refactor: rename LLM alignment score to Judge Agreement

docs/notebooks/02_inspecting_detected_entities.ipynb

@@ -671,7 +671,7 @@
        "\n",
        "Since finishing his training, <span style=\"border:1.5px solid #06b6d4;padding:1px 4px;border-radius:3px\">&lt;Bobby, first_name&gt;<span style=\"color:#06b6d4;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| first_name</span></span> has worked at <span style=\"border:1.5px solid #06b6d4;padding:1px 4px;border-radius:3px\">&lt;VCA Animal Hospital, organization_name&gt;<span style=\"color:#06b6d4;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| organization_name</span></span> and later at the <span style=\"border:1.5px solid #06b6d4;padding:1px 4px;border-radius:3px\">&lt;Colorado Veterinary Clinic, organization_name&gt;<span style=\"color:#06b6d4;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| organization_name</span></span>, where he now leads a busy mixed‑practice team. He identifies as a <span style=\"border:1.5px solid #3b82f6;padding:1px 4px;border-radius:3px\">&lt;Christian Democrat, political_view&gt;<span style=\"color:#3b82f6;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| political_view</span></span> and often volunteers at local shelters, a habit encouraged by his wife, <span style=\"border:1.5px solid #06b6d4;padding:1px 4px;border-radius:3px\">&lt;Maya, first_name&gt;<span style=\"color:#06b6d4;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| first_name</span></span>, and their two teenage children, <span style=\"border:1.5px solid #06b6d4;padding:1px 4px;border-radius:3px\">&lt;Aria, first_name&gt;<span style=\"color:#06b6d4;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| first_name</span></span> and <span style=\"border:1.5px solid #06b6d4;padding:1px 4px;border-radius:3px\">&lt;Leo, first_name&gt;<span style=\"color:#06b6d4;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| first_name</span></span>. Outside the clinic, <span style=\"border:1.5px solid #06b6d4;padding:1px 4px;border-radius:3px\">&lt;Bobby, first_name&gt;<span style=\"color:#06b6d4;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| first_name</span></span> enjoys hiking the <span style=\"border:1.5px solid #22c55e;padding:1px 4px;border-radius:3px\">&lt;Rockies, place_name&gt;<span style=\"color:#22c55e;font-size:0.75em;font-weight:600;margin-left:3px;opacity:0.85\">| place_name</span></span> with his family and mentoring veterinary students from his alma mater.</div>\n",
        "      </div>\n",
-       "      <div style='margin-bottom:16px'><div style=\"font-size:0.8em;font-weight:600;text-transform:uppercase;letter-spacing:0.05em;margin-bottom:6px;opacity:0.5\">Detection Judge</div><div style='font-size:0.9em;line-height:1.8'><strong>Detection Validity:</strong> <span style='color:#22c55e;font-weight:600'>Satisfied</span> (LLM alignment score: 19/19)<div style='font-size:0.8em;opacity:0.7;font-style:italic;margin-top:2px'>LLM alignment score: The level of alignment between the detection and evaluation LLMs across entity classification, attributes, and relationships.</div></div></div>\n",
+       "      <div style='margin-bottom:16px'><div style=\"font-size:0.8em;font-weight:600;text-transform:uppercase;letter-spacing:0.05em;margin-bottom:6px;opacity:0.5\">Detection Judge</div><div style='font-size:0.9em;line-height:1.8'><strong>Detection Validity:</strong> <span style='color:#22c55e;font-weight:600'>Satisfied</span> (Judge Agreement: 19/19)<div style='font-size:0.8em;opacity:0.7;font-style:italic;margin-top:2px'>Judge Agreement: The number of entity-label pairs accepted by the evaluation judge out of all pairs produced by the Anonymizer.</div></div></div>\n",
        "      \n",
        "      \n",
        "      \n",

src/anonymizer/engine/replace/replace_runner.py

@@ -73,7 +73,7 @@ def run(
         """Apply the replacement strategy (no LLM judges).
 
