feat(ocr): per-script-type confidence thresholds

Kurrent OCR produces much lower confidence than typewriter/Latin.
Separate thresholds allow aggressive filtering for Kurrent (0.5)
while keeping typewriter lenient (0.3).

- OCR_CONFIDENCE_THRESHOLD: default for Surya paths (0.3)
- OCR_CONFIDENCE_THRESHOLD_KURRENT: Kraken Kurrent path (0.5)
- apply_confidence_markers() now accepts threshold parameter
- get_threshold(script_type) selects the right threshold

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

docker-compose.yml

@@ -85,6 +85,7 @@ services:
     environment:
       KRAKEN_MODEL_PATH: /app/models/german_kurrent.mlmodel
       OCR_CONFIDENCE_THRESHOLD: "0.3"
+      OCR_CONFIDENCE_THRESHOLD_KURRENT: "0.5"
     networks:
       - archive-net
     healthcheck:

ocr-service/confidence.py

@@ -2,16 +2,24 @@
 
 import os
 
-CONFIDENCE_THRESHOLD = float(os.environ.get("OCR_CONFIDENCE_THRESHOLD", "0.3"))
+THRESHOLD_DEFAULT = float(os.environ.get("OCR_CONFIDENCE_THRESHOLD", "0.3"))
+THRESHOLD_KURRENT = float(os.environ.get("OCR_CONFIDENCE_THRESHOLD_KURRENT", "0.5"))
 
 ILLEGIBLE_MARKER = "[unleserlich]"
 
 
-def apply_confidence_markers(words: list[dict]) -> str:
+def get_threshold(script_type: str) -> float:
+    if script_type and script_type.upper() == "HANDWRITING_KURRENT":
+        return THRESHOLD_KURRENT
+    return THRESHOLD_DEFAULT
+
+
+def apply_confidence_markers(words: list[dict], threshold: float | None = None) -> str:
     """Replace low-confidence words with [unleserlich], collapsing adjacent markers.
 
     Args:
         words: list of {"text": str, "confidence": float} dicts
+        threshold: confidence threshold (uses THRESHOLD_DEFAULT if None)
 
     Returns:
         Reconstructed text string with [unleserlich] substitutions.
@@ -19,11 +27,14 @@ def apply_confidence_markers(words: list[dict]) -> str:
     if not words:
         return ""
 
+    if threshold is None:
+        threshold = THRESHOLD_DEFAULT
+
     result: list[str] = []
     prev_was_marker = False
 
     for word in words:
-        if word["confidence"] < CONFIDENCE_THRESHOLD:
+        if word["confidence"] < threshold:
             if not prev_was_marker:
                 result.append(ILLEGIBLE_MARKER)
             prev_was_marker = True

ocr-service/main.py

@@ -9,7 +9,7 @@ import pypdfium2 as pdfium
 from fastapi import FastAPI, HTTPException
 from PIL import Image
 
-from confidence import apply_confidence_markers
+from confidence import apply_confidence_markers, get_threshold
 from engines import kraken as kraken_engine
 from engines import surya as surya_engine
 from models import OcrBlock, OcrRequest
@@ -72,9 +72,10 @@ async def run_ocr(request: OcrRequest):
         # TYPEWRITER, HANDWRITING_LATIN, UNKNOWN — all use Surya
         blocks = surya_engine.extract_blocks(images, request.language)
 
+    threshold = get_threshold(script_type)
     for block in blocks:
         if block.get("words"):
-            block["text"] = apply_confidence_markers(block["words"])
+            block["text"] = apply_confidence_markers(block["words"], threshold)
         block.pop("words", None)
 
     return [OcrBlock(**b) for b in blocks]

ocr-service/test_confidence.py

@@ -2,7 +2,7 @@
 
 import os
 import pytest
-from confidence import apply_confidence_markers, words_from_characters
+from confidence import apply_confidence_markers, words_from_characters, get_threshold
 
 
 # ─── apply_confidence_markers ─────────────────────────────────────────────────
@@ -75,22 +75,29 @@ def test_just_below_threshold_replaced():
     assert apply_confidence_markers(words) == "[unleserlich]"
 
 
-def test_custom_threshold_via_env(monkeypatch):
+def test_custom_threshold_via_parameter():
-    monkeypatch.setenv("OCR_CONFIDENCE_THRESHOLD", "0.8")
-    # Need to reload the module to pick up the new env var
-    import importlib
-    import confidence
-    importlib.reload(confidence)
-
     words = [
         {"text": "Lieber", "confidence": 0.95},
         {"text": "Freund", "confidence": 0.5},
     ]
-    assert confidence.apply_confidence_markers(words) == "Lieber [unleserlich]"
+    assert apply_confidence_markers(words, threshold=0.8) == "Lieber [unleserlich]"
+    assert apply_confidence_markers(words, threshold=0.3) == "Lieber Freund"
 
-    # Reset
+
-    monkeypatch.setenv("OCR_CONFIDENCE_THRESHOLD", "0.3")
+def test_kurrent_threshold_is_higher_than_default():
-    importlib.reload(confidence)
+    default = get_threshold("TYPEWRITER")
+    kurrent = get_threshold("HANDWRITING_KURRENT")
+    assert kurrent > default
+
+
+def test_get_threshold_kurrent():
+    assert get_threshold("HANDWRITING_KURRENT") == 0.5
+
+
+def test_get_threshold_default():
+    assert get_threshold("TYPEWRITER") == 0.3
+    assert get_threshold("HANDWRITING_LATIN") == 0.3
+    assert get_threshold("UNKNOWN") == 0.3
 
 
 def test_low_confidence_at_start():