améliorations diverses

.gitignore

@@ -1,4 +1,6 @@
 OLD/
+/Interro*/
+/DS*/
 __pycache__/
 *.py[cod]
 .venv/

Readme.org

@@ -197,13 +197,17 @@ racine du projet). Dans la console, les boutons de copie et le clic droit
 permettent de copier la sélection ou toute la sortie ; =Ctrl+C= et
 =Ctrl+A= sont également disponibles (=Cmd= sous macOS).
 
-Pendant =Découper la marge des labels=, =s= signale la copie en erreur
-et passe à la suivante. =Entrée= valide et enregistre la découpe affichée.
+Pendant =Découper une partie à gauche pour détection des labels=, =n= décale
+la zone de 50 px vers la droite, =N= de 100 px, =t= de 50 px vers la
+gauche et =l= l’élargit de 50 px. =1= utilise les pages entières. =s=
+signale la copie en erreur et passe à la suivante ; =Entrée= valide et
+enregistre la découpe affichée.
 Une copie ignorée conserve ses anciennes découpes. Les signalements sont
 conservés dans =.copienator/copy_errors.json=, même après fermeture.
 Dans =Séparer et réordonner les pages=, =Traiter les copies signalées=
 reprend ces copies à partir des originaux conservés. Le même bouton dans
-=Découper la marge des labels= reprend leurs marges ; chaque signalement
+=Découper une partie à gauche pour détection des labels= reprend leur
+découpage ; chaque signalement
 est effacé seulement après validation et enregistrement avec =Entrée=.
 Fermer la fenêtre, appuyer à nouveau sur =s= ou rencontrer une erreur
 conserve le signalement. Les commandes =page-split= et =crop-labels=

copienator/commands/correction.py

@@ -5,6 +5,7 @@ import base64
 import concurrent.futures
 import json
 import os
+import signal
 import shlex
 import shutil
 import sys
@@ -45,15 +46,22 @@ COPIES_DIR = Path()
 GROUPS_DIR = Path()
 output_path = Path()
 progress_path = Path()
+pending_responses_path = Path()
 tasks: list[tuple] = []
 tasks_to_process: list[tuple] = []
 results: dict = {}
 completed_tasks: list = []
 errors_summary: list = []
+pending_responses: dict[str, str] = {}
 overwrite = False
 limit = None
 client = None
 start_time = 0.0
+stop_requested = threading.Event()
+
+
+class CorrectionStopRequested(Exception):
+    """Raised in a worker before it starts another Gemini request."""
 
 # --- Thread-safe Logging ---
 log_lock = threading.Lock()
@@ -82,6 +90,25 @@ def flush_thread_log(tid=None):
                 f.write("\n".join(thread_logs[tid]) + "\n\n")
             thread_logs[tid].clear()
 
+
+def report_group_progress(completed: int, total: int) -> None:
+    """Emit a stable, human-readable progress line for the GUI and CLI."""
+    print(
+        f"[Progression correction] Groupes traités : {completed}/{total}",
+        flush=True,
+    )
+
+
+def request_graceful_stop(_signum=None, _frame=None) -> None:
+    """Stop scheduling Gemini calls while allowing in-flight calls to finish."""
+    if not stop_requested.is_set():
+        stop_requested.set()
+        print(
+            "\n[Interruption] Arrêt des nouveaux appels Gemini demandé. "
+            "Attente des appels en cours et sauvegarde de leurs résultats…",
+            flush=True,
+        )
+
 # --- Lock for thread-safe file writing ---
 io_lock = threading.Lock()
 pro_lock = threading.Lock()
@@ -140,6 +167,7 @@ def configure_runtime(
     api_client=None,
 ) -> None:
     global INPUT_DIR, COPIES_DIR, GROUPS_DIR, output_path, progress_path
+    global pending_responses_path, pending_responses
     global tasks, tasks_to_process, results, completed_tasks, errors_summary
     global overwrite, limit, client, start_time
     global pro_count, flash_count, pro_quota_exhausted
@@ -149,11 +177,13 @@ def configure_runtime(
     GROUPS_DIR = workspace.groups_dir
     output_path = workspace.correction_file
     progress_path = workspace.correction_progress_file
+    pending_responses_path = workspace.correction_pending_responses_file
     tasks = list(discovered_tasks)
     overwrite = bool(args.overwrite)
     limit = args.limit
     start_time = time.time()
     errors_summary = []
+    pending_responses = {}
     completed_tasks = []
     results = {label: [] for _file, label in tasks}
     thread_logs.clear()
@@ -161,6 +191,7 @@ def configure_runtime(
     pro_count = 0
     flash_count = 0
     pro_quota_exhausted = False
+    stop_requested.clear()
 
     if not overwrite:
         if progress_path.is_file():
@@ -174,6 +205,20 @@ def configure_runtime(
                 raise TypeError("correction.json must contain a JSON object")
             results = loaded_results
 
+    # A response saved during a graceful interruption is not a completed
+    # correction. Resume its auxiliary checks even with --overwrite instead of
+    # paying for the same primary request again.
+    if pending_responses_path.is_file():
+        loaded_pending = read_json(pending_responses_path)
+        if not isinstance(loaded_pending, dict) or not all(
+            isinstance(key, str) and isinstance(value, str)
+            for key, value in loaded_pending.items()
+        ):
+            raise TypeError(
+                "correction_pending_responses.json must contain a JSON object"
+            )
+        pending_responses = loaded_pending
+
     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
@@ -184,7 +229,11 @@ def configure_runtime(
 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):
+    for path in (
+        workspace.correction_file,
+        workspace.correction_progress_file,
+        workspace.correction_pending_responses_file,
+    ):
         if path.exists():
             path.unlink()
             print(f"Deleted: {path}")
@@ -213,6 +262,8 @@ def call_gemini_with_retries(model_id, contents, config,
     delays = [60, 300]
 
     for attempt in range(3):
+        if stop_requested.is_set():
+            raise CorrectionStopRequested
         # Switch to fallback immediately if quota was exhausted by another thread
         if model_id == MODEL_ID_pro and pro_quota_exhausted and fallback_model_id:
             model_id = fallback_model_id
@@ -228,6 +279,8 @@ def call_gemini_with_retries(model_id, contents, config,
                     full_response_text += chunk.text
             return full_response_text
         except Exception as e:
+            if stop_requested.is_set():
+                raise CorrectionStopRequested from e
             error_msg = str(e).lower()
             is_quota_error = "429" in error_msg or "quota" in error_msg or "exhausted" in error_msg
             is_minute_limit = "minute" in error_msg or "rpm" in error_msg or "tpm" in error_msg
@@ -239,7 +292,8 @@ def call_gemini_with_retries(model_id, contents, config,
                 wait_time = float(retry_match.group(1)) + 1.0 if retry_match else delays[attempt]
 
                 tprint(f"\tGemini Pro minute limit hit. Waiting {wait_time:.1f}s...")
-                time.sleep(wait_time)
+                if stop_requested.wait(wait_time):
+                    raise CorrectionStopRequested
                 continue # Retry same model
 
             # Immediately fallback to Flash without waiting if it's a Pro quota error
@@ -251,7 +305,8 @@ def call_gemini_with_retries(model_id, contents, config,
 
             if attempt < 2:
                 tprint(f"\tGemini API failure: {e}. Retrying in {delays[attempt]} seconds...")
-                time.sleep(delays[attempt])
+                if stop_requested.wait(delays[attempt]):
+                    raise CorrectionStopRequested
             else:
                 tprint(f"\tGemini API failure: {e}. Maximum retries reached.")
                 raise
@@ -359,6 +414,8 @@ 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 CorrectionStopRequested:
+            raise
         except Exception:  # noqa: BLE001 - invalid auxiliary model response
             add_labels = []
 
@@ -417,6 +474,9 @@ def process_single_task(task_tuple, precomputed_response=None):
         total_height = group_data[-1][2]
         use_flash = n >= 4 or total_height <= 500
 
+        if precomputed_response is None and stop_requested.is_set():
+            raise CorrectionStopRequested
+
         # Only apply limits and counts if we are making a live call
         if precomputed_response is None:
             if not use_flash:
@@ -437,9 +497,11 @@ def process_single_task(task_tuple, precomputed_response=None):
             model_to_use = MODEL_ID_flash if use_flash else MODEL_ID_pro
 
             if precomputed_response:
-                tprint(f"Using batched response for: {label} {group_name}")
+                tprint(f"Using saved response for: {label} {group_name}")
                 full_response_text = precomputed_response
             else:
+                if stop_requested.is_set():
+                    raise CorrectionStopRequested
                 tprint(f"Asking Gemini {'Flash' if use_flash else 'Pro  '}: {label} {group_name}")
                 full_response_text = call_gemini_with_retries(model_to_use, contents, config)
 
@@ -478,8 +540,28 @@ def process_single_task(task_tuple, precomputed_response=None):
                 if res["error"] != "":
                     tprint("\tError :", res["error"], "for Copie", pid, group_name)
 
-                if can_spawn_tasks and res.get("error") in ["wrong-label", "additional-answer"]:
-                    new_tasks.extend(handle_label_errors(pid, label, res, pdf_path))
+                if can_spawn_tasks and res.get("error") in [
+                    "wrong-label",
+                    "additional-answer",
+                ]:
+                    if stop_requested.is_set():
+                        with io_lock:
+                            pending_responses[file_path] = json.dumps(json_data)
+                            atomic_write_json(
+                                pending_responses_path, pending_responses
+                            )
+                        raise CorrectionStopRequested
+                    try:
+                        new_tasks.extend(
+                            handle_label_errors(pid, label, res, pdf_path)
+                        )
+                    except CorrectionStopRequested:
+                        with io_lock:
+                            pending_responses[file_path] = json.dumps(json_data)
+                            atomic_write_json(
+                                pending_responses_path, pending_responses
+                            )
+                        raise
                     # Si "wrong-label" a déplacé le fichier courant vers _old
                     if res.get("error", "").startswith("wrg-lbl-moved-to:"):
                         current_suffix = "_old"
@@ -520,6 +602,8 @@ def process_single_task(task_tuple, precomputed_response=None):
                 if needs_correction:
                     tprint(f"\tBox anomalies detected for Copie {pid} {group_name}. \n\tRequesting isolated correction from Gemini Flash...")
                     try:
+                        if stop_requested.is_set():
+                            raise CorrectionStopRequested
                         # Pensez à passer pdf_path à la fonction modifiée !
                         res["feedback"] = correct_boxes_with_gemini(
                             pid, label, pdf_path, res["feedback"],
@@ -542,7 +626,12 @@ def process_single_task(task_tuple, precomputed_response=None):
                 # To track progress
                 completed_tasks.append((file_path, label))
                 atomic_write_json(progress_path, completed_tasks)
+                if file_path in pending_responses:
+                    del pending_responses[file_path]
+                    atomic_write_json(pending_responses_path, pending_responses)
 