         Evaluation is a separate concern — call ``evaluate()`` on the resulting
-        dataframe when you want the LLM alignment scores.
+        dataframe when you want the Judge Agreement scores.
         """
         strategy = type(replace_method).__name__
         with stage_timer(

src/anonymizer/interface/anonymizer.py

@@ -235,7 +235,7 @@ def run(
         """Run the full anonymization pipeline (detection + replacement).
 
         No LLM evaluation judges run here — call :meth:`evaluate` on the
-        result's ``trace_dataframe`` when you want the LLM alignment scores.
+        result's ``trace_dataframe`` when you want the Judge Agreement scores.
 
         Args:
             config: Workflow behavior — replace strategy, entity labels, thresholds.
@@ -345,7 +345,7 @@ def preview(
         """Run the pipeline on a subset of records for quick inspection.
 
         No LLM evaluation judges run here — call :meth:`evaluate` on the
-        result's ``trace_dataframe`` when you want the LLM alignment scores.
+        result's ``trace_dataframe`` when you want the Judge Agreement scores.
 
         Args:
             config: Workflow behavior — replace strategy, entity labels, thresholds.

src/anonymizer/interface/display.py

@@ -579,7 +579,7 @@ def _verdict_badge(valid: object, correct: int, total: int) -> tuple[str, str]:
     else:
         verdict, color = "Partially Satisfied", "#f59e0b"
     badge = f"<span style='color:{color};font-weight:600'>{verdict}</span>"
-    rate_html = f" (LLM alignment score: {correct}/{total})"
+    rate_html = f" (Judge Agreement: {correct}/{total})"
     return badge, rate_html
 
 
@@ -668,8 +668,8 @@ def _render_detection_judge_section(row: pd.Series) -> str:
         "<div style='font-size:0.9em;line-height:1.8'>"
         f"<strong>Detection Validity:</strong> {badge}{html.escape(rate_html)}"
         "<div style='font-size:0.8em;opacity:0.7;font-style:italic;margin-top:2px'>"
-        "LLM alignment score: The level of alignment between the detection and "
-        "evaluation LLMs across entity classification, attributes, and relationships."
+        "Judge Agreement: The number of entity-label pairs accepted by the evaluation "
+        "judge out of all pairs produced by the Anonymizer."
         "</div>"
         "</div>"
     )
@@ -920,7 +920,7 @@ def _render_relational_consistency_section(row: pd.Series) -> str:
     all_relations = _extract_all_relations(row)
     invalid_count = sum(1 for r in all_relations if not bool(r.get("passes", False)))
     # Fall back to the invalid-relations column when the raw output is missing,
-    # so the LLM alignment score still surfaces "at least this many failures".
+    # so Judge Agreement still surfaces "at least this many failures".
     if not all_relations:
         fallback_invalid = _normalize_relations(row.get(COL_RELATIONAL_CONSISTENCY_INVALID_RELATIONS))
         invalid_count = len(fallback_invalid)
@@ -1051,7 +1051,7 @@ def _count_detected_entity_label_pairs(row: pd.Series) -> int:
     """Count (value, label) pairs the judge had a chance to evaluate.
 
     The judge schema flags entities at the (value, label) granularity, so the
-    denominator for the LLM alignment score is the total number of such pairs in the
+    denominator for Judge Agreement is the total number of such pairs in the
     deduped entity payload, not the number of unique values.
     """
     raw = row.get(COL_ENTITIES_BY_VALUE) if COL_ENTITIES_BY_VALUE in row.index else None

tests/interface/test_display.py

@@ -158,7 +158,7 @@ def test_normalize_replacement_map_non_dict_returns_empty() -> None:
 def test_verdict_badge_satisfied_when_all_correct_and_valid_true() -> None:
     badge, rate = _verdict_badge(valid=True, correct=10, total=10)
     assert "Satisfied" in badge and "Not" not in badge
-    assert "10/10" in rate
+    assert rate == " (Judge Agreement: 10/10)"
 
 
 def test_verdict_badge_partial_for_mixed_count() -> None: