.showerror("Error", f"Failed to process PDF: {exc}")
-            backup_path = os.path.join(backup_dir, file_name)
+            self._temporary_directory.cleanup()
-            copies_path = os.path.join(copies_dir, file_name)
+            self.master.destroy()
-
+            return False
-            # Remove backup if it already exists (overwrite)
+        else:
-            if os.path.exists(backup_path):
+            self._temporary_directory.cleanup()
-                os.remove(backup_path)
+            return True
-
-            # Move the original file to "Copies Originales"
-            shutil.move(self.pdf_path, backup_path)
-
-            # Move the temp output file to replace the original
-            shutil.move(self.final_file, copies_path)
-
-            # print(f"Original moved to {backup_path}, new file saved at {self.pdf_path}")
-
-        except Exception as e:
-            messagebox.showerror("Error", f"Failed to move/replace files: {e}")
-
-        self.remove_dirs()
-
-    def _restore_original(self, path):
-        """Restores the original file from the 'Copies Originales' backup."""
-        dir_name = os.path.dirname(os.path.abspath(path))
-        file_name = os.path.basename(path)
-        backup_path = os.path.join(dir_name, "Copies Originales", file_name)
-
-        if os.path.exists(backup_path):
-            try:
-                # Moving overwrites the generated PDF with the original backup
-                shutil.move(backup_path, path)
-                print(f"Restored original file from: {backup_path}")
-            except Exception as e:
-                print(f"Failed to restore original file: {e}")
 
     def go_to_previous_file(self, event=None):
         """Goes back to the beginning of the previously completed file."""
@@ -398,14 +418,16 @@ class PDFPreviewer:
         # Close the currently open document to avoid lock issues
         if hasattr(self, 'doc'):
             self.doc.close()
+        if hasattr(self, "_temporary_directory"):
+            self._temporary_directory.cleanup()
 
         # 1. Push current file back onto the stack so it processes next
         self.inputs.append(self.pdf_path)
 
-        # 2. Get the previous file, restore its original state, and push to stack
+        # 2. Reprocess the previous file from its preserved original backup
         prev_file = self.history.pop()
-        self._restore_original(prev_file)
+        backup = self.workspace.original_copies_dir / Path(prev_file).name
-        self.inputs.append(prev_file)
+        self.inputs.append(backup if backup.is_file() else Path(prev_file))
 
         # 3. Reload environment (setup_next_file will pop prev_file back off the stack)
         self.setup_next_file()
@@ -426,13 +448,9 @@ class PDFPreviewer:
         for pdf in pdf_files:
             try:
                 os.remove(pdf)
-            except Exception as e:
+            except OSError as e:
                 print(f"Error deleting {pdf}: {e}")
 
-    def remove_dirs(self):
-        shutil.rmtree(self.split_dir)
-        shutil.rmtree(self.reorder_dir)
-
     def split_pdf(self):
         """Splits each page of the PDF according to the saved settings."""
         print("Starting PDF processing...")
@@ -588,13 +606,56 @@ class PDFPreviewer:
         print(f"Created merged PDF: {self.final_file}")
 
 
-if __name__ == "__main__":
+def _selected_inputs(
-    if len(sys.argv) != 2:
+    workspace: EvaluationWorkspace,
-        print("Usage: python script_name.py <path_to_pdf_file>")
+    target: Path,
-        sys.exit(1)
+) -> list[Path]:
+    if target.is_file():
+        if target.suffix.casefold() != ".pdf":
+            raise CliError(f"Target is not a PDF: {target}", ExitCode.INVALID_ARGUMENTS)
+        backup = workspace.original_copies_dir / target.name
+        return [backup if backup.is_file() else target]
 
-    pdf_file_path = sys.argv[1]
+    directory = target
+    if target == workspace.copies_dir:
+        candidates = list_pdf_files(workspace.copies_dir)
+        candidates = [
+            (
+                workspace.original_copies_dir / path.name
+                if (workspace.original_copies_dir / path.name).is_file()
+                else path
+            )
+            for path in candidates
+        ]
+    else:
+        candidates = list_pdf_files(directory)
+    return list(reversed(candidates))
 
+
+def run(workspace: EvaluationWorkspace, target: Path) -> ExitCode:
+    inputs = _selected_inputs(workspace, target)
+    if not inputs:
+        print(f"No PDF files found in {target}")
+        return ExitCode.SUCCESS
     root = tk.Tk()
-    app = PDFPreviewer(root, pdf_file_path)
+    application = PDFPreviewer(root, workspace, inputs)
     root.mainloop()
+    return ExitCode.FAILURE if application.failed else ExitCode.SUCCESS
+
+
+def build_parser() -> argparse.ArgumentParser:
+    return target_parser("Interactively split and reorder scanned PDF pages")
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+
+    def handle(args: argparse.Namespace) -> ExitCode:
+        workspace, target = workspace_from_target(args)
+        return run(workspace, target)
+
+    return execute(parser, argv, handle)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

plotting.py

@@ -1,24 +1,29 @@
-import json
+from __future__ import annotations
+
+import argparse
 import queue
-import sys
 import threading
 import tkinter as tk
+from collections.abc import Sequence
 from pathlib import Path
 from tkinter import messagebox
 
 from PIL import Image, ImageDraw, ImageFont, ImageTk
 
+from copienator import (
+    EvaluationWorkspace,
+    ExitCode,
+    atomic_write_json,
+    execute,
+    read_json,
+    target_parser,
+    workspace_from_target,
+)
 from platform_utils import open_path
-
+from utils import natural_key, read_all_labels
-print("o to open pdf, O original pdf, e to emacs part, p to go back, i to interro, click for coordinates")
 
 # --- Configuration & Globals ---
 padding = 60
-valid_labels_set = None
-
-# Queue payload: (pil_image, json_path, metadata)
-# metadata is a dict: {'copie': str, 'part': int, 'schema': dict}
-image_queue = queue.Queue(maxsize=5)
 
 try:
     font = ImageFont.truetype("DejaVuSans.ttf", size=30)
@@ -55,6 +60,14 @@ def convert_list(l, group_id, json_schema):
         ll.append(ee)
     return ll
 
+
+def normalized_labels(entries):
+    return [
+        str(value["label"]).removeprefix("|").removesuffix("|")
+        for value in entries
+        if str(value["label"]) != "_"
+    ]
+
 def prepare_image(image_path: str, bounding_boxes, all_labels, nb_pages, last_label_index):
     im = Image.open(image_path)
     im.load()
@@ -94,7 +107,7 @@ def prepare_image(image_path: str, bounding_boxes, all_labels, nb_pages, last_la
 
 # --- Processing Logic (Worker Thread) ---
 
-def worker_thread(base_dir, files_to_process, all_labels):
+def _worker_items(base_dir, files_to_process, all_labels, output_queue):
     """
     Iterates through files, prepares VISUALS only, and puts metadata in queue.
     Does NOT write final JSON files anymore.
@@ -111,9 +124,8 @@ def worker_thread(base_dir, files_to_process, all_labels):
         json_schema_path = base_dir / 'Cutleft' / f"{copie}_schema.json"
 
         try:
-            with open(json_schema_path, 'r') as f:
+            json_schema = read_json(json_schema_path)
-                json_schema = json.load(f)
+        except (OSError, TypeError, ValueError):
-        except:
             print("No json_schema : ", json_schema_path)
             continue
 
@@ -124,11 +136,10 @@ def worker_thread(base_dir, files_to_process, all_labels):
             bb_list = []
             json_name = ""
             try:
-                with open(json_path, 'r') as f:
+                json_result = read_json(json_path)
-                    json_result = json.load(f)
                 bb_list = json_result.get("list", [])
                 json_name = json_result.get("name", "")
-            except Exception as e:
+            except Exception as e:  # noqa: BLE001 - malformed user-editable JSON
                 print(f"Warning: {json_path.name} is malformed! Loading blank. {e}")
                 # We do NOT skip; we continue so the user can fix it in the GUI
 
@@ -138,7 +149,7 @@ def worker_thread(base_dir, files_to_process, all_labels):
                     prepare_image(str(img_path), bb_list, all_labels, nb_pages, last_label_index)
                 error_msg = None
 