+        except CorrectionStopRequested:
+            raise
         except json.JSONDecodeError:
             tprint(f"Error decoding JSON for {file_path}", file=sys.stderr)
             with io_lock:
@@ -836,36 +925,82 @@ def run_configured(args: argparse.Namespace) -> ExitCode:
         else:
             print(f"Warning: Batch results file {batch_results_path} not found.", file=sys.stderr)
 
+    report_live_progress = not any(
+        (args.batch, args.batch_from, args.deal_with_batched, args.refaire)
+    )
+    progress_total = len(tasks)
+    progress_completed = max(0, progress_total - len(tasks_to_process))
+    if report_live_progress:
+        report_group_progress(progress_completed, progress_total)
+
     made_progress = True
     while tasks_to_process or made_progress:
         if tasks_to_process:
             print(f"Starting processing on {len(tasks_to_process)} tasks with {NB_THREADS} threads...")
             with concurrent.futures.ThreadPoolExecutor(max_workers=NB_THREADS) as executor:
+                waiting_tasks = list(tasks_to_process)
                 futures = {}
-                for task in tasks_to_process:
-                    file_path = task[0]
-                    precomp = batched_responses.get(file_path)
-                    futures[executor.submit(process_single_task, task, precomp)] = task
 
-                for future in concurrent.futures.as_completed(futures):
-                    try:
-                        new_generated_tasks = future.result()
-                        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:  # 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]))
+                def submit_available_tasks() -> None:
+                    while (
+                        waiting_tasks
+                        and len(futures) < NB_THREADS
+                        and not stop_requested.is_set()
+                    ):
+                        task = waiting_tasks.pop(0)
+                        file_path = task[0]
+                        precomp = pending_responses.get(
+                            file_path, batched_responses.get(file_path)
+                        )
+                        futures[
+                            executor.submit(process_single_task, task, precomp)
+                        ] = task
+
+                submit_available_tasks()
+                while futures:
+                    completed_futures, _pending = concurrent.futures.wait(
+                        tuple(futures),
+                        return_when=concurrent.futures.FIRST_COMPLETED,
+                    )
+                    for future in completed_futures:
+                        failed_task = futures.pop(future)
+                        task_completed = False
+                        try:
+                            new_generated_tasks = future.result()
+                            task_completed = True
+                            if new_generated_tasks and not stop_requested.is_set():
+                                if report_live_progress:
+                                    progress_total += len(new_generated_tasks)
+                                waiting_tasks.extend(new_generated_tasks)
+                        except CorrectionStopRequested:
+                            pass
+                        except Exception as e:  # noqa: BLE001 - future boundary
+                            print(
+                                f"Exception during task execution: {e}",
+                                file=sys.stderr,
+                            )
+                            with io_lock:
+                                errors_summary.append((str(e), failed_task[0]))
+                        if report_live_progress and task_completed:
+                            progress_completed += 1
+                            report_group_progress(
+                                progress_completed, progress_total
+                            )
+                    submit_available_tasks()
 
             tasks_to_process = [] # Vider la liste une fois traitée
 
         # Après avoir traité toutes les tâches actuelles (live ou batched),
         # on tente de débloquer les mouvements qui étaient en attente
+        if stop_requested.is_set():
+            break
+
         delayed_tasks = resolve_delayed_moves()
         if delayed_tasks:
             print(f"Resolved {len(delayed_tasks)} delayed moves! Running executor for new tasks...")
+            if report_live_progress:
+                progress_total += len(delayed_tasks)
+                report_group_progress(progress_completed, progress_total)
             tasks_to_process.extend(delayed_tasks)
             made_progress = True
         else:
@@ -907,6 +1042,12 @@ def run_configured(args: argparse.Namespace) -> ExitCode:
             print(err, file=sys.stderr)
             escaped_path = shlex.quote(str(file))
             print(f"Run :    python -m copienator correct {escaped_path}")
+    if stop_requested.is_set():
+        print(
+            "[Interruption] Appels en cours terminés ; résultats disponibles sauvegardés.",
+            flush=True,
+        )
+        return ExitCode.INTERRUPTED
     return ExitCode.PARTIAL if errors_summary else ExitCode.SUCCESS
 
 
@@ -931,9 +1072,17 @@ def run(
     configure_runtime(workspace, discovered, args, api_client=api_client)
     if not discovered and not args.refaire:
         return ExitCode.PARTIAL
+    previous_handlers = {}
+    for signal_name in ("SIGINT", "SIGBREAK"):
+        interrupt_signal = getattr(signal, signal_name, None)
+        if interrupt_signal is not None:
+            previous_handlers[interrupt_signal] = signal.getsignal(interrupt_signal)
+            signal.signal(interrupt_signal, request_graceful_stop)
     try:
         status = run_configured(args)
     finally:
+        for interrupt_signal, previous_handler in previous_handlers.items():
+            signal.signal(interrupt_signal, previous_handler)
         for thread_id in list(thread_logs):
             flush_thread_log(thread_id)
     if warnings and status == ExitCode.SUCCESS:
@@ -983,6 +1132,12 @@ def main(argv: Sequence[str] | None = None) -> int:
     parser = build_parser()
 
     def handle(args: argparse.Namespace) -> ExitCode:
+        if args.refaire and args.overwrite:
+            raise CliError(
+                "--overwrite cannot be used with --refaire; --refaire already "
+                "replaces the corrections selected in refaire.json",
+                ExitCode.INVALID_ARGUMENTS,
+            )
         workspace, target = workspace_from_target(args)
         targets = [target]
         for additional in args.additional_targets:

copienator/commands/cutleft.py

@@ -28,6 +28,8 @@ from copienator.copy_errors import clear_copy_error, mark_copy_error, marked_cop
 DELIMITER_WIDTH = 5
 DELIMITER_COLOR = (0, 0, 0)
 OUTPUT_SIZE = (1800, 1000)
+CROP_SHIFT_STEP = 50
+CROP_WIDTH_STEP = 50
 pdf_cache_lock = threading.Lock()
 
 
@@ -80,6 +82,7 @@ def process_single_pdf(
     pdf_path: Path,
     shift_offset: int = 0,
     max_per_file: int = 5,
+    width_offset: int = 0,
 ) -> tuple[Image.Image, list[Image.Image], dict[str, object]] | None:
     """Convert one PDF into a preview, full-resolution splits and metadata."""
     try:
@@ -90,7 +93,10 @@ def process_single_pdf(
                 left, right = 0, width
             else:
                 left = max(0, 100 + shift_offset)
-                right = min(width, width // 3 + 100 + shift_offset)
+                right = min(
+                    width,
+                    width // 3 + 100 + shift_offset + max(0, width_offset),
+                )
             if right > left:
                 cropped_images.append(image.crop((left, 0, right, height)))
         if not cropped_images:
@@ -166,6 +172,7 @@ class ImageReviewer:
         self.current_result = None
         self.index = 0
         self.current_shift = 0
+        self.current_width_offset = 0
         self.default_max_per_file = default_max_per_file
         self.current_max_per_file = default_max_per_file
         self.current_preview: Image.Image | None = None
@@ -184,9 +191,10 @@ class ImageReviewer:
         self.root.bind("<Return>", self.on_next)
         self.root.bind("s", self.on_skip)
         self.root.protocol("WM_DELETE_WINDOW", self.on_close)
-        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("n", lambda _event: self.on_shift(CROP_SHIFT_STEP))
+        self.root.bind("N", lambda _event: self.on_shift(2 * CROP_SHIFT_STEP))
+        self.root.bind("t", lambda _event: self.on_shift(-CROP_SHIFT_STEP))
+        self.root.bind("l", lambda _event: self.on_enlarge(CROP_WIDTH_STEP))
         self.root.bind("1", lambda _event: self.on_set_max_pages(1))
 
         Thread(target=self.prefetch_worker, daemon=True).start()
@@ -222,6 +230,7 @@ class ImageReviewer:
             return
         self.is_processing = False
         self.current_shift = 0
+        self.current_width_offset = 0
         self.current_result = None
         self.trigger_processing(self.files[self.index], self.current_shift)
 
@@ -234,7 +243,12 @@ class ImageReviewer:
 
         def worker() -> None:
             self.manual_queue.put(
-                process_single_pdf(pdf_path, shift, self.current_max_per_file)
+                process_single_pdf(
+                    pdf_path,
+                    shift,
+                    self.current_max_per_file,
+                    self.current_width_offset,
+                )
             )
 
         Thread(target=worker, daemon=True).start()
@@ -271,10 +285,12 @@ class ImageReviewer:
         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"Shift: {self.current_shift}px | "
+                f"Extra width: {self.current_width_offset}px\n"
+                f"Files: {schema['number_of_files']} | "
                 f"Cols: {schema['columns_per_file']}\n"
                 "Enter: Save and next | s: flag error and skip | n: +50 | N: +100 | t: -50 | "
-                "1: use single column"
+                "l: widen by 50 | 1: use full pages"
             ),
             fg="black",
         )
@@ -286,6 +302,13 @@ class ImageReviewer:
         print(f"Applying shift: {self.current_shift}")
         self.trigger_processing(self.files[self.index], self.current_shift)
 
+    def on_enlarge(self, amount: int) -> None:
+        if self.is_processing:
+            return
+        self.current_width_offset += amount
+        print(f"Applying extra width: {self.current_width_offset}")
+        self.trigger_processing(self.files[self.index], self.current_shift)
+
     def on_next(self, _event: object) -> None:
         if self.is_processing or self.current_result is None:
             return
@@ -317,6 +340,7 @@ class ImageReviewer:
     def _advance(self) -> None:
         self.index += 1
         self.current_shift = 0
+        self.current_width_offset = 0
         self.current_max_per_file = self.default_max_per_file
         self.load_current_image()
 

copienator/commands/plotting.py

@@ -1,7 +1,9 @@
 from __future__ import annotations
 
 import argparse
+import copy
 import queue
+import re
 import threading
 import tkinter as tk
 from collections.abc import Sequence
@@ -24,6 +26,9 @@ from copienator.utils import natural_key, read_all_labels
 
