feat(calibration): כיול-אמפירי model×effort מול הסופיים (#208)

A/B harness שמכייל את ה-effort הנעוץ per-בלוק (#204) מול הסופיים של דפנה:
מייצר מחדש כל בלוק דרך מסלול-הייצור (write_block(effort_override=…) →
claude_session.query → claude -p, Opus 4.8, מקומי-בלבד) ומודד מול הסקשן
המתאים בסופי דרך style_distance.block_distance_to_final (change_percent,
anti_pattern_total, golden-ratio deviation, composite distance). ממליץ
per-בלוק על ה-effort הקרוב-ביותר לסופי.

- scripts/calibrate_effort.py — ההארנס (מודל eval_retrieval.py): --self-test
  (offline, מוכיח מדידה+המלצה, אפס DB/CLI) · --dry-run · --efforts/--blocks/
  --case/--repeats. דוח data/eval/effort-calibration-<ts>.{json,md} עם
  גודל-מדגם בולט — עדות-כיוון, לא רגרסיה (מעט סופיים-עלויים).
- style_distance.py — block_distance_to_final + split_final_by_section
  (מקור-מדידה יחיד, G2; reuse compute_diff_stats/count_anti_patterns/chunker).
- block_writer.py — write_block(effort_override=) להזרקת effort per-קריאה
  בלי לדרוס את ברירות-המחדל הנעוצות; רושם את ה-effort האפקטיבי.
- scripts/SCRIPTS.md — ערך חדש.

Invariants: G8 (eval-harness — מדידה אמפירית, לא הנחה) · G2 (reuse של
style_distance/learning_loop — אין מסלול-מדד מקביל) · claude_session
local-only (reference_claude_generation_path) · INV-LRN4/5 (השוואה מול
הסופי, מדידת-סגנון; אין מהות-תיק נגררת). אומת: --self-test 14/14 PASS.
הריצה החיה host-only (claude CLI) — לא ניתנת-להרצה ב-worktree/קונטיינר.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

mcp-server/src/legal_mcp/services/block_writer.py

@@ -355,6 +355,7 @@ async def write_block(
     case_id: UUID,
     block_id: str,
     instructions: str = "",
+    effort_override: str | None = None,
 ) -> dict:
     """כתיבת בלוק יחיד בהחלטה.
 
@@ -362,6 +363,11 @@ async def write_block(
         case_id: מזהה התיק
         block_id: מזהה הבלוק (block-alef, block-he, block-yod, ...)
         instructions: הנחיות נוספות
+        effort_override: optional per-call reasoning effort (low/medium/high/
+            xhigh/max). When set, overrides BLOCK_CONFIG[block_id].effort for
+            THIS call only — used by the #208 model/effort calibration harness
+            to A/B efforts without mutating the pinned defaults. Production
+            callers leave it None and get the deterministic per-block effort.
 
     Returns:
         dict עם content, word_count, block_id, generation_type
@@ -472,7 +478,7 @@ async def write_block(
     # reasoning effort so generation is structurally deterministic — these were
     # previously NOT forwarded (the source of inconsistency). model/effort flow
     # through claude_session.query → `claude -p --model … --effort …`.
-    effort = block_cfg.get("effort", DEFAULT_EFFORT)
+    effort = effort_override or block_cfg.get("effort", DEFAULT_EFFORT)
     timeout = claude_session.LONG_TIMEOUT if effort in _LONG_EFFORTS else claude_session.DEFAULT_TIMEOUT
     content = await claude_session.query(
         prompt,
@@ -485,6 +491,10 @@ async def write_block(
     sources = await _collect_block_sources(case_id, block_id)
     sources["case_law_ids"] = _precedent_case_law_ids
     result = _build_result(block_id, content, block_cfg)
+    # Record the EFFECTIVE effort (override wins) so the harness can attribute
+    # the measured distance to the effort that actually produced the text.
+    if result.get("effort") is not None:
+        result["effort"] = effort
     result["sources"] = sources
     return result
 

mcp-server/src/legal_mcp/services/style_distance.py

@@ -15,6 +15,7 @@ import re
 from uuid import UUID
 
 from legal_mcp.services import db
+from legal_mcp.services.learning_loop import compute_diff_stats
 from legal_mcp.services.lessons import ANTI_PATTERNS, GOLDEN_RATIOS, canonical_outcome
 
 logger = logging.getLogger(__name__)
@@ -122,6 +123,93 @@ def golden_ratio_adherence(block_word_counts: dict[str, int], outcome: str) -> d
     return {"outcome": outcome, "total_words": total, "sections": sections, "max_deviation": max_dev}
 