-            except Exception as e:
+            except Exception as e:  # noqa: BLE001 - keep the item editable in the GUI
                 print(f"Error processing {img_path.name}: {e}")
                 pil_image = Image.open(str(img_path))
                 error_msg = str(e)
@@ -151,15 +162,24 @@ def worker_thread(base_dir, files_to_process, all_labels):
                 "error": error_msg
             }
 
-            image_queue.put((pil_image, json_path, metadata))
+            output_queue.put((pil_image, json_path, metadata))
 
-    # Sentinel to indicate finished
+def worker_thread(base_dir, files_to_process, all_labels, output_queue):
-    image_queue.put((None, None, None))
+    """Prepare queue items and always terminate the GUI stream."""
+    failure = None
+    try:
+        _worker_items(base_dir, files_to_process, all_labels, output_queue)
+    except Exception as exc:  # noqa: BLE001 - worker boundary
+        failure = str(exc)
+        print(f"Plotting worker failed: {exc}")
+    finally:
+        metadata = {"worker_error": failure} if failure else None
+        output_queue.put((None, None, metadata))
 
 # --- GUI Logic (Main Thread) ---
 
 class ImageViewer:
-    def __init__(self, root, base_dir):
+    def __init__(self, root, workspace, valid_labels, input_queue):
         self.root = root
         self.root.resizable(False, False) # If you resize, coordinates will be wrong
 
@@ -171,7 +191,10 @@ class ImageViewer:
 
         root.geometry(f"+{x}+{y}")
 
-        self.base_dir = base_dir
+        self.workspace = workspace
+        self.base_dir = workspace.root
+        self.valid_labels = valid_labels
+        self.image_queue = input_queue
         self.root.title("Bounding Box Viewer")
         self.label = tk.Label(root, text="Waiting for images...")
         self.label.pack(expand=True, fill="both")
@@ -192,6 +215,7 @@ class ImageViewer:
         self.history = []
         self.forward_stack = []
         self.current_pil_image = None
+        self.failed = False
 
         from config import PLOTTING_KB
 
@@ -202,7 +226,8 @@ class ImageViewer:
         self.root.bind(PLOTTING_KB["open pdf"], self.on_open_pdf)
         self.root.bind(PLOTTING_KB["open original pdf"], self.on_open_ori_pdf)
         self.root.bind(PLOTTING_KB["open eval"], self.on_open_interro)
-        self.root.bind('<Escape>', lambda e: self.root.quit())
+        self.root.bind('<Escape>', lambda _event: self.close())
+        self.root.protocol("WM_DELETE_WINDOW", self.close)
         self.label.bind('<Button-1>', self.on_click)
 
         self.poll_queue()
@@ -214,10 +239,15 @@ class ImageViewer:
                 if self.forward_stack:
                     pil_image, json_path, metadata = self.forward_stack.pop()
                 else:
-                    pil_image, json_path, metadata = image_queue.get_nowait()
+                    pil_image, json_path, metadata = self.image_queue.get_nowait()
 
                 # Handle End of Stream
                 if pil_image is None:
+                    if metadata and metadata.get("worker_error"):
+                        self.failed = True
+                        messagebox.showerror(
+                            "Processing Error", metadata["worker_error"]
+                        )
                     self.save_current_batch() # Save any remaining data
                     print("All images processed.")
                     self.root.quit()
@@ -241,10 +271,12 @@ class ImageViewer:
         if self.active_copie_name and self.accumulated_results:
             main_json_path = self.base_dir / "Copies" / f"{self.active_copie_name}.json"
             print(f"Writing aggregated result to {main_json_path}")
-            with open(main_json_path, 'w') as f:
+            atomic_write_json(main_json_path, self.accumulated_results)
-                json.dump(self.accumulated_results, f)
             self.accumulated_results = None
 
+    def close(self):
+        self.root.quit()
+
 
     def on_previous(self, event):
         if self.is_viewing and self.history:
@@ -288,8 +320,7 @@ class ImageViewer:
             num_added = 0 # ADD THIS LINE
 
             try:
-                with open(self.current_json_path, 'r') as f:
+                current_data = read_json(self.current_json_path)
-                    current_data = json.load(f)
 
                 # Perform the conversion now, post-edit
                 converted_items = convert_list(
@@ -298,11 +329,10 @@ class ImageViewer:
                     self.current_meta["schema"]
                 )
 
-                labels = [v["label"] for v in current_data["list"]]
+                labels = normalized_labels(current_data["list"])
-                labels = [label for label in labels if label != "_"]
+                false_labels = [
-                labels = [label[1:] for label in labels if label[0] == "|"]
+                    label for label in labels if label not in self.valid_labels
-                labels = [label[:-1] for label in labels if label[-1] == "|"]
+                ]
-                false_labels = [label for label in labels if label not in valid_labels_set]
 
                 if false_labels:
                     msg = f"Wrong label in {self.current_json_path.name}: {false_labels}\n\n\tPlease press 'e' to fix it, then press Enter again."
@@ -318,7 +348,7 @@ class ImageViewer:
                     if "name" in current_data and current_data["name"] != "Continued":
                         self.accumulated_results["name"] = current_data["name"]
 
-            except Exception as e:
+            except Exception as e:  # noqa: BLE001 - interactive validation boundary
                 # Warn user and STOP (do not advance to next image)
                 msg = f"Error reading {self.current_json_path.name}:\n\n{e}\n\nPlease press 'e' to fix it, then press Enter again."
                 print(msg)
@@ -387,51 +417,58 @@ class ImageViewer:
         self.root.clipboard_clear()
         self.root.clipboard_append(box_str)
 
-from utils import natural_key, read_all_labels
+def _selected_images(workspace: EvaluationWorkspace, target: Path) -> list[Path]:
+    workspace.require_directories("Cutleft", "Copies")
+    if target.is_file():
+        stem = target.stem
+        exact = workspace.cutleft_dir / f"{stem}.jpg"
+        if exact.is_file():
+            return [exact]
+        return sorted(
+            workspace.cutleft_dir.glob(f"{stem}_*.jpg"),
+            key=natural_key,
+        )
+    return sorted(workspace.cutleft_dir.glob("*.jpg"), key=natural_key)
+
+
+def run(workspace: EvaluationWorkspace, target: Path) -> ExitCode:
+    workspace.require_files("labels")
+    all_labels = read_all_labels(workspace.root)
+    files_to_process = _selected_images(workspace, target)
+    if not files_to_process:
+        print(f"No Cutleft images found for {target}")
+        return ExitCode.PARTIAL
+
+    print(
+        "o to open pdf, O original pdf, e to edit part, p to go back, "
+        "i to open the statement, click for coordinates"
+    )
+    input_queue = queue.Queue(maxsize=5)
+    worker = threading.Thread(
+        target=worker_thread,
+        args=(workspace.root, files_to_process, all_labels, input_queue),
+        daemon=True,
+    )
+    worker.start()
+    root = tk.Tk()
+    application = ImageViewer(root, workspace, set(all_labels), input_queue)
+    root.mainloop()
+    return ExitCode.PARTIAL if application.failed else ExitCode.SUCCESS
+
+
+def build_parser() -> argparse.ArgumentParser:
+    return target_parser("Interactively verify detected label coordinates")
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+
+    def handle(args: argparse.Namespace) -> ExitCode:
+        workspace, target = workspace_from_target(args)
+        return run(workspace, target)
+
+    return execute(parser, argv, handle)
+
 
 if __name__ == "__main__":
-    if len(sys.argv) < 2:
+    raise SystemExit(main())
-        print("Usage: python plotting.py <directory_or_file>")
-        sys.exit(1)
-
-    input_path = Path(sys.argv[1])
-    files_to_process = []
-
-
-    if input_path.is_file():
-        # Correctly identify base_dir if we are in 'Copies' or 'Cutleft'
-        if input_path.parent.name in ["Copies", "Cutleft"]:
-            base_dir = input_path.parent.parent
-        else:
-            base_dir = input_path.parent
-
-        stem = input_path.stem
-        cutleft_dir = base_dir / "Cutleft"
-        img_path = cutleft_dir / f"{stem}.jpg"
-
-        if img_path.exists():
-            files_to_process = [img_path]
-        else:
-            # We're given something like Copie01.pdf, look for its split image parts
-            files_to_process = sorted(list(cutleft_dir.glob(f"{stem}_*.jpg")), key=natural_key)
-    else:
-        base_dir = input_path
-        cutleft_dir = base_dir / "Cutleft"
-        if not cutleft_dir.exists():
-            print(f"Error: {cutleft_dir} does not exist.")
-            sys.exit(1)
-        files_to_process = sorted(cutleft_dir.glob("*.jpg"))
-
-    try:
-        all_labels = read_all_labels(base_dir)
-    except FileNotFoundError:
-        all_labels = []
-    valid_labels_set = set(all_labels)
-
-    t = threading.Thread(target=worker_thread, args=(base_dir, files_to_process, all_labels))
-    t.daemon = True
-    t.start()
-
-    root = tk.Tk()
-    app = ImageViewer(root, base_dir)
-    root.mainloop()

splitting_int.py

@@ -1,223 +1,266 @@
-import fitz  # PyMuPDF
+from __future__ import annotations
-from pypdf import PdfWriter
+
-from pypdf import PdfReader
+import argparse
-import os
-import sys
-import json
 import shutil
-from pathlib import Path
+import tempfile
 from collections import defaultdict
+from collections.abc import Sequence
+from pathlib import Path
 