 # --- Configuration & Globals ---
 padding = 60
+MISSING_LABEL_COLOR = "orange"
+COMMON_MISSING_LABEL_COLOR = "#403a00"
+COMMON_MISSING_THRESHOLD = 0.66
 
 try:
     font = ImageFont.truetype("DejaVuSans.ttf", size=30)
@@ -81,13 +86,81 @@ def normalized_labels(entries):
         if str(value["label"]) != "_"
     ]
 
-def prepare_image(image_path: str, bounding_boxes, all_labels, nb_pages, last_label_index):
+
+def frequently_missing_labels(
+    copies_dir: Path,
+    all_labels: list[str],
+    threshold: float = COMMON_MISSING_THRESHOLD,
+) -> set[str]:
+    """Return labels absent from at least ``threshold`` of detected copies."""
+    labels_by_copy: dict[str, set[str]] = {}
+    for json_path in copies_dir.glob("*.json"):
+        match = re.fullmatch(r"(.+)_\d+", json_path.stem)
+        if match is None:
+            continue
+        try:
+            data = read_json(json_path)
+            entries = data["list"]
+            if not isinstance(entries, list):
+                continue
+            present = set(normalized_labels(entries))
+        except (OSError, KeyError, TypeError, ValueError):
+            continue
+        labels_by_copy.setdefault(match.group(1), set()).update(present)
+
+    copy_count = len(labels_by_copy)
+    if copy_count == 0:
+        return set()
+    return {
+        label
+        for label in all_labels
+        if sum(label not in present for present in labels_by_copy.values())
+        >= threshold * copy_count
+    }
+
+
+def label_color(
+    label: str | None,
+    all_labels: list[str],
+    last_label_index: int,
+    common_missing: set[str],
+) -> tuple[str, int]:
+    """Choose a label color and return the updated chronological index."""
+    color = "black"
+    if not label or label not in all_labels:
+        return color, last_label_index
+
+    current_index = all_labels.index(label)
+    if current_index < last_label_index or (
+        last_label_index == -1 and current_index != 0
+    ):
+        color = "red"
+    elif current_index > last_label_index + 1:
+        only_previous_is_missing = current_index == last_label_index + 2
+        previous_label = all_labels[current_index - 1]
+        color = (
+            COMMON_MISSING_LABEL_COLOR
+            if only_previous_is_missing and previous_label in common_missing
+            else MISSING_LABEL_COLOR
+        )
+    return color, current_index
+
+
+def prepare_image(
+    image_path: str,
+    bounding_boxes,
+    all_labels,
+    nb_pages,
+    last_label_index,
+    common_missing: set[str] | None = None,
+):
     im = Image.open(image_path)
     im.load()
     width, height = im.size
     new_im = Image.new(im.mode, (width + padding, height), "white")
     new_im.paste(im, (0, 0))
     draw = ImageDraw.Draw(new_im)
+    common_missing = common_missing or set()
 
     for bbox in sort_bounding_boxes(bounding_boxes, nb_pages):
         raw_y_min = int(bbox["box_2d"][0] * height / 1000)
@@ -99,15 +172,13 @@ def prepare_image(image_path: str, bounding_boxes, all_labels, nb_pages, last_la
         abs_y_max = min(height, raw_y_max + 10)
         abs_x_max = min(width, raw_x_max + 10)
 
-        color = "black"
         label = bbox.get("label")
-        if label and label in all_labels:
-            current_index = all_labels.index(label)
-            if current_index < last_label_index or (last_label_index == -1 and current_index != 0):
-                color = "red"
-            elif current_index > last_label_index + 1:
-                color = "orange"
-            last_label_index = current_index
+        color, last_label_index = label_color(
+            label,
+            all_labels,
+            last_label_index,
+            common_missing,
+        )
 
         draw.rectangle(((abs_x_min, abs_y_min), (abs_x_max, abs_y_max)), outline=color, width=4)
         if label:
@@ -126,6 +197,7 @@ def _worker_items(base_dir, files_to_process, all_labels, output_queue):
     """
     previous_copie = None
     last_label_index = None
+    common_missing = frequently_missing_labels(base_dir / "Copies", all_labels)
     for img_path in files_to_process:
         json_path = base_dir / "Copies" / f"{img_path.stem}.json"
         copie_part = int(img_path.stem[-2:])
@@ -158,7 +230,14 @@ def _worker_items(base_dir, files_to_process, all_labels, output_queue):
             try:
                 print(f"Buffering {img_path.name}...")
                 (pil_image, last_label_index) = \
-                    prepare_image(str(img_path), bb_list, all_labels, nb_pages, last_label_index)
+                    prepare_image(
+                        str(img_path),
+                        bb_list,
+                        all_labels,
+                        nb_pages,
+                        last_label_index,
+                        common_missing,
+                    )
                 error_msg = None
 
             except Exception as e:  # noqa: BLE001 - keep the item editable in the GUI
@@ -271,7 +350,6 @@ class ImageViewer:
                     # Start new batch
                     self.active_copie_name = metadata["copie"]
                     self.accumulated_results = {"name": metadata["name"], "list": []}
-                    self.history.clear()
 
                 self.display_image(pil_image, json_path, metadata)
             except queue.Empty:
@@ -293,16 +371,23 @@ class ImageViewer:
     def on_previous(self, event):
         if self.is_viewing and self.history:
             print("Going back to previous image...")
-            prev_pil, prev_json, prev_meta, num_added = self.history.pop()
-
-            # Undo the accumulation to prevent duplicates when we hit Enter again
-            if self.accumulated_results and num_added > 0:
-                self.accumulated_results["list"] = self.accumulated_results["list"][:-num_added]
+            (
+                prev_pil,
+                prev_json,
+                prev_meta,
+                previous_copie_name,
+                previous_results,
+            ) = self.history.pop()
 
             # Push current image to the forward stack so we don't lose it
             self.forward_stack.append((self.current_pil_image,
                                        self.current_json_path, self.current_meta))
 
+            # Restore the aggregation exactly as it was before validating the
+            # previous image. This also makes navigation across copies safe.
+            self.active_copie_name = previous_copie_name
+            self.accumulated_results = previous_results
+
             # Display the previous image immediately
             self.display_image(prev_pil, prev_json, prev_meta)
     def display_image(self, pil_image, json_path, metadata):
@@ -329,7 +414,8 @@ class ImageViewer:
     def on_enter(self, event):
         if self.is_viewing:
             print(f"Committing data for {self.current_json_path.name}...")
-            num_added = 0 # ADD THIS LINE
+            previous_copie_name = self.active_copie_name
+            previous_results = copy.deepcopy(self.accumulated_results)
 
             try:
                 current_data = read_json(self.current_json_path)
@@ -352,8 +438,6 @@ class ImageViewer:
                     print(msg)
                     messagebox.showerror("Label Error", msg)
                     return
-                num_added = len(converted_items)
-
                 # Add to accumulator
                 if self.accumulated_results:
                     self.accumulated_results["list"].extend(converted_items)
@@ -368,8 +452,15 @@ class ImageViewer:
                 messagebox.showerror("JSON Error", msg)
                 return  # Abort advancement
 
-            self.history.append((self.current_pil_image, self.current_json_path,
-                                 self.current_meta, num_added))
+            self.history.append(
+                (
+                    self.current_pil_image,
+                    self.current_json_path,
+                    self.current_meta,
+                    previous_copie_name,
+                    previous_results,
+                )
+            )
 
             # Advance UI
             self.is_viewing = False

copienator/commands/splitting_int.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import argparse
+import math
 import shutil
 import tempfile
 from collections import defaultdict
@@ -23,6 +24,7 @@ from copienator import (
 from copienator.filesystem import staged_directory
 
 SQUARE = 1000 // 38
+ANSWER_TOP_PADDING_POINTS = 4 * 72 / 25.4
 Coordinate = tuple[str, int, int, int, int, int]
 ParsedCoordinate = tuple[str, str, int, int, int, int, int]
 
@@ -171,7 +173,15 @@ def _render_split_outputs(
 
                 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
+                    y0 = (
+                        math.floor(max(
+                            0,
+                            (y_start / 1000) * page.rect.height
+                            - ANSWER_TOP_PADDING_POINTS,
+                        ))
+                        if page_number == start_page
+                        else 0
+                    )
                     y1 = (end_y / 1000) * page.rect.height if page_number == end_page else page.rect.height
                     clip = pymupdf.Rect(
                         fraction_x0 * page.rect.width, y0,

copienator/crop_blank_margins.py

@@ -110,7 +110,8 @@ def main() -> None:
               flush=True)
         (args.output/'report.json').write_text(json.dumps(rows, indent=2)+'\n')
     with (args.output/'report.csv').open('w') as stream:
-        writer = csv.DictWriter(stream, fieldnames=list(rows[0]))
+        fieldnames = list(dict.fromkeys(key for row in rows for key in row))
+        writer = csv.DictWriter(stream, fieldnames=fieldnames)
         writer.writeheader()
         writer.writerows(rows)
     cards = []

copienator/ink_detection.py

@@ -65,13 +65,26 @@ def _large_blank_ink(gray: np.ndarray, clean: np.ndarray, dpi: float):
     n, labels, stats, centers = cv2.connectedComponentsWithStats(
         (residual > 35).astype(np.uint8), 8)
     keep = np.zeros(n, bool)
+    xx = stats[1:, 0]
     ww, hh, area = stats[1:, 2], stats[1:, 3], stats[1:, 4]
-    keep[1:] = ((area >= .8*px*px) & (np.minimum(ww, hh) >= .6*px)
-                & (area / (ww*hh) > .3)
-                & (np.maximum(ww, hh) < 4*np.minimum(ww, hh)))
+    density = area / (ww*hh)
+    shortest, longest = np.minimum(ww, hh), np.maximum(ww, hh)
+    compact_ink = ((area >= .8*px*px) & (shortest >= .6*px)
+                   & (density > .3) & (longest < 4*shortest))
+    # Ruling suppression can fragment faint pencil handwriting into sparse,
+    # elongated components. Admit those moderately more readily in the central
+    # 80% of the sheet. The outermost 9% deliberately uses a stricter filter:
+    # punched holes, torn binding edges, and page numbers usually occur there.
+    center_x = xx + ww/2
+    central = (center_x > width*.1) & (center_x < width*.9)
+    central_ink = (central & (area >= .55*px*px) & (shortest >= .45*px)
+                   & (density > .18) & (longest < 6*shortest))
+    outer = (center_x < width*.09) | (center_x > width*.91)
+    outer_ink = (outer & (area >= 1.2*px*px) & (shortest >= .8*px)
+                 & (density >= .4) & (longest < 3*shortest))
+    keep[1:] = np.where(outer, outer_ink, compact_ink | central_ink)
     # Use side columns as a prior, then require repeated size and alignment. An
     # isolated note in the same column remains eligible to protect the margin.
-    xx = stats[1:, 0]
     candidates = np.flatnonzero(
         ((xx < 15*px) | (xx+ww > width-15*px))
         & (ww > px) & (ww < 9*px) & (hh > px) & (hh < 12*px)) + 1

copienator/workspace.py

@@ -109,6 +109,10 @@ class EvaluationWorkspace:
     def correction_progress_file(self) -> Path:
         return self.root / "correction_progress.json"
 
