Provide Gemini with the list of IDs in correction request

tests/test_correction_ids.py

@@ -0,0 +1,69 @@
+import json
+import tempfile
+import unittest
+from pathlib import Path
+from unittest.mock import Mock, patch
+
+from copienator import EvaluationWorkspace, atomic_write_json
+from copienator.commands import correction
+
+
+class CorrectionIdTests(unittest.TestCase):
+    def test_mismatched_ids_are_rejected_before_any_side_effects(self):
+        for ids, diagnostic in (
+            (["37", "03", "05"], "unexpected IDs: ['03']"),
+            (["37"], "missing IDs: ['05']"),
+            (["37", "05", "05"], "duplicate IDs: ['05']"),
+        ):
+            with self.subTest(ids=ids), tempfile.TemporaryDirectory() as directory:
+                root = Path(directory)
+                group = root / "Par label" / "Ex 1" / "Group_1.jpg"
+                group.parent.mkdir(parents=True)
+                group.touch()
+                atomic_write_json(group.with_suffix(".json"), [
+                    ["37", 0, 500, 1], ["05", 550, 1000, 1],
+                ])
+                pdf = root / "Copies" / "Copie37" / "Ex 1.pdf"
+                pdf.parent.mkdir(parents=True)
+                pdf.write_bytes(b"original")
+                workspace = EvaluationWorkspace(root)
+                prior = {"Other": [[{"id": "22", "result": {"score": 4}}]]}
+                atomic_write_json(workspace.correction_file, prior)
+                before = workspace.correction_file.read_bytes()
+                args = correction.build_parser().parse_args([str(root)])
+                correction.configure_runtime(workspace, [(str(group), "Ex 1")], args, api_client=Mock())
+                response = [{"id": pid, "result": {
+                    "score": 0, "error": "empty-answer", "feedback": [],
+                }} for pid in ids]
+                with patch.object(correction.prompting, "generate_request", return_value=([], {})), patch.object(
+                    correction, "handle_label_errors"
+                ) as handle:
+                    correction.process_single_task((str(group), "Ex 1"), json.dumps(response))
+                handle.assert_not_called()
+                self.assertEqual(pdf.read_bytes(), b"original")
+                self.assertFalse(pdf.with_name("Ex 1_old.pdf").exists())
+                self.assertEqual(workspace.correction_file.read_bytes(), before)
+                self.assertEqual(correction.completed_tasks, [])
+                self.assertEqual(correction.results, prior)
+                self.assertIn(diagnostic, correction.errors_summary[0][0])
+
+    def test_valid_ids_can_be_returned_in_different_order(self):
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            group = root / "Par label" / "Ex 1" / "Group_1.jpg"
+            group.parent.mkdir(parents=True)
+            group.touch()
+            atomic_write_json(group.with_suffix(".json"), [
+                ["37", 0, 500, 1], ["05", 550, 1000, 1],
+            ])
+            workspace = EvaluationWorkspace(root)
+            args = correction.build_parser().parse_args([str(root)])
+            correction.configure_runtime(workspace, [(str(group), "Ex 1")], args, api_client=Mock())
+            response = [{"id": pid, "result": {
+                "score": 4, "error": "", "feedback": [],
+            }} for pid in ["05", "37"]]
+            with patch.object(correction.prompting, "generate_request", return_value=([], {})):
+                correction.process_single_task((str(group), "Ex 1"), json.dumps(response))
+            self.assertEqual(correction.errors_summary, [])
+            self.assertEqual(correction.results["Ex 1"], [response])
+            self.assertEqual(correction.completed_tasks, [(str(group), "Ex 1")])