-from utils import read_all_labels
+import fitz
+from pypdf import PdfReader, PdfWriter
 
-carreau = 1000 // 38
+import utils
+from copienator import (
+    CliError,
+    EvaluationWorkspace,
+    ExitCode,
+    execute,
+    read_json,
+    target_parser,
+    workspace_from_target,
+)
+from copienator.filesystem import staged_directory
+
+SQUARE = 1000 // 38
+Coordinate = tuple[str, int, int, int, int, int]
+ParsedCoordinate = tuple[str, str, int, int, int, int, int]
 
 
-def decode_json(pdf_file):
+def decode_json(pdf_file: str | Path) -> tuple[str, list[Coordinate]]:
-    file_path = Path(pdf_file)
+    """Read verified label coordinates associated with one copy PDF."""
-    with open(file_path.with_suffix(".json"), "r") as f:
+    pdf_path = Path(pdf_file)
-        json_result = json.load(f)
+    loaded = read_json(pdf_path.with_suffix(".json"))
-
+    if not isinstance(loaded, dict):
-    nb_pages = len(PdfReader(file_path).pages)
+        raise TypeError(f"Expected a JSON object for {pdf_path}")
-
+    boxes = loaded.get("list")
-    bb_list = json_result["list"]
+    if not isinstance(boxes, list):
-    name = json_result["name"]
+        raise TypeError(f"Expected a list of labels for {pdf_path}")
-    column_width = 1000 // nb_pages
+    page_count = len(PdfReader(pdf_path).pages)
-
+    if page_count == 0:
-    def page_number(b):
+        raise ValueError(f"PDF contains no pages: {pdf_path}")
-        return ((b[1] + b[3]) // 2) // column_width
+    column_width = 1000 // page_count
-
+    result: list[Coordinate] = []
-    result = []
+    for entry in boxes:
-    for d in bb_list:
+        if not isinstance(entry, dict):
-        (b, label) = d["box_2d"], d["label"]
+            raise TypeError(f"Malformed label entry for {pdf_path}: {entry!r}")
-        pn = page_number(b)
+        box = entry["box_2d"]
-        result.append((label, pn, b[0] - carreau, b[2]-carreau, b[1], b[3]))
+        label = str(entry["label"])
-    result.sort(key=lambda x: (x[1], x[2]))
+        page_number = ((box[1] + box[3]) // 2) // column_width
-    return (name, result)
+        result.append(
+            (label, page_number, box[0] - SQUARE, box[2] - SQUARE, box[1], box[3])
+        )
+    result.sort(key=lambda item: (item[1], item[2]))
+    return str(loaded.get("name", "")), result
 
 
-def split_an_interro(base_dir, input_pdf, coords_list):
+def _parse_coordinates(coords_list: list[Coordinate]) -> list[ParsedCoordinate]:
-    doc = fitz.open(input_pdf)
+    parsed: list[ParsedCoordinate] = []
-
+    for label, page, y0, y1, x0, x1 in coords_list:
-    output_dir = base_dir / "Copies" / input_pdf.stem
-    generated_files = set()
-    parts_by_label = defaultdict(list)
-
-    # 1. Parse labels to strip '|' and determine type: L (Left), R (Right), N (Normal)
-    parsed_coords = []
-    for item in coords_list:
-        label, pn, y0, y1, x0, x1 = item
         if label.startswith("|"):
-            c_type, clean_label = "L", label[1:]
+            kind, clean_label = "L", label[1:]
         elif label.endswith("|"):
-            c_type, clean_label = "R", label[:-1]
+            kind, clean_label = "R", label[:-1]
         else:
-            c_type, clean_label = "N", label
+            kind, clean_label = "N", label
-        parsed_coords.append((clean_label, c_type, pn, y0, y1, x0, x1))
+        parsed.append((clean_label, kind, page, y0, y1, x0, x1))
+    filtered: list[ParsedCoordinate] = []
+    for item in parsed:
+        if not filtered or item[0] != filtered[-1][0]:
+            filtered.append(item)
+    return filtered
 
-    # 2. Filter consecutive duplicate labels based on the cleaned name
-    filtered_coords = []
-    if parsed_coords:
-        filtered_coords.append(parsed_coords[0])
-        for item in parsed_coords[1:]:
-            if item[0] != filtered_coords[-1][0]:
-                filtered_coords.append(item)
-    coords_list = filtered_coords
 
-    def scale_coord(y, page):
+def _save_cropped_page(
-        """Scale y from 0–1000 range to PDF points."""
+    document: fitz.Document,
-        page_height = page.rect.height
+    page_number: int,
-        return (y / 1000) * page_height
+    x0: float,
-
+    y0: float,
-    def save_cropped_page(doc, page_num, x0, y0, x1, y1, out_path):
+    x1: float,
-        """Saves a cropped portion of a page as a new PDF."""
+    y1: float,
-        page = doc[page_num]
+    output_path: Path,
-        rotated_rect = page.rect * page.transformation_matrix
+) -> None:
-        visual_crop_rect = fitz.Rect(rotated_rect.x0 + x0, y0, rotated_rect.x0 + x1, y1)
+    page = document[page_number]
-        unrotated_clip_rect = visual_crop_rect * page.derotation_matrix
+    rotated_rectangle = page.rect * page.transformation_matrix
-
+    visual_crop = fitz.Rect(
-        temp_doc = fitz.open()
+        rotated_rectangle.x0 + x0,
-        temp_page = temp_doc.new_page(
+        y0,
-            width=visual_crop_rect.width,
+        rotated_rectangle.x0 + x1,
-            height=visual_crop_rect.height
+        y1,
-        )
+    )
-        temp_page.show_pdf_page(
+    unrotated_clip = visual_crop * page.derotation_matrix
-            temp_page.rect,
+    cropped = fitz.open()
-            doc,
+    try:
-            page_num,
+        target_page = cropped.new_page(width=visual_crop.width, height=visual_crop.height)
+        target_page.show_pdf_page(
+            target_page.rect,
+            document,
+            page_number,
             rotate=-page.rotation,
-            clip=unrotated_clip_rect
+            clip=unrotated_clip,
         )
-        temp_doc.save(out_path)
+        cropped.save(output_path)
-        temp_doc.close()
+    finally:
+        cropped.close()
 