 
+def split_final_by_section(final_text: str) -> dict[str, str]:
+    """Group a signed final decision into golden-ratio sections (#208 calibration).
+
+    Reuses the SAME structure-aware splitter as measure_corpus_ratios
+    (chunker._split_into_sections + _CHUNK_SECTION_TO_GOLDEN) — no parallel
+    parsing path (G2). Returns {golden_section: concatenated_text} for the
+    sections that map to an AI block (background/claims/discussion/summary).
+    A section type that does not map (e.g. headers) is dropped, never silently
+    folded into another section. Best-effort: an unsplittable final returns {}.
+    """
+    from legal_mcp.services.chunker import _split_into_sections
+
+    by_section: dict[str, list[str]] = {}
+    for stype, stext in _split_into_sections(final_text or ""):
+        g = _CHUNK_SECTION_TO_GOLDEN.get(stype)
+        if g and stext.strip():
+            by_section.setdefault(g, []).append(stext.strip())
+    return {sec: "\n\n".join(parts) for sec, parts in by_section.items()}
+
+
+def block_distance_to_final(
+    block_id: str,
+    regenerated_text: str,
+    final_section_text: str,
+    outcome: str,
+    section_target_total_words: int | None = None,
+) -> dict:
+    """Distance of ONE regenerated block from the chair's matching final section.
+
+    The per-(block, effort) measurement cell for the #208 model/effort
+    calibration harness. Pure/deterministic (no LLM, no DB) — reuses the
+    existing style-distance primitives so the harness has no parallel metric
+    path (G2 / INV-G8 eval-harness):
+
+      • change_percent          — compute_diff_stats(regen, final_section)
+                                   (learning_loop, the SAME diff the pairing
+                                   ledger stores). Lower ⇒ the draft already
+                                   reads like the final ⇒ less chair rewriting.
+      • anti_pattern_total      — count_anti_patterns(regen) (lessons.ANTI_PATTERNS).
+                                   Lower ⇒ closer to Dafna's continuous-narrative
+                                   voice; the CLEANEST style signal (07-learning §0.7).
+      • golden_ratio_deviation_pp — |regen %-of-total − final %-of-total| for this
+                                   block's section. 0 ⇒ same structural weight as
+                                   the final. Requires the final's total words
+                                   (section_target_total_words); otherwise None
+                                   (we never fabricate a denominator).
+
+    Returns the three metrics + a single composite `distance` (normalized,
+    lower=closer) the harness ranks efforts by.
+    """
+    outcome = canonical_outcome(outcome)
+    diff = compute_diff_stats(regenerated_text or "", final_section_text or "")
+    change_percent = diff["change_percent"]
+    anti_total = count_anti_patterns(regenerated_text or "")["total"]
+
+    section = _BLOCK_TO_SECTION.get(block_id)
+    regen_words = len((regenerated_text or "").split())
+    final_words = len((final_section_text or "").split())
+    ratio_dev: float | None = None
+    if section and section_target_total_words and section_target_total_words > 0:
+        # Replace the final's own block contribution with the regen's, holding
+        # the rest of the final constant, to compare structural weight fairly.
+        regen_total = section_target_total_words - final_words + regen_words
+        if regen_total > 0:
+            regen_pct = regen_words / regen_total * 100
+            final_pct = final_words / section_target_total_words * 100
+            ratio_dev = round(abs(regen_pct - final_pct), 1)
+
+    # Composite: normalize each component to ~[0,1] and average the present ones.
+    # change_percent/100, anti_total/10 (10+ hits is already very bad), ratio/20.
+    comps: list[float] = [min(change_percent / 100.0, 1.0), min(anti_total / 10.0, 1.0)]
+    if ratio_dev is not None:
+        comps.append(min(ratio_dev / 20.0, 1.0))
+    distance = round(sum(comps) / len(comps), 4)
+
+    return {
+        "block_id": block_id,
+        "section": section,
+        "regen_words": regen_words,
+        "final_words": final_words,
+        "change_percent": change_percent,
+        "anti_pattern_total": anti_total,
+        "golden_ratio_deviation_pp": ratio_dev,
+        "distance": distance,
+    }
+
+
 async def style_distance(case_number: str) -> dict:
     """Assemble the 3 style-distance components for one case (T7)."""
     case = await db.get_case_by_number(case_number)