From 638dbf9b9fcdaf94a49478b209873e2dacc80580 Mon Sep 17 00:00:00 2001
From: Chaim <chaim@marcus-law.co.il>
Date: Sun, 28 Jun 2026 22:10:39 +0000
Subject: [PATCH] feat(learning): feed corpus-measured section ratios to the
 writer (P2 #7, structural half)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Golden ratios were measured from the corpus (style_distance.measure_corpus_ratios)
but never reached the writer — it used only the hardcoded STRUCTURE_GUIDANCE text.
So the structural target was frozen, and drafts kept deviating (e.g. discussion
block 10% vs a 40-47% target) with no corrective signal.

_structure_guidance_with_corpus(outcome, practice_area) now appends the
corpus-measured section ratios for the block's outcome to the structure guidance,
at both writer entry points (write_block + get_block_context) via one shared helper
(G2). Because measure_corpus_ratios reads style_corpus — which grows with every
enrolled final (see exemplar-growth, #343) — the writer's structural target now
reflects Dafna's ACTUAL current distribution and improves as the corpus grows.
Deterministic (no LLM); best-effort.

This is the STRUCTURAL half of the voice-profile-refresh (#7). The remaining half —
LLM-regenerating the abstract voice-fingerprint PROSE from the corpus (chair-gated)
— is a distinct larger initiative, deliberately deferred rather than rushed.

Invariants: G1 (the writer gets the live measured target, not a frozen constant),
G2 (one structure-guidance builder shared by both entry points; reuses the existing
measure_corpus_ratios — no parallel measurement), INV-LRN4 (the corpus the loop
grows now feeds the writer's structural target).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../src/legal_mcp/services/block_writer.py    | 42 +++++++++++++------
 1 file changed, 30 insertions(+), 12 deletions(-)

diff --git a/mcp-server/src/legal_mcp/services/block_writer.py b/mcp-server/src/legal_mcp/services/block_writer.py
index 38caccb..31edd77 100644
--- a/mcp-server/src/legal_mcp/services/block_writer.py
+++ b/mcp-server/src/legal_mcp/services/block_writer.py
@@ -297,6 +297,32 @@ STRUCTURE_GUIDANCE = {
     "partial_acceptance": "קבלה חלקית — מיפוי מתחים: מה מתקבל ולמה, מה נדחה ולמה, איזון.",
 }
 
+_RATIO_SECTION_HE = {"background": "רקע", "claims": "טענות", "discussion": "דיון", "summary": "סיכום"}
+
+
+async def _structure_guidance_with_corpus(outcome: str, practice_area: str) -> str:
+    """Outcome structure guidance + the CORPUS-MEASURED section ratios (T10). The
+    measured ratios come from style_distance.measure_corpus_ratios over style_corpus,
+    which grows with every enrolled final — so the writer's structural target reflects
+    Dafna's ACTUAL current distribution, not a frozen constant (this is the structural
+    half of the voice-profile that learns from each decision). Deterministic (no LLM);
+    best-effort. Shared by both writer entry points (G2)."""
+    sg = STRUCTURE_GUIDANCE.get(outcome, "")
+    if practice_area == "betterment_levy":
+        sg = (sg + " | היטל השבחה: "
+              + " ".join(PRACTICE_AREA_OVERRIDES["betterment_levy"]["discussion_rules"])).strip()
+    try:
+        from legal_mcp.services.style_distance import measure_corpus_ratios
+        entry = (await measure_corpus_ratios() or {}).get(outcome) or {}
+        secs = entry.get("sections") or {}
+        if secs:
+            parts = [f"{_RATIO_SECTION_HE.get(s, s)} ~{round(p)}%" for s, p in secs.items()]
+            sg = (sg + f" | מבנה-יעד מדוד מהקורפוס ({entry.get('n', 0)} החלטות {outcome}): "
+                  + " · ".join(parts)).strip()
+    except Exception as e:
+        logger.warning("corpus ratios not loaded for structure guidance: %s", e)
+    return sg
+
 
 async def write_block(
     case_id: UUID,
@@ -352,12 +378,8 @@ async def write_block(
     post_hearing_context = await _build_post_hearing_context(case_id)
 
     outcome = canonical_outcome((decision or {}).get("outcome", "rejection"))
-    structure_guidance = STRUCTURE_GUIDANCE.get(outcome, "")
-    if case.get("practice_area") == "betterment_levy":
-        structure_guidance = (
-            structure_guidance + " | היטל השבחה: "
-            + " ".join(PRACTICE_AREA_OVERRIDES["betterment_levy"]["discussion_rules"])
-        ).strip()
+    structure_guidance = await _structure_guidance_with_corpus(
+        outcome, case.get("practice_area", ""))
 
     # Content checklist — tells block-yod WHAT topics to cover
     content_checklist = ""
@@ -1094,12 +1116,8 @@ async def get_block_context(case_id: UUID, block_id: str, instructions: str = ""
     post_hearing_context = await _build_post_hearing_context(case_id)
 
     outcome = canonical_outcome((decision or {}).get("outcome", "rejection"))
-    structure_guidance = STRUCTURE_GUIDANCE.get(outcome, "")
-    if case.get("practice_area") == "betterment_levy":
-        structure_guidance = (
-            structure_guidance + " | היטל השבחה: "
-            + " ".join(PRACTICE_AREA_OVERRIDES["betterment_levy"]["discussion_rules"])
-        ).strip()
+    structure_guidance = await _structure_guidance_with_corpus(
+        outcome, case.get("practice_area", ""))
 
     # Content checklist + methodology for block-yod
     content_checklist = ""