-    # Iterate through all labels
+
-    for idx, (clean_label, c_type, start_page, y_start_raw, y_end_box, x0_raw, x1_raw) in enumerate(coords_list):
+def _render_split_outputs(
-        if clean_label == "_":
+    input_pdf: Path,
+    coords_list: list[Coordinate],
+    staging: Path,
+) -> set[str]:
+    """Render every current answer into an otherwise empty staging directory."""
+    document = fitz.open(input_pdf)
+    try:
+        parsed = _parse_coordinates(coords_list)
+        parts_by_label: defaultdict[str, list[Path]] = defaultdict(list)
+        with tempfile.TemporaryDirectory(prefix="copienator-split-") as temp_directory:
+            temporary = Path(temp_directory)
+            for index, item in enumerate(parsed):
+                clean_label, kind, start_page, y_start, _y_end, x0_raw, _x1_raw = item
+                if clean_label == "_":
+                    continue
+                if not 0 <= start_page < document.page_count:
+                    raise ValueError(
+                        f"Invalid page {start_page} for {input_pdf.name}"
+                    )
+                end_page = document.page_count - 1
+                end_y = 1000
+                for next_item in parsed[index + 1 :]:
+                    _next_label, next_kind, next_page, next_y, *_rest = next_item
+                    if (
+                        (kind == "L" and next_kind in {"L", "N"})
+                        or (kind == "R" and next_kind in {"R", "N"})
+                        or kind == "N"
+                    ):
+                        end_page = next_page
+                        end_y = min(next_y + int(1.5 * SQUARE), 1000)
+                        break
+
+                column_width = 1000 / document.page_count
+                if kind == "L":
+                    fraction_x0 = (x0_raw % column_width) / column_width
+                    fraction_x1 = 1.0
+                    end_y = min(1000, end_y + 40)
+                elif kind == "R":
+                    fraction_x0 = 0.0
+                    left_labels = [entry for entry in parsed if entry[1] == "L"]
+                    if left_labels:
+                        closest = min(left_labels, key=lambda entry: abs(entry[3] - y_start))
+                        center = (closest[5] + closest[6]) / 2.0
+                        fraction_x1 = (center % column_width) / column_width
+                        if fraction_x1 <= fraction_x0:
+                            fraction_x1 = 1.0
+                    else:
+                        fraction_x1 = 1.0
+                else:
+                    fraction_x0, fraction_x1 = 0.0, 1.0
+
+                for page_number in range(start_page, end_page + 1):
+                    page = document[page_number]
+                    y0 = (y_start / 1000) * page.rect.height if page_number == start_page else 0
+                    y1 = (end_y / 1000) * page.rect.height if page_number == end_page else page.rect.height
+                    if y1 <= y0 + 1:
+                        continue
+                    part_path = temporary / f"part-{index}-{page_number}.pdf"
+                    _save_cropped_page(
+                        document,
+                        page_number,
+                        fraction_x0 * page.rect.width,
+                        y0,
+                        fraction_x1 * page.rect.width,
+                        y1,
+                        part_path,
+                    )
+                    parts_by_label[clean_label].append(part_path)
+
+            generated: set[str] = set()
+            for label, parts in parts_by_label.items():
+                filename = f"{label}.pdf"
+                merger = PdfWriter()
+                try:
+                    for part in parts:
+                        merger.append(part)
+                    merger.write(staging / filename)
+                finally:
+                    merger.close()
+                generated.add(filename)
+            return generated
+    finally:
+        document.close()
+
+
+def _preserve_previous_outputs(
+    output_dir: Path,
+    staging: Path,
+    generated_files: set[str],
+) -> None:
+    if not output_dir.is_dir():
+        return
+    for directory in (path for path in output_dir.iterdir() if path.is_dir()):
+        shutil.copytree(directory, staging / directory.name, dirs_exist_ok=True)
+    missing_dir = staging / "Missing"
+    for item in (path for path in output_dir.iterdir() if path.is_file()):
+        if item.name in generated_files:
             continue
+        print(f"ALERT: File '{item.name}' not generated. Moving to {missing_dir}")
+        missing_dir.mkdir(exist_ok=True)
+        shutil.copy2(item, missing_dir / item.name)
 
-        temp_parts = []
-        end_page = doc.page_count - 1
-        end_y_target_raw = 1000
 
-        # RULE 2: Determine stopping label
+def split_an_interro(
-        for next_item in coords_list[idx + 1:]:
+    workspace: EvaluationWorkspace,
-            n_clean, n_type, n_pn, n_y_start, n_y_end, _, _ = next_item
+    input_pdf: Path,
+    coords_list: list[Coordinate],
+) -> None:
+    """Regenerate one copy's answers and preserve obsolete ones under Missing."""
+    output_dir = workspace.copies_dir / input_pdf.stem
+    with staged_directory(output_dir) as staging:
+        generated = _render_split_outputs(input_pdf, coords_list, staging)
+        _preserve_previous_outputs(output_dir, staging, generated)
 
-            if c_type == "L":
-                is_stop = (n_type in ("L", "N"))
-            elif c_type == "R":
-                is_stop = (n_type in ("R", "N"))
-            else:
-                is_stop = True # Normal labels stop at anything
 
-            if is_stop:
+def _selected_pdfs(workspace: EvaluationWorkspace, target: Path) -> list[Path]:
-                end_page = n_pn
+    workspace.require_directories("Copies")
-                # end_y_target_raw = n_y_start
+    if target.is_file():
-                # On avait retiré un carreau précédemment inutilement, on le rajoute, plus un demi carreau
+        if target.suffix.casefold() != ".pdf":
-                end_y_target_raw = min(n_y_start + int(1.5 * carreau), 1000)
+            raise CliError(f"Target is not a PDF: {target}", ExitCode.INVALID_ARGUMENTS)
-                break
+        return [target]
+    return sorted(workspace.copies_dir.glob("*.pdf"), key=lambda path: path.name.casefold())
 
-        # RULES 3 & 4: Calculate horizontal boundaries (0.0 to 1.0 fraction of local page width)
-        col_w = 1000 / doc.page_count
-        if c_type == "L": # |name
-            fraction_x0 = (x0_raw % col_w) / col_w
-            fraction_x1 = 1.0
-            end_y_target_raw = min(1000, end_y_target_raw + 40)
-        elif c_type == "R": # name|
-            fraction_x0 = 0.0
-            # Find the closest 'L' label in y-distance
-            L_labels = [it for it in parsed_coords if it[1] == "L"]
-            if L_labels:
-                closest_L = min(L_labels, key=lambda it: abs(it[3] - y_start_raw))
-                closest_L_x_center = (closest_L[5] + closest_L[6]) / 2.0
-                fraction_x1 = (closest_L_x_center % col_w) / col_w
-                if fraction_x1 <= fraction_x0: fraction_x1 = 1.0 # Fallback
-            else:
-                fraction_x1 = 1.0
-        else: # Normal
-            fraction_x0 = 0.0
-            fraction_x1 = 1.0
 
-        current_p = start_page
+def run(workspace: EvaluationWorkspace, target: Path) -> ExitCode:
-        while current_p <= end_page:
+    workspace.require_files("labels")
-            page = doc[current_p]
+    utils.read_all_labels(workspace.root)
+    pdf_files = _selected_pdfs(workspace, target)
+    status = ExitCode.SUCCESS
+    for pdf_path in pdf_files:
+        json_path = pdf_path.with_suffix(".json")
+        if not json_path.is_file():
+            print(f"Warning: No JSON found for {pdf_path.name}")
+            status = ExitCode.PARTIAL
+            continue
+        name, coordinates = decode_json(pdf_path)
+        print(f"Decoded name: {name}")
+        split_an_interro(workspace, pdf_path, coordinates)
+    if not pdf_files:
+        print("No PDF copies found.")
+    return status
 
-            y0 = scale_coord(y_start_raw, page) if current_p == start_page else 0
-            y1 = scale_coord(end_y_target_raw, page) if current_p == end_page else page.rect.height
 
-            if y1 > y0 + 1:
+def build_parser() -> argparse.ArgumentParser:
-                # Convert fractions to absolute PDF points
+    return target_parser("Split verified PDF copies into answers by label")
-                x0_pdf = fraction_x0 * page.rect.width
-                x1_pdf = fraction_x1 * page.rect.width
 
-                temp_path = f"_part_{idx}_{current_p}.pdf"
-                save_cropped_page(doc, current_p, x0_pdf, y0, x1_pdf, y1, temp_path)
-                temp_parts.append(temp_path)
 
-            current_p += 1
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
 
-        parts_by_label[clean_label].extend(temp_parts)
+    def handle(args: argparse.Namespace) -> ExitCode:
+        workspace, target = workspace_from_target(args)
+        return run(workspace, target)
 
-    output_dir.mkdir(parents=True, exist_ok=True)
+    return execute(parser, argv, handle)
-
-    # Process aggregated parts by label
-    for title, parts in parts_by_label.items():
-        merger = PdfWriter()
-        for part in parts:
-            if os.path.exists(part):
-                merger.append(part)
-
-        filename = f"{title}.pdf"
-        merger.write(output_dir / filename)
-        merger.close()
-        generated_files.add(filename)
-
-        # Cleanup
-        for part in parts:
-            if os.path.exists(part):
-                os.remove(part)
-
-    doc.close()
-
-    # Move files not generated in this run to 'Missing' folder
-    if output_dir.exists():
-        missing_dir = output_dir / "Missing"
-        for item in output_dir.iterdir():
-            if item.is_file() and item.name not in generated_files:
-                print(f"ALERT: File '{item.name}' not generated. Moving to {missing_dir}")
-                missing_dir.mkdir(exist_ok=True)
-                item.rename(missing_dir / item.name)
 
 
 if __name__ == "__main__":
-    if len(sys.argv) < 2:
+    raise SystemExit(main())
-        print("Usage: python script.py <directory or pdf_file>")
-        sys.exit(1)
-
-    input_arg = Path(sys.argv[1])
-
-    if input_arg.is_file():
-        base_dir = input_arg.parent
-        if base_dir.name == "Copies":
-            base_dir = base_dir.parent
-        pdf_files = [input_arg]
-    elif input_arg.is_dir():
-        base_dir = input_arg
-        copies_dir = base_dir / "Copies"
-        pdf_files = sorted(copies_dir.glob("*.pdf"))
-    else:
-        print(f"Error: {input_arg} is not a valid file or directory.")
-        sys.exit(1)
-
-    read_all_labels(base_dir)
-
-    for pdf_path in pdf_files:
-            json_path = pdf_path.with_suffix(".json")
-            # print("Debug :", json_path)
-            if json_path.exists():
-                (name, coords) = decode_json(pdf_path)
-                print("Decoded name : ", name)
-                split_an_interro(base_dir, pdf_path, coords)
-            else:
-                print(f"Warning: No JSON found for {pdf_path.name}")

submit_batches.py

@@ -1,77 +1,96 @@
-import os
+from __future__ import annotations
-import sys
+
 import argparse
-from pathlib import Path
+from collections.abc import Sequence
+
 from google import genai
 from google.genai import types
 
-def main():
+import config
-    parser = argparse.ArgumentParser(description="Upload JSONL files and create Gemini Batch jobs.")
+from copienator import (
-    parser.add_argument("root_dir", type=str, help="Root directory containing the batch JSONL files")
+    CliError,
-    args = parser.parse_args()
+    EvaluationWorkspace,
+    ExitCode,
+    atomic_write_json,
+    evaluation_parser,
+    execute,
+    read_json,
+    workspace_from_args,
+)
 
-    root_dir = Path(args.root_dir)
 
-    if "GEMINI_API_KEY" not in os.environ:
+def run(workspace: EvaluationWorkspace, *, client=None) -> ExitCode:
-        sys.exit("Error: GEMINI_API_KEY environment variable not set.")
+    if client is None:
-
+        if not config.API_KEY:
-    client = genai.Client()
+            raise CliError("GEMINI_API_KEY is not configured")
-
+        client = genai.Client(api_key=config.API_KEY)
-    # Define the batch files and their corresponding models
+    batches = (
-    batches_to_create = [
+        (
-        {
+            "flash",
-            "file_path": root_dir / "batch_requests_flash.jsonl",
+            workspace.root / "batch_requests_flash.jsonl",
-            "model_id": "gemini-3-flash-preview",
+            config.MODEL_FLASH_ID,
-            "display_name": f"flash-correction-{root_dir.name}"
+            f"flash-correction-{workspace.name}",
-        },
+        ),
-        {
+        (
-            "file_path": root_dir / "batch_requests_pro.jsonl",
+            "pro",
-            "model_id": "gemini-3.1-pro-preview",
+            workspace.root / "batch_requests_pro.jsonl",
-            "display_name": f"pro-correction-{root_dir.name}"
+            config.MODEL_PRO_ID,
-        }
+            f"pro-correction-{workspace.name}",
-    ]
+        ),
-
+    )
-    for batch in batches_to_create:
+    manifest = {
-        file_path = batch["file_path"]
+        "version": 1,
-        model_id = batch["model_id"]
+        "evaluation": workspace.name,
-        display_name = batch["display_name"]
+        "jobs": {},
-
+    }
-        # Check if the file exists
+    if workspace.batch_jobs_file.is_file():
-        if not file_path.exists():
+        previous = read_json(workspace.batch_jobs_file)
+        if isinstance(previous, dict) and isinstance(previous.get("jobs"), dict):
+            manifest["jobs"] = previous["jobs"]
+    started = 0
+    for tier, file_path, model_id, display_name in batches:
+        if not file_path.is_file():
             print(f"Skipping {model_id}: {file_path.name} does not exist.")
             continue
-
-        # Check if the file is empty (e.g., if all tasks went to Flash, Pro might be empty)
         if file_path.stat().st_size == 0:
             print(f"Skipping {model_id}: {file_path.name} is empty.")
             continue
-
+        print(f"Uploading {file_path.name} for model {model_id}...")
-        print(f"Processing {file_path.name} for model {model_id}...")
+        uploaded = client.files.upload(
-
-        # 1. Upload the file to the File API
-        print(f"  Uploading file...")
-        uploaded_file = client.files.upload(
             file=str(file_path),
             config=types.UploadFileConfig(
                 display_name=f"{display_name}-input",
-                mime_type='jsonl'
+                mime_type="jsonl",
-            )
+            ),
         )
-        print(f"  Uploaded successfully! File ID: {uploaded_file.name}")
+        job = client.batches.create(
-
-        # 2. Create the batch job
-        print(f"  Starting batch job...")
-        batch_job = client.batches.create(
             model=model_id,
-            src=uploaded_file.name,
+            src=uploaded.name,
-            config={
+            config={"display_name": display_name},
-                'display_name': display_name,
-            },
         )
-        print(f"  Success! Batch Job Name: {batch_job.name}\n")
+        started += 1
+        manifest["jobs"][tier] = {
+            "name": job.name,
+            "display_name": display_name,
+            "model": model_id,
+            "request_file": file_path.name,
+        }
+        atomic_write_json(workspace.batch_jobs_file, manifest)
+        print(f"Started batch job: {job.name}")
+    if not started:
+        print("No non-empty batch request files were found.")
+        return ExitCode.PARTIAL
+    return ExitCode.SUCCESS
+
+
+def build_parser() -> argparse.ArgumentParser:
+    return evaluation_parser("Upload correction JSONL files and start Gemini batches")
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = build_parser()
+    return execute(parser, argv, lambda args: run(workspace_from_args(args)))
 
-    print("-" * 50)
-    print("All batch jobs have been initiated.")
 
 if __name__ == "__main__":
-    main()
+    raise SystemExit(main())

tests/test_gui_core.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import importlib.util
 import io
+import json
 import os
 import queue
 import sys
@@ -11,7 +12,8 @@ import unittest
 from concurrent.futures import ThreadPoolExecutor
 from contextlib import redirect_stderr
 from pathlib import Path
-from unittest.mock import patch
+from types import SimpleNamespace
+from unittest.mock import Mock, patch
 
 from PIL import Image
 from pypdf import PdfReader, PdfWriter
@@ -21,6 +23,7 @@ from copienator import (
     WorkspaceNotFoundError,
     WorkspaceValidationError,
     atomic_update_json,
+    atomic_write_bytes,
     atomic_write_json,
     read_json,
     workspace_from_target,
@@ -114,6 +117,12 @@ class WorkspaceTests(unittest.TestCase):
 