+    @property
+    def correction_pending_responses_file(self) -> Path:
+        return self.root / "correction_pending_responses.json"
+
     @property
     def batch_jobs_file(self) -> Path:
         return self.root / "batch_jobs.json"

copienator_gui/app.py

@@ -22,6 +22,7 @@ from copienator.platform import (
 
 from .diagnostics import collect_diagnostics
 from .batch_monitor import BatchMonitor
+from .giving_names import GivingNamesPanel, find_name_issues
 from .notifications import notify_desktop
 from .manual_resolution import ManualResolutionPanel
 from .refaire import SECTION as REFAIRE_SECTION
@@ -58,6 +59,9 @@ STATUS_LABELS = {
 }
 
 DEFAULT_HTTPS_PROXY = "http://10.0.0.1:3128"
+CORRECTION_PROGRESS_RE = re.compile(
+    r"\[Progression correction\] Groupes traités : (\d+)/(\d+)"
+)
 PERSONAL_INTERRO_SOURCE = Path("/home/sebastien/Prépa/Staging/Interro")
 ANNOTATION_DIRECTORIES = ("BGnot", "Bnot", "Anot")
 ANNOTATION_VARIANT_DIRECTORIES = {
@@ -65,6 +69,10 @@ ANNOTATION_VARIANT_DIRECTORIES = {
     "checks": "Bnot",
     "grouped": "BGnot",
 }
+ANNOTATION_VARIANT_READERS = {
+    "checks": "standard",
+    "grouped": "grouped",
+}
 
 
 def get_personal_interro_files(
@@ -259,8 +267,11 @@ class CopienatorApp(tk.Tk):
         self.copy_paths: dict[str, Path] = {}
         self._rendering = False
         self.refaire_panel: RefaireSelection | None = None
+        self.giving_names_panel: GivingNamesPanel | None = None
+        self.complete_giving_names_button: ttk.Button | None = None
         self.pending_refaire_commands: list[list[str]] = []
         self.refaire_command_index = 0
+        self._correction_progress_buffer = ""
 
         self.title("Copienator — assistant de correction")
         self.geometry("1180x820")
@@ -752,7 +763,15 @@ class CopienatorApp(tk.Tk):
             return
         self._rendering = True
         redo = step.section == REFAIRE_SECTION
-        expanded_form = redo or step.id in {"page_splitter", "crop_blank_margins", "cutleft", "labels", "manual_resolution"}
+        expanded_form = redo or step.id in {
+            "page_splitter",
+            "crop_blank_margins",
+            "cutleft",
+            "labels",
+            "plotting",
+            "manual_resolution",
+            "giving_names",
+        }
         self.console.configure(height=8 if expanded_form else 14)
         self.rowconfigure(1, weight=4 if expanded_form else 3)
         self.rowconfigure(2, weight=1 if expanded_form else 2)
@@ -763,6 +782,8 @@ class CopienatorApp(tk.Tk):
             child.destroy()
         self.arg_vars.clear()
         self.refaire_panel = None
+        self.giving_names_panel = None
+        self.complete_giving_names_button = None
         entry = self.state_store.step(step.id) if self.state_store.evaluation else {}
         saved_variant = entry.get("variant", step.variants[0].id)
         if saved_variant not in {variant.id for variant in step.variants}:
@@ -909,6 +930,31 @@ class CopienatorApp(tk.Tk):
                 )
                 row += 1
             return row
+        if step.id == "giving_names" and self.evaluation:
+            self.giving_names_panel = GivingNamesPanel(
+                self.form, self.evaluation
+            )
+            self.giving_names_panel.grid(
+                row=row,
+                column=0,
+                columnspan=2,
+                sticky="ew",
+                pady=(0, 8),
+            )
+            row += 1
+            self.complete_giving_names_button = ttk.Button(
+                self.form,
+                text="Marquer terminée",
+                command=self._complete_giving_names,
+            )
+            self.complete_giving_names_button.grid(
+                row=row,
+                column=0,
+                columnspan=2,
+                sticky="w",
+                pady=(0, 8),
+            )
+            row += 1
         if step.section == REFAIRE_SECTION:
             evaluation = self.evaluation
             if step.id in {"refaire_selection", "refaire_correct"}:
@@ -935,7 +981,14 @@ class CopienatorApp(tk.Tk):
             return row + 1
         if step.id in {"review_persp", "correction"}:
             row = self._correction_folder_buttons(row)
-        if step.id in {"page_splitter", "crop_blank_margins", "cutleft", "labels"}:
+        copy_selection_steps = {
+            "page_splitter",
+            "crop_blank_margins",
+            "cutleft",
+            "labels",
+            "plotting",
+        }
+        if step.id in copy_selection_steps:
             evaluation = self.evaluation
             paths = copy_pdf_paths(evaluation) if evaluation else []
             self.copy_paths = {path.name: path for path in paths}
@@ -966,7 +1019,7 @@ class CopienatorApp(tk.Tk):
                 command=self._open_selected_copy,
                 state="normal" if names else "disabled",
             ).grid(row=1, column=1, padx=(6, 0), pady=(7, 0))
-            if step.id in {"page_splitter", "crop_blank_margins", "cutleft", "labels"}:
+            if step.id in copy_selection_steps:
                 actions = ttk.Frame(copies)
                 actions.grid(row=2, column=0, columnspan=2, sticky="w", pady=(7, 0))
                 label = "Refaire la copie sélectionnée" if step.id == "page_splitter" else "Cibler la copie sélectionnée"
@@ -978,6 +1031,8 @@ class CopienatorApp(tk.Tk):
                     if step.id == "page_splitter"
                     else "Seule la copie ciblée sera analysée par Gemini. "
                     if step.id == "labels"
+                    else "Seule la copie ciblée sera vérifiée. "
+                    if step.id == "plotting"
                     else "Seule la copie ciblée sera traitée. "
                 )
                 ttk.Label(copies, text=help_text + "Vérifiez la commande puis cliquez sur Exécuter.",
@@ -1352,8 +1407,7 @@ class CopienatorApp(tk.Tk):
                 return
         if step.is_manual:
             self._mark_step("success")
-            if step.section == REFAIRE_SECTION:
-                self._move_selection_from(step.id, 1)
+            self._move_selection_from(step.id, 1)
             return
         if os.name == "nt" and step.id != "statement":
             labels_path = evaluation / "labels"
@@ -1400,6 +1454,9 @@ class CopienatorApp(tk.Tk):
             last_run_values=run_values,
         )
         self.active_step_id = step.id
+        if step.id == "correction":
+            self._correction_progress_buffer = ""
+            self.info_var.set("Correction : préparation des groupes…")
         workspace = self.state_store.workspace
         assert workspace is not None
         log_path = workspace.log_path(step.id)
@@ -1463,6 +1520,33 @@ class CopienatorApp(tk.Tk):
         self._populate_tree()
         self.info_var.set("Renommage validé — aucun fichier modifié.")
 
+    def _complete_giving_names(self) -> None:
+        if (
+            not self.current_step
+            or self.current_step.id != "giving_names"
+            or not self.evaluation
+            or self.active_step_id
+            or self.runner.running
+        ):
+            return
+        return_dir = EvaluationWorkspace(self.evaluation).return_dir
+        if not return_dir.is_dir():
+            messagebox.showerror(
+                "A Rendre absent",
+                "Exécutez d’abord l’attribution des noms.",
+            )
+            return
+        issues = find_name_issues(self.evaluation)
+        if issues and not messagebox.askyesno(
+            "Noms encore à vérifier",
+            f"{len(issues)} copie(s) ont encore un nom inconnu ou dupliqué. "
+            "Marquer quand même l’étape comme terminée ?",
+            icon="warning",
+        ):
+            return
+        self._mark_step("success")
+        self._move_selection_from("giving_names", 1)
+
     def _skip_step(self) -> None:
         if self.current_step and self.current_step.optional:
             step_id = self.current_step.id
@@ -1477,7 +1561,9 @@ class CopienatorApp(tk.Tk):
             except queue.Empty:
                 break
             if event == "output":
-                self._append_console(str(payload))
+                output = str(payload)
+                self._track_correction_progress(output)
+                self._append_console(output)
             elif event == "input_echo":
                 self._append_console(f"> {payload}")
             elif event == "runner_error":
@@ -1487,6 +1573,24 @@ class CopienatorApp(tk.Tk):
                 self._finish_process(int(return_code), bool(interrupted))
         self.after(60, self._poll_runner)
 
