fix(csv): preserve finished targets and honor batch size env

- Queue only untranslated rows from mixed skip-check batches
- Let .env batchsize and frequency_penalty override model defaults
- Cover skip behavior and pricing batch size override in tests

gui/csv_tab.py

@@ -193,8 +193,8 @@ class CSVTab(QWidget):
         self.skip_if_target_translated_cb = QCheckBox("Skip if Target Translated")
         self.skip_if_target_translated_cb.setToolTip(
             "Check candidate rows in batches matching the global Batch Size setting. "
-            "Skip a batch only if every target is already translated (non-empty, no Japanese). "
-            "If any row still needs work, the whole batch is translated."
+            "Skip a batch when every target is already translated (non-empty, no Japanese). "
+            "If any row still needs work, translate only the unfinished rows in that batch."
         )
         left_column.addWidget(self.skip_if_target_translated_cb)
         

modules/csv.py

@@ -42,13 +42,13 @@ FILENAME = None
 BRACKETNAMES = False
 
 # CSV Configuration Settings (configurable via GUI)
-CSV_DELIMITER = "	"  # CSV delimiter character (comma, semicolon, tab)
-SOURCE_COLUMN = 2  # Which column has the source text to translate (0-indexed)
-TARGET_COLUMN = 3  # Which column to write translations to
+CSV_DELIMITER = ","  # CSV delimiter character (comma, semicolon, tab)
+SOURCE_COLUMN = 0  # Which column has the source text to translate (0-indexed)
+TARGET_COLUMN = 1  # Which column to write translations to
 SPEAKER_COLUMN = 1  # Which column has speaker names (-1 = none)
-SKIP_HEADER_ROW = False  # Skip the first row (header)
+SKIP_HEADER_ROW = True  # Skip the first row (header)
 USE_TARGET_IF_NOT_EMPTY = False  # Use target column text if not empty (T++ style)
-SKIP_IF_TARGET_TRANSLATED = False  # Skip batches whose targets are already translated
+SKIP_IF_TARGET_TRANSLATED = True  # Skip batches whose targets are already translated
 WRITE_TO_NEXT_COLUMN = False  # Write to column after target instead of overwriting
 PARSE_NAME_TAGS = False  # Parse :name[] tags in text
 PARSE_M_MARKERS = False  # Parse \M markers in text
@@ -256,9 +256,10 @@ def _collect_process_indices(data):
     """Row indices to translate.
 
     When SKIP_IF_TARGET_TRANSLATED is on, candidates are checked in groups of
-    BATCHSIZE (same as the translation batch size). A group is skipped only if
-    every target in it is already translated; if any row still needs work, the
-    whole group runs.
+    BATCHSIZE (same as the translation batch size):
+    - If every target in the group is already translated, skip the group.
+    - If any row still needs work, keep the group but only queue the
+      untranslated rows so finished targets are left alone.
     """
     candidates = [i for i in range(len(data)) if _is_candidate_row(data, i)]
     if not SKIP_IF_TARGET_TRANSLATED:
@@ -270,7 +271,7 @@ def _collect_process_indices(data):
         batch = candidates[start:start + batch_size]
         if all(_target_is_translated(data[i]) for i in batch):
             continue
-        process.extend(batch)
+        process.extend(i for i in batch if not _target_is_translated(data[i]))
     return process
 
 

tests/test_csv_skip_translated.py

@@ -89,7 +89,7 @@ class SkipTranslatedBatchIndexTests(unittest.TestCase):
         # First batch (0,1) all translated -> skip; second batch (2,3) needs work
         self.assertEqual(csv_mod._collect_process_indices(data), [2, 3])
 
-    def test_partial_batch_keeps_entire_batch(self):
+    def test_partial_batch_queues_only_untranslated_rows(self):
         csv_mod.SKIP_IF_TARGET_TRANSLATED = True
         csv_mod.BATCHSIZE = 3
         data = [
@@ -100,9 +100,9 @@ class SkipTranslatedBatchIndexTests(unittest.TestCase):
             ["お", "O"],
             ["か", "Ka"],
         ]
-        # Batch [0,1,2] has untranslated row 1 -> keep all three
+        # Batch [0,1,2] has work -> keep only untranslated row 1
         # Batch [3,4,5] all translated -> skip
-        self.assertEqual(csv_mod._collect_process_indices(data), [0, 1, 2])
+        self.assertEqual(csv_mod._collect_process_indices(data), [1])
 
     def test_uses_global_batch_size(self):
         csv_mod.SKIP_IF_TARGET_TRANSLATED = True
@@ -112,6 +112,18 @@ class SkipTranslatedBatchIndexTests(unittest.TestCase):
         # First 30 fully translated -> skipped; last singleton batch needs work
         self.assertEqual(csv_mod._collect_process_indices(data), [30])
 
+    def test_header_with_english_target_skipped_when_opt_in(self):
+        csv_mod.SKIP_IF_TARGET_TRANSLATED = True
+        csv_mod.SKIP_HEADER_ROW = False
+        csv_mod.BATCHSIZE = 10
+        data = [
+            ["Source", "Target"],
+            ["こんにちは", ""],
+            ["ありがとう", "Thank you."],
+        ]
+        # Header target looks translated; thank-you row preserved; only empty target queued
+        self.assertEqual(csv_mod._collect_process_indices(data), [1])
+
 
 class SkipIfTargetTranslatedCollectTests(unittest.TestCase):
     def setUp(self):
@@ -176,14 +188,14 @@ class SkipIfTargetTranslatedCollectTests(unittest.TestCase):
             with patch.object(csv_mod, "dazedwrap") as mock_wrap:
                 mock_wrap.wrapText.side_effect = lambda text, _width: text
                 csv_mod.translateCSV(data, pbar, None, MagicMock(), "test.csv", None)
-        # First batch fully translated -> skipped; second batch kept whole
+        # First batch fully translated -> skipped; second batch keeps unfinished only
         self.assertEqual(mock_ai.call_args[0][0], ["さようなら", "ありがとう"])
         self.assertEqual(data[0][1], "Hello")
         self.assertEqual(data[1][1], "Good morning")
         self.assertEqual(data[2][1], "Goodbye")
         self.assertEqual(data[3][1], "Thank you")
 
-    def test_partial_batch_retranslates_already_done_rows(self):
+    def test_partial_batch_preserves_already_done_rows(self):
         csv_mod.SKIP_IF_TARGET_TRANSLATED = True
         csv_mod.BATCHSIZE = 2
         data = [
@@ -192,14 +204,14 @@ class SkipIfTargetTranslatedCollectTests(unittest.TestCase):
         ]
         pbar = MagicMock()
         with patch.object(
-            csv_mod, "translateAI", return_value=(["Hi again", "Goodbye"], [0, 0])
+            csv_mod, "translateAI", return_value=(["Goodbye"], [0, 0])
         ) as mock_ai:
             with patch.object(csv_mod, "dazedwrap") as mock_wrap:
                 mock_wrap.wrapText.side_effect = lambda text, _width: text
                 csv_mod.translateCSV(data, pbar, None, MagicMock(), "test.csv", None)
-        # Mixed batch: keep both rows, including the already-translated one
-        self.assertEqual(mock_ai.call_args[0][0], ["こんにちは", "さようなら"])
-        self.assertEqual(data[0][1], "Hi again")
+        # Mixed batch: only the unfinished row is sent; existing translation kept
+        self.assertEqual(mock_ai.call_args[0][0], ["さようなら"])
+        self.assertEqual(data[0][1], "Hello")
         self.assertEqual(data[1][1], "Goodbye")
 
 

tests/test_pricing_batchsize_override.py

@@ -0,0 +1,22 @@
+import os
+import unittest
+from unittest.mock import patch
+
+from util.translation import getPricingConfig
+
+
+class PricingBatchSizeOverrideTests(unittest.TestCase):
+    def test_env_batchsize_overrides_model_default(self):
+        with patch.dict(os.environ, {"batchsize": "2"}, clear=False):
+            cfg = getPricingConfig("claude-sonnet-4-6")
+        self.assertEqual(cfg["batchSize"], 2)
+
+    def test_missing_env_keeps_model_default(self):
+        env = {k: v for k, v in os.environ.items() if k != "batchsize"}
+        with patch.dict(os.environ, env, clear=True):
+            cfg = getPricingConfig("claude-sonnet-4-6")
+        self.assertEqual(cfg["batchSize"], 30)
+
+
+if __name__ == "__main__":
+    unittest.main()

util/translation.py

@@ -1756,6 +1756,21 @@ def getPricingConfig(model):
     if _live_override:
         cfg.update(_live_override)
 
+    # Config / .env overrides win over model defaults so the Batch Size and
+    # Frequency Penalty controls actually affect translation batching.
+    env_batch = os.getenv("batchsize")
+    if env_batch not in (None, ""):
+        try:
+            cfg["batchSize"] = max(1, int(env_batch))
+        except (TypeError, ValueError):
+            pass
+    env_penalty = os.getenv("frequency_penalty")
+    if env_penalty not in (None, ""):
+        try:
+            cfg["frequencyPenalty"] = float(env_penalty)
+        except (TypeError, ValueError):
+            pass
+
     return cfg