 
 class AtomicJsonTests(unittest.TestCase):
+    def test_atomic_binary_round_trip(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            path = Path(directory) / "result.jsonl"
+            atomic_write_bytes(path, b'{"one":1}\n')
+            self.assertEqual(path.read_bytes(), b'{"one":1}\n')
+
     def test_atomic_round_trip_and_unicode(self) -> None:
         with tempfile.TemporaryDirectory() as directory:
             path = Path(directory) / "nested" / "state.json"
@@ -302,6 +311,29 @@ class StandardCliTests(unittest.TestCase):
             "reading_grouped_annotations": load_script_module(
                 "reading_grouped_annotations.py", "reading_grouped_annotations"
             ),
+            "cutleft": load_script_module("cutleft.py", "cutleft"),
+            "splitting_int": load_script_module(
+                "splitting_int.py", "splitting_int"
+            ),
+            "page_splitter": load_script_module(
+                "page_splitter.py", "page_splitter"
+            ),
+            "plotting": load_script_module("plotting.py", "plotting"),
+            "gemini_for_labels": load_script_module(
+                "gemini_for_labels.py", "gemini_for_labels"
+            ),
+            "gemini_for_enonce": load_script_module(
+                "gemini_for_enonce.py", "gemini_for_enonce"
+            ),
+            "enonce_info": load_script_module("enonce_info.py", "enonce_info"),
+            "correction": load_script_module("correction.py", "correction"),
+            "submit_batches": load_script_module(
+                "submit_batches.py", "submit_batches"
+            ),
+            "batch_status": load_script_module("batch_status.py", "batch_status"),
+            "fetch_batched_results": load_script_module(
+                "fetch_batched_results.py", "fetch_batched_results"
+            ),
             "copies_tools": load_script_module(
                 "copies_tools.py", "copienator_copies_tools_test"
             ),
@@ -339,6 +371,16 @@ class StandardCliTests(unittest.TestCase):
                 "annotating_by_label": [missing],
                 "reading_annotations": [missing],
                 "reading_grouped_annotations": [missing],
+                "cutleft": [missing],
+                "splitting_int": [missing],
+                "page_splitter": [missing],
+                "plotting": [missing],
+                "gemini_for_labels": [missing],
+                "gemini_for_enonce": [missing],
+                "enonce_info": [missing],
+                "correction": [missing],
+                "submit_batches": [missing],
+                "fetch_batched_results": [missing],
             }
             for name, arguments in invocations.items():
                 with self.subTest(script=name), redirect_stderr(io.StringIO()):
@@ -383,6 +425,16 @@ class StandardCliTests(unittest.TestCase):
         steps = {step.id: step for step in build_workflow(True)}
         evaluation = "Evaluation with spaces"
         cases = {
+            "gemini_for_enonce": (
+                "statement",
+                "gemini",
+                {"target": evaluation, "restart": True},
+            ),
+            "enonce_info": (
+                "statement",
+                "personal",
+                {"target": evaluation},
+            ),
             "export": ("export", "default", {"target": evaluation, "refaire": True}),
             "import": ("import", "default", {"target": evaluation, "refaire": True}),
             "giving_names": (
@@ -431,6 +483,46 @@ class StandardCliTests(unittest.TestCase):
                 "grouped",
                 {"target": evaluation, "update_score": True, "refaire": True},
             ),
+            "cutleft": (
+                "cutleft",
+                "default",
+                {"target": evaluation, "fullpage": True},
+            ),
+            "splitting_int": (
+                "splitting",
+                "default",
+                {"target": evaluation},
+            ),
+            "page_splitter": (
+                "page_splitter",
+                "default",
+                {"target": evaluation},
+            ),
+            "plotting": (
+                "plotting",
+                "default",
+                {"target": evaluation},
+            ),
+            "gemini_for_labels": (
+                "labels",
+                "default",
+                {"target": evaluation, "overwrite": True},
+            ),
+            "correction": (
+                "correction",
+                "live",
+                {"target": evaluation, "overwrite": True, "limit": 5},
+            ),
+            "submit_batches": (
+                "submit_batches",
+                "default",
+                {"target": evaluation},
+            ),
+            "fetch_batched_results": (
+                "fetch_batches",
+                "default",
+                {"target": evaluation},
+            ),
         }
         for module_name, (step_id, variant_id, values) in cases.items():
             step = steps[step_id]
@@ -458,6 +550,21 @@ class StandardCliTests(unittest.TestCase):
                 self.assertEqual(parsed.operation, step_id)
                 self.assertEqual(str(parsed.evaluation), evaluation)
 
+    def test_enonce_info_preserves_labels_when_no_blocks_are_found(self) -> None:
+        module = self.modules["enonce_info"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            evaluation.mkdir()
+            (evaluation / "source.tex").write_text(
+                "No SHEETINFO blocks\n", encoding="utf-8"
+            )
+            (evaluation / "labels").write_text("Existing\n", encoding="utf-8")
+            self.assertEqual(
+                module.process_directory(EvaluationWorkspace(evaluation)), 4
+            )
+            self.assertEqual((evaluation / "labels").read_text(), "Existing\n")
+            self.assertFalse(list(evaluation.glob(".labels.*.tmp")))
+
     def test_export_main_copies_outputs(self) -> None:
         module = self.modules["export"]
         with tempfile.TemporaryDirectory() as directory:
@@ -513,6 +620,481 @@ class StandardCliTests(unittest.TestCase):
             self.assertEqual(module.main([str(evaluation)]), 0)
             self.assertTrue((evaluation / "Par label").is_dir())
 