+    def _track_correction_progress(self, output: str) -> None:
+        if self.active_step_id != "correction":
+            return
+        self._correction_progress_buffer = (
+            self._correction_progress_buffer + output
+        )[-512:]
+        matches = list(CORRECTION_PROGRESS_RE.finditer(
+            self._correction_progress_buffer
+        ))
+        if not matches:
+            return
+        completed, total = (int(value) for value in matches[-1].groups())
+        percent = round(100 * completed / total) if total else 100
+        self.info_var.set(
+            f"Correction : {completed} groupe(s) traité(s) sur {total} "
+            f"({percent} %)."
+        )
+
     def _batch_watch_status_changed(self, message: str) -> None:
         self.batch_watch_status.set(message)
         self._update_controls()
@@ -1565,7 +1669,25 @@ class CopienatorApp(tk.Tk):
                 "été conservés.",
             )
             return
-        self.state_store.update_step(step_id, status=status, return_code=return_code)
+        stored_status = (
+            "detected"
+            if step_id == "giving_names" and status == "success"
+            else status
+        )
+        self.state_store.update_step(
+            step_id, status=stored_status, return_code=return_code
+        )
+        if step_id == "annotation" and status == "success":
+            generation_variant = self.state_store.step("annotation").get(
+                "last_run_variant"
+            )
+            reader_variant = ANNOTATION_VARIANT_READERS.get(
+                str(generation_variant)
+            )
+            if reader_variant:
+                self.state_store.update_step(
+                    "read_annotations", variant=reader_variant
+                )
         self.state_store.add_history(
             {
                 "step": step_id,
@@ -1580,12 +1702,27 @@ class CopienatorApp(tk.Tk):
         self.active_step_id = None
         self._populate_tree()
         self._update_controls()
+        if step_id == "giving_names" and status == "success":
+            if self.giving_names_panel:
+                self.giving_names_panel.reload()
+            self.info_var.set(
+                "A Rendre a été préparé. Vérifiez les noms, puis cliquez "
+                "sur « Marquer terminée »."
+            )
+            return
         if status == "success":
             if step_id == "batch_status" and self.batch_monitor.active:
                 self.batch_monitor.stop()
                 self._batch_results_ready()
                 return
-            self.after_idle(lambda completed_id=step_id: self._move_selection_from(completed_id, 1))
+            if step_id == "manual_resolution":
+                self.after_idle(self._select_refaire_correction)
+            else:
+                self.after_idle(
+                    lambda completed_id=step_id: self._move_selection_from(
+                        completed_id, 1
+                    )
+                )
 
     def _send_input(self) -> None:
         text = self.stdin_var.get()
@@ -1597,12 +1734,21 @@ class CopienatorApp(tk.Tk):
         self.stdin_var.set("")
 
     def _interrupt(self) -> None:
-        if self.runner.running and messagebox.askyesno(
-            "Interrompre", "Envoyer une interruption au script en cours ?"
-        ):
+        graceful_correction = self.active_step_id == "correction"
+        question = (
+            "Arrêter de lancer de nouveaux appels Gemini ? Les appels déjà en "
+            "cours pourront se terminer et leurs résultats seront sauvegardés."
+            if graceful_correction
+            else "Envoyer une interruption au script en cours ?"
+        )
+        if self.runner.running and messagebox.askyesno("Interrompre", question):
             try:
                 self.runner.interrupt()
-                self.info_var.set("Interruption demandée. Utilisez « Forcer l’arrêt » si le script ne répond pas.")
+                self.info_var.set(
+                    "Arrêt demandé : attente des appels Gemini en cours…"
+                    if graceful_correction
+                    else "Interruption demandée. Utilisez « Forcer l’arrêt » si le script ne répond pas."
+                )
             except OSError as exc:
                 messagebox.showerror("Interruption impossible", str(exc))
 
@@ -1645,6 +1791,10 @@ class CopienatorApp(tk.Tk):
         self.force_button.configure(state="normal" if running else "disabled")
         self.send_button.configure(state="normal" if running else "disabled")
         self.stdin_entry.configure(state="normal" if running else "disabled")
+        if self.complete_giving_names_button is not None:
+            self.complete_giving_names_button.configure(
+                state="disabled" if running else "normal"
+            )
 
     def _move_selection(self, delta: int) -> None:
         if not self.current_step:
@@ -1663,6 +1813,14 @@ class CopienatorApp(tk.Tk):
         self.tree.selection_set(ids[target])
         self.tree.see(ids[target])
 
+    def _select_refaire_correction(self) -> None:
+        """Return from manual conflict resolution to the main redo correction."""
+        if self.runner.running or self.active_step_id or not self.state_store.evaluation:
+            return
+        self.state_store.update_step("correction", variant="refaire")
+        self.tree.selection_set("correction")
+        self.tree.see("correction")
+
     def _on_close(self) -> None:
         if self.runner.running:
             if not messagebox.askyesno(

copienator_gui/workflow.py

@@ -259,8 +259,11 @@ def build_workflow(show_personal_steps: bool) -> list[StepDefinition]:
         StepDefinition(
             "cutleft",
             "Prétraitement des copies",
-            "Découper la marge des labels",
-            "Produit les images de la partie gauche des copies. Une copie précise peut être ciblée.",
+            "Découper une partie à gauche pour détection des labels",
+            "Produit les images de la partie gauche des copies. Dans la fenêtre de découpe : "
+            "n décale la zone de 50 px vers la droite, N de 100 px, t de 50 px vers la gauche, "
+            "l l’élargit de 50 px, 1 utilise les pages entières, Entrée valide et s signale une erreur. "
+            "Une copie précise peut être ciblée.",
             (python("default", "Découpe", "crop-labels"),),
             arguments=(
                 arg_target(),
@@ -309,7 +312,8 @@ def build_workflow(show_personal_steps: bool) -> list[StepDefinition]:
             "plotting",
             "Labels et regroupement",
             "Vérifier visuellement les labels",
-            "Ouvre la fenêtre de vérification. La cible peut être toute l’évaluation ou une copie.",
+            "Ouvre la fenêtre de vérification. La cible peut être toute l’évaluation ou une copie. "
+            "Le raccourci p revient à l’image précédente, y compris dans la copie précédente.",
             (python("default", "Vérification", "review-labels"),),
             arguments=(arg_target(),),
             requires=("labels", "Cutleft"),
@@ -364,7 +368,10 @@ def build_workflow(show_personal_steps: bool) -> list[StepDefinition]:
             "correction",
             "Correction",
             "Lancer ou intégrer la correction",
-            "Choisir une correction immédiate, batch, hybride, une recorrection, ou l’intégration d’un batch.",
+            "Choisir une correction immédiate, batch, hybride, une recorrection, ou l’intégration d’un batch. "
+            "En correction immédiate, la barre d’état affiche le nombre de groupes traités et le total. "
+            "Interrompre bloque les nouveaux appels Gemini, attend ceux déjà en cours, sauvegarde leurs résultats puis arrête la commande. "
+            "Au prochain lancement, les validations encore nécessaires (mauvais label ou contenu supplémentaire) reprennent sans refaire la requête principale.",
             (
                 python("live", "Correction immédiate", "correct"),
                 python("batch", "Préparer toutes les requêtes batch", "correct", fixed_args=("--batch",)),
@@ -386,7 +393,7 @@ def build_workflow(show_personal_steps: bool) -> list[StepDefinition]:
                     "bool",
                     "--overwrite",
                     help="Relance les corrections demandées même lorsqu’un résultat existe déjà.",
-                    variants=("live", "batch", "hybrid", "refaire", "integrate"),
+                    variants=("live", "batch", "hybrid", "integrate"),
                 ),
                 ArgumentSpec(
                     "limit",
@@ -443,6 +450,18 @@ def build_workflow(show_personal_steps: bool) -> list[StepDefinition]:
             optional=True,
             skip_for_live_correction=True,
         ),
+        StepDefinition(
+            "manual_resolution",
+            "Correction",
+            "Résoudre les conflits manuels",
+            "Étape conditionnelle, uniquement si manual_resolutions.txt contient des conflits à traiter. "
+            "Après succès, le GUI revient à la correction avec « Recorrection depuis refaire.json » sélectionné.",
+            (python("default", "Résolution", "resolve-manual"),),
+            arguments=(arg_target("le dossier de l’évaluation"),),
+            optional=True,
+            requires=("manual_resolutions.txt", "correction.json"),
+            skip_without_manual_conflicts=True,
+        ),
         StepDefinition(
             "post_correction",
             "Correction",
@@ -452,17 +471,6 @@ def build_workflow(show_personal_steps: bool) -> list[StepDefinition]:
             arguments=(arg_target("le dossier de l’évaluation"),),
             requires=("correction.json",),
         ),
-        StepDefinition(
-            "manual_resolution",
-            "Correction",
-            "Résoudre les conflits manuels",
-            "Étape conditionnelle, uniquement si manual_resolutions.txt contient des conflits à traiter.",
-            (python("default", "Résolution", "resolve-manual"),),
-            arguments=(arg_target("le dossier de l’évaluation"),),
-            optional=True,
-            requires=("manual_resolutions.txt", "correction.json"),
-            skip_without_manual_conflicts=True,
-        ),
         StepDefinition(
             "annotation",
             "Génération des annotations",
@@ -596,7 +604,9 @@ def build_workflow(show_personal_steps: bool) -> list[StepDefinition]:
             "giving_names",
             "Finalisation",
             "Attribuer les noms et préparer A Rendre",
-            "Crée le dossier A Rendre à partir du dossier d’annotations choisi.",
+            "Crée le dossier A Rendre à partir du dossier d’annotations choisi, sans passer automatiquement à l’étape suivante. "
+            "Après l’exécution, vous pouvez renommer chaque dossier, ainsi que son fichier .jpg et son fichier .pdf, avec un autre nom ; le suffixe (id) du dossier est conservé. "
+            "Les outils affichés permettent d’identifier et corriger les noms Unknown ou attribués à plusieurs copies. Cliquez ensuite sur « Marquer terminée ».",
             (python("default", "Attribution des noms", "giving-names"),),
             arguments=(
                 arg_target("le dossier de l’évaluation"),

docs/crop_blank_margins.md

@@ -46,6 +46,11 @@ alignment in the outer 15 mm are treated as punched holes only when at least
 three span a substantial part of the page. Writing in the same side column still
 protects its margin.
 