tests/test_cutleft_narrow_pages.py

@@ -0,0 +1,73 @@
+import tempfile
+import unittest
+from pathlib import Path
+from unittest.mock import patch
+
+import pymupdf
+from PIL import Image
+
+from copienator.commands import cutleft
+
+
+class NarrowPageCropTests(unittest.TestCase):
+    def test_mixed_pages_crop_narrow_margin_from_very_left(self):
+        normal = Image.new('RGB', (900, 600), 'red')
+        normal.paste('green', (100, 0, 400, 600))
+        narrow = Image.new('RGB', (450, 600), 'red')
+        narrow.paste('blue', (0, 0, 300, 600))
+        thin = Image.new('RGB', (200, 600), 'yellow')
+        with patch.object(cutleft, 'get_pdf_pages', return_value=[normal, narrow, thin]):
+            result = cutleft.process_single_pdf(Path('/tmp/no-original/Copy.pdf'))
+        _, images, schema = result
+        image = images[0]
+        self.assertEqual(schema['full_width_pages'], [3])
+        self.assertEqual(schema['columns_per_file'], [3])
+        self.assertEqual(image.size, (910, 600))
+        self.assertEqual(image.getpixel((0, 0)), (0, 128, 0))
+        self.assertEqual(image.getpixel((299, 0)), (0, 128, 0))
+        self.assertEqual(image.getpixel((300, 0)), (0, 0, 0))
+        self.assertEqual(image.getpixel((305, 0)), (0, 0, 255))
+        self.assertEqual(image.getpixel((604, 0)), (0, 0, 255))
+        self.assertEqual(image.getpixel((610, 0)), (255, 255, 0))
+        self.assertEqual(image.getpixel((809, 0)), (255, 255, 0))
+        self.assertEqual(image.getpixel((810, 0)), (255, 255, 255))
+        self.assertEqual(image.getpixel((909, 0)), (255, 255, 255))
+
+    def test_narrow_margin_can_be_shifted_and_enlarged(self):
+        normal = Image.new('RGB', (900, 600), 'white')
+        narrow = Image.new('RGB', (450, 600), 'red')
+        narrow.paste('blue', (50, 0, 450, 600))
+        with patch.object(cutleft, 'get_pdf_pages', return_value=[normal, narrow]):
+            result = cutleft.process_single_pdf(
+                Path('/tmp/no-original/Copy.pdf'), shift_offset=50, width_offset=100)
+        self.assertEqual(result[1][0].size, (805, 600))
+        self.assertEqual(result[1][0].getpixel((405, 0)), (0, 0, 255))
+        self.assertEqual(result[1][0].getpixel((804, 0)), (0, 0, 255))
+
+    def test_legacy_original_identifies_narrow_pages_without_normal_siblings(self):
+        with tempfile.TemporaryDirectory() as temporary:
+            root = Path(temporary)
+            originals = root / 'Copies Originales'
+            originals.mkdir()
+            with pymupdf.open() as document:
+                document.new_page(width=1191, height=842)
+                document.save(originals / 'Copy.pdf')
+            images = [Image.new('RGB', (900, 2339)), Image.new('RGB', (750, 2339))]
+            with patch.object(cutleft, 'get_pdf_pages', return_value=images):
+                result = cutleft.process_single_pdf(root / 'Copies' / 'Copy.pdf')
+            self.assertEqual(result[2]['full_width_pages'], [])
+            self.assertEqual(result[1][0].width, 2 * int((1191 / 2 * 200 / 72) // 3) + 5)
+
+    def test_small_main_split_adjustments_keep_margin_cropping(self):
+        images = [Image.new('RGB', (1600, 2339)), Image.new('RGB', (1694, 2339))]
+        with patch.object(cutleft, 'get_pdf_pages', return_value=images):
+            result = cutleft.process_single_pdf(Path('/tmp/no-original/Copy.pdf'))
+        self.assertEqual(result[2]['full_width_pages'], [])
+        self.assertEqual(result[1][0].width, 2 * (1694 // 3) + 5)
+
+    def test_fullpage_option_still_keeps_every_page_separately(self):
+        images = [Image.new('RGB', (900, 600)), Image.new('RGB', (450, 600))]
+        with patch.object(cutleft, 'get_pdf_pages', return_value=images):
+            result = cutleft.process_single_pdf(Path('/tmp/no-original/Copy.pdf'), max_per_file=1)
+        self.assertEqual([image.size for image in result[1]], [(900, 600), (450, 600)])
+        self.assertEqual(result[2]['columns_per_file'], [1, 1])

tests/test_gemini_config.py

@@ -26,6 +26,7 @@ class GeminiConfigTests(unittest.TestCase):
         with tempfile.TemporaryDirectory() as directory:
             image = Path(directory) / "answer.jpg"
             image.write_bytes(b"image")
+            image.with_suffix(".json").write_text('[["01", 0, 400, 1.0, "Ex 1"]]')
             for level in ("minimal", "low", "medium", "high", None):
                 with self.subTest(level=level), patch.object(prompting, "make_prompt", return_value="Grade"):
                     _, config = prompting.generate_request(directory, image, "Ex 1", thinking_level=level)

tests/test_gui_core.py

@@ -1775,8 +1775,9 @@ class StandardCliTests(unittest.TestCase):
                 args,
                 api_client=Mock(),
             )
-            with patch.object(module.prompting, "make_prompt", return_value="prompt"):
+            with patch.object(module.prompting, "make_prompt", return_value="prompt") as make_prompt:
                 self.assertEqual(module.run_configured(args), 0)
+            make_prompt.assert_called_once_with(evaluation, "Ex 1", expected_ids=["01"])
             lines = (evaluation / "batch_requests_flash.jsonl").read_text().splitlines()
             self.assertEqual(len(lines), 1)
             self.assertEqual(json.loads(lines[0])["key"], str(image))

tests/test_prompting.py

@@ -1,10 +1,26 @@
 import unittest
+import tempfile
+from pathlib import Path
 from unittest.mock import patch
 
 from copienator import prompting
 
 
 class PromptingTests(unittest.TestCase):
+    def test_live_prompt_includes_only_metadata_ids_in_order(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            image = Path(directory) / "Group_16.jpg"
+            image.write_bytes(b"image")
+            image.with_suffix(".json").write_text(
+                '[["22", 50, 834, 1], ["37", 904, 3250, 1], '
+                '["13", 3320, 4072, 1], ["05", 4142, 4892, 1]]'
+            )
+            contents, _ = prompting.generate_request(directory, image, "I)12)")
+        prompt = contents[0].parts[1].text
+        self.assertIn('["22", "37", "13", "05"]', prompt)
+        self.assertIn("exactement un résultat par identifiant", prompt)
+        self.assertIn("ignore les numéros manuscrits", prompt)
+
     def test_perspective_is_inserted_with_alternative_method_guidance(self) -> None:
         with patch.object(
             prompting, "get_label_text_content", return_value="Question"