+    def test_cutleft_can_target_one_copy_without_rendering_at_import(self) -> None:
+        module = self.modules["cutleft"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            copy_pdf = evaluation / "Copies" / "Copie01.pdf"
+            copy_pdf.parent.mkdir(parents=True)
+            copy_pdf.write_bytes(b"pdf")
+            with patch.object(module, "ImageReviewer") as reviewer:
+                self.assertEqual(module.main([str(copy_pdf), "--fullpage"]), 0)
+            files, output_dir = reviewer.call_args.args[:2]
+            self.assertEqual(files, [copy_pdf])
+            self.assertEqual(output_dir, evaluation / "Cutleft")
+            self.assertEqual(reviewer.call_args.kwargs["default_max_per_file"], 1)
+
+    def test_cutleft_atomic_save_removes_only_obsolete_copy_outputs(self) -> None:
+        module = self.modules["cutleft"]
+        with tempfile.TemporaryDirectory() as directory:
+            output_dir = Path(directory) / "Cutleft"
+            output_dir.mkdir()
+            (output_dir / "Copie01_01.jpg").write_bytes(b"old-one")
+            (output_dir / "Copie01_02.jpg").write_bytes(b"old-two")
+            (output_dir / "Copie02_01.jpg").write_bytes(b"other-copy")
+            (output_dir / "Copie01_schema.json").write_text(
+                "{}", encoding="utf-8"
+            )
+            image = Image.new("RGB", (5, 5), "white")
+            result = (
+                image,
+                [image],
+                {
+                    "original_filename": "Copie01.pdf",
+                    "total_pages": 1,
+                    "number_of_files": 1,
+                    "columns_per_file": [1],
+                },
+            )
+            module.save_results(result, Path("Copie01.pdf"), output_dir)
+            self.assertTrue((output_dir / "Copie01_01.jpg").is_file())
+            self.assertFalse((output_dir / "Copie01_02.jpg").exists())
+            self.assertEqual(
+                (output_dir / "Copie02_01.jpg").read_bytes(), b"other-copy"
+            )
+            self.assertEqual(
+                read_json(output_dir / "Copie01_schema.json")["total_pages"], 1
+            )
+
+    def test_splitting_missing_json_is_partial(self) -> None:
+        module = self.modules["splitting_int"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            copy_pdf = evaluation / "Copies" / "Copie01.pdf"
+            copy_pdf.parent.mkdir(parents=True)
+            copy_pdf.write_bytes(b"pdf")
+            (evaluation / "labels").write_text("Ex 1\n", encoding="utf-8")
+            self.assertEqual(module.main([str(evaluation)]), 4)
+
+    def test_splitting_failure_preserves_previous_copy_outputs(self) -> None:
+        module = self.modules["splitting_int"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            copy_pdf = evaluation / "Copies" / "Copie01.pdf"
+            output_dir = evaluation / "Copies" / "Copie01"
+            output_dir.mkdir(parents=True)
+            copy_pdf.write_bytes(b"pdf")
+            (output_dir / "sentinel.pdf").write_bytes(b"old")
+            workspace = EvaluationWorkspace(evaluation)
+            with patch.object(
+                module,
+                "_render_split_outputs",
+                side_effect=RuntimeError("render failed"),
+            ), self.assertRaises(RuntimeError):
+                module.split_an_interro(workspace, copy_pdf, [])
+            self.assertEqual((output_dir / "sentinel.pdf").read_bytes(), b"old")
+
+    def test_splitting_moves_obsolete_outputs_to_missing_on_commit(self) -> None:
+        module = self.modules["splitting_int"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            copy_pdf = evaluation / "Copies" / "Copie01.pdf"
+            output_dir = evaluation / "Copies" / "Copie01"
+            missing_dir = output_dir / "Missing"
+            missing_dir.mkdir(parents=True)
+            copy_pdf.write_bytes(b"pdf")
+            (output_dir / "Old.pdf").write_bytes(b"obsolete")
+            (missing_dir / "Earlier.pdf").write_bytes(b"earlier")
+
+            def render(_pdf, _coordinates, staging):
+                (staging / "Ex 1.pdf").write_bytes(b"new")
+                return {"Ex 1.pdf"}
+
+            with patch.object(module, "_render_split_outputs", side_effect=render):
+                module.split_an_interro(EvaluationWorkspace(evaluation), copy_pdf, [])
+            self.assertEqual((output_dir / "Ex 1.pdf").read_bytes(), b"new")
+            self.assertEqual(
+                (output_dir / "Missing" / "Old.pdf").read_bytes(), b"obsolete"
+            )
+            self.assertEqual(
+                (output_dir / "Missing" / "Earlier.pdf").read_bytes(), b"earlier"
+            )
+
+    def test_splitting_renders_a_real_one_page_answer(self) -> None:
+        module = self.modules["splitting_int"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            copy_pdf = evaluation / "Copies" / "Copie01.pdf"
+            copy_pdf.parent.mkdir(parents=True)
+            document = module.fitz.open()
+            page = document.new_page(width=600, height=800)
+            page.insert_text((100, 200), "Student answer")
+            document.save(copy_pdf)
+            document.close()
+            (evaluation / "labels").write_text("Ex 1\n", encoding="utf-8")
+            atomic_write_json(
+                copy_pdf.with_suffix(".json"),
+                {
+                    "name": "Copie01",
+                    "list": [{"label": "Ex 1", "box_2d": [100, 100, 300, 300]}],
+                },
+            )
+
+            self.assertEqual(module.main([str(evaluation)]), 0)
+            answer = evaluation / "Copies" / "Copie01" / "Ex 1.pdf"
+            self.assertTrue(answer.is_file())
+            self.assertEqual(len(PdfReader(answer).pages), 1)
+
+    def test_page_splitter_commits_original_and_generated_copy(self) -> None:
+        module = self.modules["page_splitter"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            evaluation.mkdir()
+            original = evaluation / "Copie01.pdf"
+            generated = evaluation / ".generated.pdf"
+            original.write_bytes(b"original")
+            generated.write_bytes(b"processed")
+            workspace = EvaluationWorkspace(evaluation)
+
+            output = module.commit_processed_pdf(workspace, original, generated)
+            self.assertEqual(output, evaluation / "Copies" / "Copie01.pdf")
+            self.assertEqual(output.read_bytes(), b"processed")
+            self.assertEqual(
+                (evaluation / "Copies Originales" / "Copie01.pdf").read_bytes(),
+                b"original",
+            )
+            self.assertFalse(original.exists())
+
+    def test_page_splitter_commit_failure_rolls_back_both_files(self) -> None:
+        module = self.modules["page_splitter"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            original = evaluation / "Copie01.pdf"
+            backup = evaluation / "Copies Originales" / "Copie01.pdf"
+            output = evaluation / "Copies" / "Copie01.pdf"
+            backup.parent.mkdir(parents=True)
+            output.parent.mkdir()
+            original.write_bytes(b"new original")
+            backup.write_bytes(b"previous original")
+            output.write_bytes(b"previous output")
+            missing_generated = evaluation / "missing-generated.pdf"
+
+            with self.assertRaises(FileNotFoundError):
+                module.commit_processed_pdf(
+                    EvaluationWorkspace(evaluation), original, missing_generated
+                )
+            self.assertEqual(original.read_bytes(), b"new original")
+            self.assertEqual(backup.read_bytes(), b"previous original")
+            self.assertEqual(output.read_bytes(), b"previous output")
+
+    def test_page_splitter_reprocesses_from_preserved_original(self) -> None:
+        module = self.modules["page_splitter"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            generated = evaluation / "Copies" / "Copie01.pdf"
+            original = evaluation / "Copies Originales" / "Copie01.pdf"
+            generated.parent.mkdir(parents=True)
+            original.parent.mkdir()
+            generated.write_bytes(b"processed")
+            original.write_bytes(b"original")
+            workspace = EvaluationWorkspace(evaluation)
+            self.assertEqual(
+                module._selected_inputs(workspace, generated),
+                [original],
+            )
+
+    def test_plotting_batch_save_is_atomic(self) -> None:
+        module = self.modules["plotting"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            (evaluation / "Copies").mkdir(parents=True)
+            viewer = module.ImageViewer.__new__(module.ImageViewer)
+            viewer.base_dir = evaluation
+            viewer.active_copie_name = "Copie01"
+            viewer.accumulated_results = {
+                "name": "Test",
+                "list": [{"label": "Ex 1"}],
+            }
+            viewer.save_current_batch()
+            self.assertEqual(
+                read_json(evaluation / "Copies" / "Copie01.json"),
+                {"name": "Test", "list": [{"label": "Ex 1"}]},
+            )
+            self.assertIsNone(viewer.accumulated_results)
+
+    def test_plotting_validates_plain_and_directional_labels(self) -> None:
+        module = self.modules["plotting"]
+        self.assertEqual(
+            module.normalized_labels(
+                [
+                    {"label": "Ex 1"},
+                    {"label": "|Ex 2"},
+                    {"label": "Ex 3|"},
+                    {"label": "_"},
+                ]
+            ),
+            ["Ex 1", "Ex 2", "Ex 3"],
+        )
+
+    def test_plotting_worker_failure_still_terminates_its_queue(self) -> None:
+        module = self.modules["plotting"]
+        output_queue = queue.Queue()
+        with patch.object(
+            module, "_worker_items", side_effect=RuntimeError("worker failed")
+        ):
+            module.worker_thread(Path("."), [], [], output_queue)
+        image, json_path, metadata = output_queue.get_nowait()
+        self.assertIsNone(image)
+        self.assertIsNone(json_path)
+        self.assertEqual(metadata, {"worker_error": "worker failed"})
+
+    def test_label_detection_resolves_copy_and_cutleft_targets(self) -> None:
+        module = self.modules["gemini_for_labels"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            copy_pdf = evaluation / "Copies" / "Copie01.pdf"
+            image_one = evaluation / "Cutleft" / "Copie01_01.jpg"
+            image_two = evaluation / "Cutleft" / "Copie01_02.jpg"
+            copy_pdf.parent.mkdir(parents=True)
+            image_one.parent.mkdir()