+On confirmed ruled paper, faint sparse strokes in the central 80% of the page
+use a moderately more sensitive component filter. The outermost 9% on each side
+uses a stricter filter because punched holes, torn binding edges, and page
+numbers normally appear there.
+
 The large-blank refinement changes an edge only when it finds at least 30 mm of
 additional empty paper. A 2 mm recovery neighbourhood is applied before the
 normal padding. Apparently blank pages and pages without reliable ink seeds are

tests/test_gui_convenience.py

@@ -85,6 +85,53 @@ class GuiConvenienceTests(unittest.TestCase):
             self.app.manual_panel.editor_button.invoke()
         opened.assert_called_once_with(other_manual)
 
+    def test_successful_manual_resolution_returns_to_refaire_correction(self):
+        manual = self.evaluation / "manual_resolutions.txt"
+        manual.write_text("Copie01 A -> B|\n", encoding="utf-8")
+        self.app.state_store.update_step("manual_resolution", visited=True)
+        self.app.tree.selection_set("manual_resolution")
+        self.app.update()
+
+        self.app.active_step_id = "manual_resolution"
+        self.app._finish_process(0, False)
+        self.app.update()
+
+        self.assertEqual(self.app.current_step.id, "correction")
+        self.assertEqual(self.app.variant_var.get(), "refaire")
+        self.assertEqual(
+            self.app.state_store.step("correction")["variant"], "refaire"
+        )
+        self.assertNotIn("overwrite", self.app.arg_vars)
+        self.assertIn("--refaire", self.app._make_command())
+
+    def test_giving_names_waits_for_explicit_manual_completion(self):
+        return_dir = self.evaluation / "A Rendre" / "Student (01)"
+        return_dir.mkdir(parents=True)
+        (return_dir / "Student.jpg").write_bytes(b"jpg")
+        (return_dir / "Student.pdf").write_bytes(b"pdf")
+        self.app.tree.selection_set("giving_names")
+        self.app.update()
+
+        self.app.active_step_id = "giving_names"
+        self.app._finish_process(0, False)
+        self.app.update()
+
+        self.assertEqual(self.app.current_step.id, "giving_names")
+        self.assertEqual(
+            self.app.state_store.step("giving_names")["status"], "detected"
+        )
+        self.assertEqual(
+            self.app.complete_giving_names_button.cget("text"),
+            "Marquer terminée",
+        )
+
+        self.app.complete_giving_names_button.invoke()
+        self.app.update()
+        self.assertEqual(
+            self.app.state_store.step("giving_names")["status"], "success"
+        )
+        self.assertNotEqual(self.app.current_step.id, "giving_names")
+
     def test_batch_status_waits_until_all_jobs_are_ready(self):
         self.app.state_store.update_step("correction", variant="batch")
         self.app.state_store.update_step("batch_status", visited=True)
@@ -213,6 +260,28 @@ class GuiConvenienceTests(unittest.TestCase):
         self.assertIn("Cibler la copie sélectionnée", controls)
         self.assertIn("Cibler tout le dossier", controls)
 
+    def test_label_review_can_show_and_target_one_copy(self):
+        copies = self.evaluation / "Copies"
+        copies.mkdir()
+        for name in ("Copie01.pdf", "Copie02.pdf"):
+            (copies / name).touch()
+
+        self.app.tree.selection_set("plotting")
+        self.app.update()
+        self.app.copy_var.set("Copie02.pdf")
+
+        with patch("copienator_gui.app.open_path") as opened:
+            self.app._open_selected_copy()
+        opened.assert_called_once_with(copies / "Copie02.pdf")
+
+        self.app._target_selected_copy()
+        command = self.app._make_command()
+        self.assertEqual(Path(command[-1]), copies / "Copie02.pdf")
+        labels = self._label_texts(self.app.form)
+        self.assertTrue(
+            any("Seule la copie ciblée sera vérifiée" in text for text in labels)
+        )
+
     def test_free_form_arguments_are_shown_only_when_documented(self):
         self.app.tree.selection_set("labels")
         self.app.update()

tests/test_gui_core.py

@@ -43,6 +43,11 @@ from copienator_gui.app import (
     plotting_shortcut_lines,
     process_status,
 )
+from copienator_gui.giving_names import (
+    find_name_issues,
+    original_copy_path,
+    rename_return_copy,
+)
 from copienator_gui.diagnostics import collect_diagnostics
 from copienator_gui.runner import ProcessRunner
 from copienator_gui.state import StateStore
@@ -911,6 +916,43 @@ class StandardCliTests(unittest.TestCase):
             )
             self.assertIsNone(viewer.accumulated_results)
 
+    def test_plotting_detects_labels_missing_from_two_thirds_of_copies(self) -> None:
+        module = self.modules["plotting"]
+        with tempfile.TemporaryDirectory() as directory:
+            copies = Path(directory)
+            detected = {
+                "Copie01_01.json": ["Ex 1", "Ex 3"],
+                "Copie02_01.json": ["Ex 1", "Ex 3"],
+                "Copie03_01.json": ["Ex 1", "Ex 2", "Ex 3"],
+            }
+            for filename, labels in detected.items():
+                atomic_write_json(
+                    copies / filename,
+                    {"list": [{"label": label} for label in labels]},
+                )
+
+            missing = module.frequently_missing_labels(
+                copies,
+                ["Ex 1", "Ex 2", "Ex 3", "Ex 4"],
+            )
+
+        self.assertEqual(missing, {"Ex 2", "Ex 4"})
+
+    def test_plotting_mutes_only_a_single_commonly_missing_label(self) -> None:
+        module = self.modules["plotting"]
+        labels = ["Ex 1", "Ex 2", "Ex 3", "Ex 4"]
+
+        muted, index = module.label_color("Ex 3", labels, 0, {"Ex 2"})
+        ordinary, _ = module.label_color("Ex 3", labels, 0, set())
+        two_missing, _ = module.label_color(
+            "Ex 4", labels, 0, {"Ex 2", "Ex 3"}
+        )
+
+        self.assertEqual(muted, module.COMMON_MISSING_LABEL_COLOR)
+        self.assertEqual(index, 2)
+        self.assertEqual(ordinary, module.MISSING_LABEL_COLOR)
+        self.assertEqual(two_missing, module.MISSING_LABEL_COLOR)
+
     def test_plotting_validation_preserves_manually_edited_order(self) -> None:
         module = self.modules["plotting"]
         with tempfile.TemporaryDirectory() as directory:
@@ -931,6 +973,7 @@ class StandardCliTests(unittest.TestCase):
                 "part": 1,
             }
             viewer.valid_labels = {"Ex 1", "Ex 2"}
+            viewer.active_copie_name = "Copie01"
             viewer.accumulated_results = {"name": "Student", "list": []}
             viewer.history = []
             viewer.current_pil_image = Mock()
@@ -944,6 +987,49 @@ class StandardCliTests(unittest.TestCase):
                 ["Ex 2", "Ex 1"],
             )
 
+    def test_plotting_previous_restores_state_across_copies(self) -> None:
+        module = self.modules["plotting"]
+        viewer = module.ImageViewer.__new__(module.ImageViewer)
+        previous_image = Mock()
+        previous_json = Path("Copie01_02.json")
+        previous_meta = {"copie": "Copie01"}
+        previous_results = {
+            "name": "Student 1",
+            "list": [{"label": "Ex 1"}],
+        }
+        current_image = Mock()
+        current_json = Path("Copie02_01.json")
+        current_meta = {"copie": "Copie02"}
+        viewer.is_viewing = True
+        viewer.history = [
+            (
+                previous_image,
+                previous_json,
+                previous_meta,
+                "Copie01",
+                previous_results,
+            )
+        ]
+        viewer.forward_stack = []
+        viewer.current_pil_image = current_image
+        viewer.current_json_path = current_json
+        viewer.current_meta = current_meta
+        viewer.active_copie_name = "Copie02"
+        viewer.accumulated_results = {"name": "Student 2", "list": []}
+        viewer.display_image = Mock()
+
+        viewer.on_previous(None)
+
+        self.assertEqual(viewer.active_copie_name, "Copie01")
+        self.assertEqual(viewer.accumulated_results, previous_results)
+        self.assertEqual(
+            viewer.forward_stack,
+            [(current_image, current_json, current_meta)],
+        )
+        viewer.display_image.assert_called_once_with(
+            previous_image, previous_json, previous_meta
+        )
+
     def test_plotting_validates_plain_and_directional_labels(self) -> None:
         module = self.modules["plotting"]
         self.assertEqual(
@@ -1226,6 +1312,27 @@ class StandardCliTests(unittest.TestCase):
             self.assertEqual(module.results, {"Ex 1": []})
             self.assertEqual(module.tasks_to_process, [("new.jpg", "Ex 1")])
 
+    def test_correction_rejects_overwrite_with_refaire_without_writing(self) -> None:
+        module = self.modules["correction"]
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory) / "Exam"
+            evaluation.mkdir()
+            correction = evaluation / "correction.json"
+            original = {"Ex 1": [[{"id": "01", "result": {"feedback": []}}]]}
+            atomic_write_json(correction, original)
+
+            errors = io.StringIO()
+            with redirect_stderr(errors):
+                status = module.main(
+                    [str(evaluation), "--refaire", "--overwrite"]
+                )
+
+            self.assertEqual(status, module.ExitCode.INVALID_ARGUMENTS)
+            self.assertIn(
+                "--overwrite cannot be used with --refaire", errors.getvalue()
+            )
+            self.assertEqual(read_json(correction), original)
+
     def test_correction_reserves_unique_group_indices_concurrently(self) -> None:
         module = self.modules["correction"]
         with tempfile.TemporaryDirectory() as directory:
@@ -1247,6 +1354,251 @@ class StandardCliTests(unittest.TestCase):
             self.assertEqual(sorted(indices), list(range(3, 11)))
             self.assertEqual(len(indices), len(set(indices)))
 
+    def test_live_correction_reports_progress_for_generated_groups(self) -> None:
+        module = self.modules["correction"]
+        with tempfile.TemporaryDirectory() as directory:
+            workspace = EvaluationWorkspace(Path(directory))
+            args = module.build_parser().parse_args(
+                [str(workspace.root), "--overwrite"]
+            )
+            initial_tasks = [("group-1.jpg", "Ex 1"), ("group-2.jpg", "Ex 2")]
+            module.configure_runtime(
+                workspace,
+                initial_tasks,
+                args,
+                api_client=Mock(),
+            )
+            processed = []
+
+            def process(task, _precomputed=None):
+                processed.append(task[0])
+                if task[0] == "group-1.jpg":
+                    return [("group-3.jpg", "Ex 3", False)]
+                return []
+
+            output = io.StringIO()
+            with patch.object(
+                module, "process_single_task", side_effect=process
+            ), patch.object(
+                module, "resolve_delayed_moves", return_value=[]
+            ), redirect_stdout(output):
+                self.assertEqual(module.run_configured(args), 0)
+
+        self.assertCountEqual(
+            processed,
+            ["group-1.jpg", "group-2.jpg", "group-3.jpg"],
+        )
+        progress = [
+            line
+            for line in output.getvalue().splitlines()
+            if line.startswith("[Progression correction]")
+        ]
+        self.assertEqual(progress[0], "[Progression correction] Groupes traités : 0/2")
+        self.assertEqual(progress[-1], "[Progression correction] Groupes traités : 3/3")
+
+    def test_live_correction_stops_scheduling_and_waits_for_active_task(self) -> None:
+        module = self.modules["correction"]
+        self.addCleanup(module.stop_requested.clear)
+        with tempfile.TemporaryDirectory() as directory:
+            workspace = EvaluationWorkspace(Path(directory))
+            args = module.build_parser().parse_args(
+                [str(workspace.root), "--overwrite"]
+            )
+            module.configure_runtime(
+                workspace,
+                [("group-1.jpg", "Ex 1"), ("group-2.jpg", "Ex 2")],
+                args,
+                api_client=Mock(),
+            )
+            processed = []
+
+            def process(task, _precomputed=None):
+                processed.append(task[0])
+                module.request_graceful_stop()
+                (workspace.root / "saved-result").write_text(task[0])
+                return []
+
+            with patch.object(module, "NB_THREADS", 1), patch.object(
+                module, "process_single_task", side_effect=process
+            ), patch.object(
+                module, "resolve_delayed_moves", return_value=[]
+            ):
+                status = module.run_configured(args)
+            saved_result = (workspace.root / "saved-result").read_text()
+
+        self.assertEqual(status, module.ExitCode.INTERRUPTED)
+        self.assertEqual(processed, ["group-1.jpg"])
+        self.assertEqual(saved_result, "group-1.jpg")
+
+    def test_gemini_call_finishes_inflight_stream_but_does_not_retry(self) -> None:
+        module = self.modules["correction"]
+        self.addCleanup(module.stop_requested.clear)
+        module.stop_requested.clear()
+        api_client = Mock()
+
+        def stream(**_kwargs):
+            yield SimpleNamespace(text="first")
+            module.stop_requested.set()
+            yield SimpleNamespace(text="-second")
+
+        api_client.models.generate_content_stream.side_effect = stream
+        with patch.object(module, "client", api_client):
+            self.assertEqual(
+                module.call_gemini_with_retries("model", [], Mock()),
+                "first-second",
+            )
+            with self.assertRaises(module.CorrectionStopRequested):
+                module.call_gemini_with_retries("model", [], Mock())
+        self.assertEqual(api_client.models.generate_content_stream.call_count, 1)
+
+    def test_interrupted_live_correction_resumes_only_remaining_groups(self) -> None:
+        module = self.modules["correction"]
+        self.addCleanup(module.stop_requested.clear)
+        with tempfile.TemporaryDirectory() as directory:
+            workspace = EvaluationWorkspace(Path(directory))
+            workspace.copies_dir.mkdir()
+            tasks = []
+            for index, label in enumerate(("Ex 1", "Ex 2"), start=1):
+                group_dir = workspace.groups_dir / label
+                group_dir.mkdir(parents=True)
+                image = group_dir / f"Group_{index}.jpg"
+                image.write_bytes(b"image")
+                atomic_write_json(
+                    image.with_suffix(".json"),
+                    [["01", 0, 400, 1.0, label]],
+                )
+                tasks.append((str(image), label))
+
+            response = json.dumps(
+                [{"id": "01", "result": {"error": "", "feedback": []}}]
+            )
+            call_count = 0
+
+            def stream(**_kwargs):
+                nonlocal call_count
+                call_count += 1
+                if call_count == 1:
+                    module.stop_requested.set()
+                yield SimpleNamespace(text=response)
+
+            api_client = Mock()
+            api_client.models.generate_content_stream.side_effect = stream
+            args = module.build_parser().parse_args([str(workspace.root)])
+
+            with patch.object(module, "NB_THREADS", 1), patch.object(
+                module.prompting, "generate_request", return_value=([], Mock())
+            ):
+                module.configure_runtime(
+                    workspace, tasks, args, api_client=api_client
+                )
+                first_status = module.run_configured(args)
+                first_progress = read_json(workspace.correction_progress_file)
+                first_results = read_json(workspace.correction_file)
+
+                module.configure_runtime(
+                    workspace, tasks, args, api_client=api_client
+                )
+                pending_on_resume = list(module.tasks_to_process)
+                second_status = module.run_configured(args)
+
+            final_progress = read_json(workspace.correction_progress_file)
+            final_results = read_json(workspace.correction_file)
+
+        self.assertEqual(first_status, module.ExitCode.INTERRUPTED)
+        self.assertEqual(first_progress, [list(tasks[0])])
+        self.assertEqual(len(first_results["Ex 1"]), 1)
+        self.assertEqual(first_results["Ex 2"], [])
+        self.assertEqual(pending_on_resume, [tasks[1]])
+        self.assertEqual(second_status, module.ExitCode.SUCCESS)
+        self.assertEqual(final_progress, [list(tasks[0]), list(tasks[1])])
+        self.assertEqual(len(final_results["Ex 1"]), 1)
+        self.assertEqual(len(final_results["Ex 2"]), 1)
+        self.assertEqual(call_count, 2)
+
+    def test_interrupted_label_errors_resume_auxiliary_requests(self) -> None:
+        module = self.modules["correction"]
+        self.addCleanup(module.stop_requested.clear)
+
+        for error_type in ("wrong-label", "additional-answer"):
+            with self.subTest(error_type=error_type), tempfile.TemporaryDirectory() as directory:
+                workspace = EvaluationWorkspace(Path(directory))
+                workspace.copies_dir.mkdir()
+                group_dir = workspace.groups_dir / "Ex 1"
+                group_dir.mkdir(parents=True)
+                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"]],
+                )
+                task = (str(image), "Ex 1")
+                response = json.dumps(
+                    [
+                        {
+                            "id": "01",
+                            "result": {"error": error_type, "feedback": []},
+                        }
+                    ]
+                )
+                primary_call_count = 0
+
+                def stream(**_kwargs):
+                    nonlocal primary_call_count
+                    primary_call_count += 1
+                    module.stop_requested.set()
+                    yield SimpleNamespace(text=response)
+
+                def resolve_error(_pid, _label, result, _pdf_path):
+                    self.assertEqual(result["error"], error_type)
+                    result["error"] = ""
+                    return []
+
+                api_client = Mock()
+                api_client.models.generate_content_stream.side_effect = stream
+                args = module.build_parser().parse_args(
+                    [str(workspace.root), "--overwrite"]
+                )
+
+                with patch.object(module, "NB_THREADS", 1), patch.object(
+                    module.prompting, "generate_request", return_value=([], Mock())
+                ), patch.object(
+                    module, "handle_label_errors", side_effect=resolve_error
+                ) as handler:
+                    module.configure_runtime(
+                        workspace, [task], args, api_client=api_client
+                    )
+                    first_status = module.run_configured(args)
+                    saved_pending = read_json(
+                        workspace.correction_pending_responses_file
+                    )
+                    first_progress_exists = (
+                        workspace.correction_progress_file.exists()
+                    )
+                    first_correction_exists = workspace.correction_file.exists()
+
+                    module.configure_runtime(
+                        workspace, [task], args, api_client=api_client
+                    )
+                    second_status = module.run_configured(args)
+
+                self.assertEqual(first_status, module.ExitCode.INTERRUPTED)
+                self.assertFalse(first_progress_exists)
+                self.assertFalse(first_correction_exists)
+                self.assertEqual(
+                    json.loads(saved_pending[str(image)]), json.loads(response)
+                )
+                self.assertEqual(second_status, module.ExitCode.SUCCESS)
+                self.assertEqual(primary_call_count, 1)
+                handler.assert_called_once()
+                self.assertEqual(
+                    read_json(workspace.correction_progress_file), [list(task)]
+                )
+                corrected = read_json(workspace.correction_file)
+                self.assertEqual(corrected["Ex 1"][0][0]["result"]["error"], "")
+                self.assertEqual(
+                    read_json(workspace.correction_pending_responses_file), {}
+                )
+
     def test_correction_reset_restores_old_and_deletes_new_files(self) -> None:
         module = self.modules["correction"]
         with tempfile.TemporaryDirectory() as directory:
@@ -1255,6 +1607,9 @@ class StandardCliTests(unittest.TestCase):
             copy_dir.mkdir(parents=True)
             atomic_write_json(evaluation / "correction.json", {"old": True})
             atomic_write_json(evaluation / "correction_progress.json", ["old"])
+            atomic_write_json(
+                evaluation / "correction_pending_responses.json", {"group": "response"}
+            )
             (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")
@@ -1262,6 +1617,9 @@ class StandardCliTests(unittest.TestCase):
             self.assertEqual(module.main([str(evaluation), "--reset"]), 0)
             self.assertFalse((evaluation / "correction.json").exists())
             self.assertFalse((evaluation / "correction_progress.json").exists())
+            self.assertFalse(
+                (evaluation / "correction_pending_responses.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())
@@ -1798,6 +2156,13 @@ class WorkflowTests(unittest.TestCase):
         )
         self.assertEqual(skip_without_conflicts, {"manual_resolution"})
 
+    def test_manual_resolution_precedes_post_correction(self) -> None:
+        ordered_ids = [step.id for step in build_workflow(False)]
+        self.assertLess(
+            ordered_ids.index("manual_resolution"),
+            ordered_ids.index("post_correction"),
+        )
+
     def test_crop_auto_start_follows_always_crop_configuration(self) -> None:
         for enabled in (False, True):
             with self.subTest(enabled=enabled), patch(
@@ -1823,6 +2188,16 @@ class WorkflowTests(unittest.TestCase):
     def test_review_persp_has_shorter_title(self) -> None:
         self.assertEqual(self.steps["review_persp"].title, "Relire les barèmes")
 
+    def test_cutleft_title_and_shortcuts_are_documented(self) -> None:
+        step = self.steps["cutleft"]
+        self.assertEqual(
+            step.title,
+            "Découper une partie à gauche pour détection des labels",
+        )
+        for shortcut in ("n", "N", "t", "l", "1", "Entrée", "s"):
+            with self.subTest(shortcut=shortcut):
+                self.assertIn(shortcut, step.description)
+
     def test_export_has_shorter_title_and_annotation_argument(self) -> None:
         self.assertEqual(self.steps["export"].title, "Exporter")
         command = self.command(
@@ -1857,6 +2232,47 @@ class WorkflowTests(unittest.TestCase):
                 detected_annotation_directories(evaluation), ("Bnot", "Anot")
             )
 
+    def test_giving_names_utilities_find_and_rename_problematic_copies(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            evaluation = Path(directory)
+            return_dir = evaluation / "A Rendre"
+            original_dir = evaluation / "Copies Originales"
+            return_dir.mkdir()
+            original_dir.mkdir()
+            (original_dir / "Copie01.pdf").write_bytes(b"original")
+            folders = {
+                "Unknown (01)": (b"unknown-jpg", b"unknown-pdf"),
+                "Dupont (02)": (b"two-jpg", b"two-pdf"),
+                "Dupont (03)": (b"three-jpg", b"three-pdf"),
+                "Unique (04)": (b"four-jpg", b"four-pdf"),
+            }
+            for folder_name, (jpg, pdf) in folders.items():
+                folder = return_dir / folder_name
+                folder.mkdir()
+                base_name = folder_name.rsplit(" (", 1)[0]
+                (folder / f"{base_name}.jpg").write_bytes(jpg)
+                (folder / f"{base_name}.pdf").write_bytes(pdf)
+                (folder / "score.json").write_text("{}", encoding="utf-8")
+                (folder / "answers").mkdir()
+
+            issues = find_name_issues(evaluation)
+            self.assertEqual([issue.copy_id for issue in issues], ["01", "02", "03"])
+            self.assertEqual(
+                original_copy_path(evaluation, "01"),
+                original_dir / "Copie01.pdf",
+            )
+
+            renamed = rename_return_copy(return_dir / "Unknown (01)", "Alice")
+            self.assertEqual(renamed.name, "Alice (01)")
+            self.assertEqual((renamed / "Alice.jpg").read_bytes(), b"unknown-jpg")
+            self.assertEqual((renamed / "Alice.pdf").read_bytes(), b"unknown-pdf")
+            self.assertTrue((renamed / "score.json").is_file())
+            self.assertTrue((renamed / "answers").is_dir())
+            self.assertEqual(
+                [issue.copy_id for issue in find_name_issues(evaluation)],
+                ["02", "03"],
+            )
+
     def test_export_and_import_defaults_follow_previous_runs(self) -> None:
         with tempfile.TemporaryDirectory() as directory:
             evaluation = Path(directory)
@@ -1883,10 +2299,27 @@ class WorkflowTests(unittest.TestCase):
 
     def test_plotting_shortcuts_describe_open_actions(self) -> None:
         rendered = "\n".join(plotting_shortcut_lines())
+        self.assertIn("p : revenir à la précédente", rendered)
         self.assertIn("ouvrir l’énoncé", rendered)
         self.assertIn("ouvrir la copie traitée", rendered)
         self.assertIn("ouvrir la copie originale", rendered)
 
+    def test_gui_tracks_correction_progress_across_output_chunks(self) -> None:
+        app = object.__new__(CopienatorApp)
+        app.active_step_id = "correction"
+        app._correction_progress_buffer = ""
+        app.info_var = Mock()
+
+        app._track_correction_progress(
+            "[Progression correction] Groupes trai"
+        )
+        app.info_var.set.assert_not_called()
+        app._track_correction_progress("tés : 3/8\n")
+
+        app.info_var.set.assert_called_once_with(
+            "Correction : 3 groupe(s) traité(s) sur 8 (38 %)."
+        )
+
     def test_proxy_is_opt_in_and_prefilled(self) -> None:
         base = {"HTTPS_PROXY": "http://system-proxy", "OTHER": "kept"}
         without_proxy = build_runner_environment(
@@ -1922,6 +2355,21 @@ class WorkflowTests(unittest.TestCase):
         )
         self.assertEqual(command[4], "correct")
 
+    def test_refaire_correction_does_not_offer_overwrite(self) -> None:
+        step = self.steps["correction"]
+        overwrite = next(
+            argument for argument in step.arguments if argument.name == "overwrite"
+        )
+        self.assertNotIn("refaire", overwrite.variants)
+
+        command = self.command(
+            "correction",
+            "refaire",
+            {"target": self.evaluation, "overwrite": True},
+        )
+        self.assertIn("--refaire", command)
+        self.assertNotIn("--overwrite", command)
+
     def test_hybrid_correction_arguments(self) -> None:
         command = self.command(
             "correction", "hybrid", {"target": self.evaluation, "batch_from": "Ex 4"}

tests/test_ink_detection.py

@@ -108,6 +108,14 @@ class InkDetectionTests(unittest.TestCase):
                     .7,(190,190,190),2)
         self.assertGreater(detect_bounds(image,dpi=150)['bottom_px'],1100)
 
+    def test_sparse_central_pencil_on_dark_grid_survives(self):
+        image = self.dark_grid()
+        # Thin, pale handwriting crossing the ruling is split into sparse
+        # components. Its central position distinguishes it from page holes.
+        cv2.putText(image,'result',(330,1090),cv2.FONT_HERSHEY_SCRIPT_SIMPLEX,
+                    .85,(155,155,155),1)
+        self.assertGreater(detect_bounds(image,dpi=150)['bottom_px'],1090)
+
     def test_large_unruled_diagram_is_not_paper(self):
         image = np.full((1200,850,3),255,np.uint8)
         cv2.rectangle(image,(100,100),(750,1100),(20,20,20),3)
@@ -151,6 +159,12 @@ class InkDetectionTests(unittest.TestCase):
                 r = detect_bounds(image,dpi=150)
                 self.assertGreater(r['bottom_px'],1060)
 
+    def test_faint_page_number_in_outer_band_does_not_block_crop(self):
+        image = self.dark_grid()
+        cv2.putText(image,'4/',(3,1160),cv2.FONT_HERSHEY_SIMPLEX,
+                    .55,(130,130,130),1)
+        self.assertLess(detect_bounds(image,dpi=150)['bottom_px'],850)
+
     def test_faded_neutral_grid(self):
         image = self.dark_grid()
         image[np.all(image == 65,axis=2)] = 145

tests/test_marked_copies.py

@@ -36,6 +36,7 @@ class MarkedCopiesTests(unittest.TestCase):
         review.had_errors = False
         review.completed = False
         review.current_shift = 50
+        review.current_width_offset = 0
         review.default_max_per_file = 5
         review.current_max_per_file = 1
         review.root = Mock()
@@ -70,6 +71,7 @@ class MarkedCopiesTests(unittest.TestCase):
         self.assertIn("Copie01.pdf", copy_errors(self.workspace))
         self.assertEqual(review.index, 1)
         self.assertEqual(review.current_max_per_file, 5)
+        self.assertEqual(review.current_width_offset, 0)
         self.assertTrue(review.had_errors)
         review.load_current_image.assert_called_once_with()
 
@@ -96,6 +98,30 @@ class MarkedCopiesTests(unittest.TestCase):
         self.assertTrue(review.stop_prefetch.is_set())
         self.assertIn("Copie01.pdf", copy_errors(self.workspace))
 
+    def test_enlarge_reprocesses_with_a_wider_selection(self):
+        review = self.reviewer()
+        review.trigger_processing = Mock()
+
+        review.on_enlarge(cutleft.CROP_WIDTH_STEP)
+
+        self.assertEqual(review.current_width_offset, 50)
+        review.trigger_processing.assert_called_once_with(self.files[0], 50)
+
+    def test_process_single_pdf_enlarges_crop_and_caps_it_at_page_edge(self):
+        page = Image.new("RGB", (900, 300), "white")
+        with patch.object(cutleft, "get_pdf_pages", return_value=[page]):
+            regular = cutleft.process_single_pdf(self.files[0])
+            enlarged = cutleft.process_single_pdf(self.files[0], width_offset=200)
+            capped = cutleft.process_single_pdf(self.files[0], width_offset=1000)
+            fullpage = cutleft.process_single_pdf(
+                self.files[0], max_per_file=1, width_offset=200
+            )
+
+        self.assertEqual(regular[1][0].size, (300, 300))
+        self.assertEqual(enlarged[1][0].size, (500, 300))
+        self.assertEqual(capped[1][0].size, (800, 300))
+        self.assertEqual(fullpage[1][0].size, (900, 300))
+
     def test_processing_blocks_skip_and_failed_conversion_is_flagged(self):
         review = self.reviewer()
         review.is_processing = True

tests/test_splitting_geometry.py

@@ -8,7 +8,10 @@ import numpy as np
 import pymupdf
 from pdf2image import convert_from_path
 
-from copienator.commands.splitting_int import _render_split_outputs
+from copienator.commands.splitting_int import (
+    ANSWER_TOP_PADDING_POINTS,
+    _render_split_outputs,
+)
 
 
 class SplittingGeometryTests(unittest.TestCase):
@@ -60,7 +63,8 @@ class SplittingGeometryTests(unittest.TestCase):
                             else:
                                 coordinates = [("A", 0, 250, 260, 0, 100),
                                                ("_", 0, 711, 721, 0, 100)]
-                                bounds = (bounds[0], max(bounds[1], height // 4),
+                                answer_top = int(height // 4 - ANSWER_TOP_PADDING_POINTS)
+                                bounds = (bounds[0], max(bounds[1], answer_top),
                                           bounds[2], min(bounds[3], height * 3 // 4))
                             _render_split_outputs(source, coordinates, root)
                             self.assert_matches_preview(root / "A.pdf", preview, bounds)
@@ -85,7 +89,7 @@ class SplittingGeometryTests(unittest.TestCase):
             rendered = convert_from_path(root / "A.pdf", dpi=72)
             self.assertEqual(len(rendered), 2)
             for actual, preview, bounds in zip(rendered, previews,
-                                               [(40, 700, 780, 920), (20, 80, 760, 250)]):
+                                               [(40, 688, 780, 920), (20, 80, 760, 250)]):
                 expected = np.asarray(preview.crop(bounds)).astype(float)
                 actual = np.asarray(actual).astype(float)
                 self.assertEqual(actual.shape, expected.shape)