+            copy_pdf.write_bytes(b"pdf")
+            image_one.write_bytes(b"image")
+            image_two.write_bytes(b"image")
+            workspace = EvaluationWorkspace(evaluation)
+
+            images, warnings = module.selected_images(workspace, [copy_pdf])
+            self.assertEqual(images, [image_one, image_two])
+            self.assertEqual(warnings, [])
+            images, warnings = module.selected_images(workspace, [image_two])
+            self.assertEqual(images, [image_two])
+            self.assertEqual(warnings, [])
+
+    def test_label_detection_preserves_context_and_writes_atomically(self) -> None:
+        module = self.modules["gemini_for_labels"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            copies = evaluation / "Copies"
+            cutleft = evaluation / "Cutleft"
+            copies.mkdir(parents=True)
+            cutleft.mkdir()
+            first = cutleft / "Copie01_01.jpg"
+            second = cutleft / "Copie01_02.jpg"
+            first.write_bytes(b"first image")
+            second.write_bytes(b"second image")
+            atomic_write_json(
+                copies / "Copie01_01.json",
+                {
+                    "name": "Student",
+                    "list": [{"box_2d": [1, 2, 3, 4], "label": "Ex 1"}],
+                },
+            )
+            response = Mock(
+                text=(
+                    '{"name":"Continued","list":'
+                    '[{"box_2d":[10,20,30,40],"label":"Ex 2"}]}'
+                )
+            )
+            client = Mock()
+            client.models.generate_content.return_value = response
+
+            generated = module.process_copy_group(
+                EvaluationWorkspace(evaluation),
+                "Copie01",
+                [first, second],
+                client=client,
+                labels_text="Ex 1\nEx 2\n",
+                names_text="Student\n",
+                valid_labels={"Ex 1", "Ex 2"},
+                valid_names={"Student", "Continued", "Unknown"},
+                overwrite=False,
+                sleep=lambda _seconds: None,
+                target_interval=0,
+            )
+            self.assertEqual(generated, 1)
+            self.assertEqual(client.models.generate_content.call_count, 1)
+            self.assertEqual(
+                read_json(copies / "Copie01_02.json"),
+                {
+                    "name": "Continued",
+                    "list": [{"box_2d": [10, 20, 30, 40], "label": "Ex 2"}],
+                },
+            )
+
+    def test_label_detection_retries_unknown_labels(self) -> None:
+        module = self.modules["gemini_for_labels"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            (evaluation / "Copies").mkdir(parents=True)
+            image = evaluation / "Cutleft" / "Copie01_01.jpg"
+            image.parent.mkdir()
+            image.write_bytes(b"image")
+            client = Mock()
+            client.models.generate_content.side_effect = [
+                Mock(
+                    text=(
+                        '{"name":"Student","list":'
+                        '[{"box_2d":[1,2,3,4],"label":"Wrong"}]}'
+                    )
+                ),
+                Mock(
+                    text=(
+                        '{"name":"Student","list":'
+                        '[{"box_2d":[1,2,3,4],"label":"Ex 1"}]}'
+                    )
+                ),
+            ]
+            sleeps = []
+            module.process_copy_group(
+                EvaluationWorkspace(evaluation),
+                "Copie01",
+                [image],
+                client=client,
+                labels_text="Ex 1\n",
+                names_text="Student\n",
+                valid_labels={"Ex 1"},
+                valid_names={"Student", "Unknown", "Continued"},
+                overwrite=True,
+                sleep=sleeps.append,
+                target_interval=0,
+            )
+            self.assertEqual(client.models.generate_content.call_count, 2)
+            self.assertIn(10, sleeps)
+            self.assertEqual(
+                read_json(evaluation / "Copies" / "Copie01_01.json")["list"][0][
+                    "label"
+                ],
+                "Ex 1",
+            )
+
+    def test_correction_overwrite_keeps_previous_state_until_a_commit(self) -> None:
+        module = self.modules["correction"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            evaluation.mkdir()
+            correction = evaluation / "correction.json"
+            progress = evaluation / "correction_progress.json"
+            atomic_write_json(correction, {"Ex 1": [[{"id": "01"}]]})
+            atomic_write_json(progress, [["old.jpg", "Ex 1"]])
+            args = module.build_parser().parse_args(
+                [str(evaluation), "--overwrite"]
+            )
+            module.configure_runtime(
+                EvaluationWorkspace(evaluation),
+                [("new.jpg", "Ex 1")],
+                args,
+                api_client=Mock(),
+            )
+            self.assertEqual(read_json(correction), {"Ex 1": [[{"id": "01"}]]})
+            self.assertEqual(read_json(progress), [["old.jpg", "Ex 1"]])
+            self.assertEqual(module.results, {"Ex 1": []})
+            self.assertEqual(module.tasks_to_process, [("new.jpg", "Ex 1")])
+
+    def test_correction_reset_restores_old_and_deletes_new_files(self) -> None:
+        module = self.modules["correction"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            copy_dir = evaluation / "Copies" / "Copie01"
+            copy_dir.mkdir(parents=True)
+            atomic_write_json(evaluation / "correction.json", {"old": True})
+            atomic_write_json(evaluation / "correction_progress.json", ["old"])
+            (copy_dir / "Ex 1.pdf").write_bytes(b"current")
+            (copy_dir / "Ex 1_old.pdf").write_bytes(b"original")
+            (copy_dir / "Ex 2_new.pdf").write_bytes(b"generated")
+
+            self.assertEqual(module.main([str(evaluation), "--reset"]), 0)
+            self.assertFalse((evaluation / "correction.json").exists())
+            self.assertFalse((evaluation / "correction_progress.json").exists())
+            self.assertEqual((copy_dir / "Ex 1.pdf").read_bytes(), b"original")
+            self.assertFalse((copy_dir / "Ex 1_old.pdf").exists())
+            self.assertFalse((copy_dir / "Ex 2_new.pdf").exists())
+
+    def test_correction_batch_request_files_are_written_atomically(self) -> None:
+        module = self.modules["correction"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            group_dir = evaluation / "Par label" / "Ex 1"
+            (evaluation / "Copies").mkdir(parents=True)
+            group_dir.mkdir(parents=True)
+            (evaluation / "labels").write_text("Ex 1\n", encoding="utf-8")
+            image = group_dir / "Group_1.jpg"
+            image.write_bytes(b"image")
+            atomic_write_json(
+                image.with_suffix(".json"),
+                [["01", 0, 400, 1.0, "Ex 1"]],
+            )
+            args = module.build_parser().parse_args([str(evaluation), "--batch"])
+            module.configure_runtime(
+                EvaluationWorkspace(evaluation),
+                [(str(image), "Ex 1")],
+                args,
+                api_client=Mock(),
+            )
+            with patch.object(module.prompting, "make_prompt", return_value="prompt"):
+                self.assertEqual(module.run_configured(args), 0)
+            lines = (evaluation / "batch_requests_flash.jsonl").read_text().splitlines()
+            self.assertEqual(len(lines), 1)
+            self.assertEqual(json.loads(lines[0])["key"], str(image))
+            self.assertEqual(
+                (evaluation / "batch_requests_pro.jsonl").read_text(), ""
+            )
+
+    def test_batch_helpers_use_mocked_api_and_atomic_combination(self) -> None:
+        submit = self.modules["submit_batches"]
+        fetch = self.modules["fetch_batched_results"]
+        status = self.modules["batch_status"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            evaluation.mkdir()
+            (evaluation / "batch_requests_flash.jsonl").write_text(
+                "{}\n", encoding="utf-8"
+            )
+            client = Mock()
+            client.files.upload.return_value = SimpleNamespace(name="files/input")
+            client.batches.create.return_value = SimpleNamespace(name="batches/1")
+            self.assertEqual(
+                submit.run(EvaluationWorkspace(evaluation), client=client), 0
+            )
+            self.assertEqual(client.batches.create.call_count, 1)
+            manifest = json.loads((evaluation / "batch_jobs.json").read_text())
+            self.assertEqual(manifest["jobs"]["flash"]["name"], "batches/1")
+
+            jobs = [
+                SimpleNamespace(
+                    name="batches/1",
+                    display_name=f"flash-correction-{evaluation.name}",
+                    state=SimpleNamespace(name="JOB_STATE_SUCCEEDED"),
+                    dest=SimpleNamespace(file_name="files/flash-result"),
+                ),
+                SimpleNamespace(
+                    name="batches/2",
+                    display_name=f"pro-correction-{evaluation.name}",
+                    state=SimpleNamespace(name="JOB_STATE_SUCCEEDED"),
+                    dest=SimpleNamespace(file_name="files/pro-result"),
+                ),
+            ]
+            client.batches.get.return_value = jobs[0]
+            client.files.download.side_effect = [b'{"flash":1}\n', b'{"pro":1}']
+            self.assertEqual(
+                fetch.run(EvaluationWorkspace(evaluation), client=client), 0
+            )
+            self.assertEqual(
+                (evaluation / "batched_correction_result.jsonl").read_bytes(),
+                b'{"flash":1}\n',
+            )
+
+            job = SimpleNamespace(
+                state=SimpleNamespace(name="JOB_STATE_SUCCEEDED"),
+                dest=SimpleNamespace(file_name="files/result"),
+            )
+            client.batches.get.return_value = job
+            client.files.download.side_effect = None
+            client.files.download.return_value = b'{"downloaded":true}\n'
+            output = evaluation / "one-result.jsonl"
+            self.assertEqual(
+                status.download_job("batches/1", output=output, client=client), 0
+            )
+            self.assertEqual(output.read_bytes(), b'{"downloaded":true}\n')
+
     def test_post_correction_main_cleans_json_atomically(self) -> None:
         module = self.modules["post_correction"]
         with tempfile.TemporaryDirectory() as directory: