diff --git a/docs/spec/07-learning.md b/docs/spec/07-learning.md
index 0eae7eb..8929e67 100644
--- a/docs/spec/07-learning.md
+++ b/docs/spec/07-learning.md
@@ -228,6 +228,27 @@ Dimensions for Data Quality* (2013) · ISO 8000 (Data quality) | סטטוס: ver
 (`/precedents`, PR#300) לאישור-יו"ר; CSV-audit ב-`data/audit/canonical-synthesis-*.csv`.
 **הפרה ידועה:** — (חדש)
 
+### INV-LRN7: חילוץ-עקרונות מגודר-פאנל + טרמינולוגיה נכונה (#152 → G2/G10/INV-AH)
+**כלל:** חילוץ עקרונות-משפטיים מפסיקה (להבא ורטרואקטיבית) עובר משטר-פאנל אחיד:
+**3 מודלים עצמאיים** (Claude מקומי + DeepSeek + Gemini) מנתחים לעומק כל החלטה,
+מציעים מועמדים עם ציון, המועמדים מותאמים בין-מודלית (cosine), ולכל אחד `votes`
+(# מודלים) ו-`score` (ממוצע-המצביעים). **כלל-אישור:** 3 קולות→אישור · 2 וציון≥0.85→
+אישור · 2 ו<0.85→`pending_review` (יו"ר, G10) · ≤1→נדחה. **תקרה:** עד
+`HALACHA_PANEL_MAX_NEW`=5 עקרונות חדשים לכל החלטה (לפי ציון); עיקרון מוכר מקושר
+ל-canonical קיים (cosine, V41) ואינו נספר בתקרה. **טרמינולוגיה (מהות, לא קוסמטיקה):**
+ועדת-ערר **מיישמת** דין ואינה יוצרת הלכה — עיקרון מפס"ד מחוזי/עליון מחייב = **הלכה**,
+מהחלטת ועדת-ערר = **כלל פרשני**, מפסיקה משכנעת = **עיקרון**; המטרייה = **עקרונות
+משפטיים**. הסיווג נגזר מ-`first_established_in` (source_kind/is_binding), ללא עמודה חדשה.
+**מקור-יחיד (G2):** extractor (`_extract_via_panel`), סינון רטרואקטיבי (`cull_principles.py`),
+ושני הם דרך `services/panel_extraction` + `panel_judges` — אין נתיב-פאנל מקביל.
+**מקורות:** gold-set tri-model consensus (AC1=0.92, [[project_goldset_tri_model_consensus]]) ·
+LegalBench (gemini-2.5-flash) · Trust-or-Escalate (ICLR 2025) | סטטוס: verified
+**אכיפה:** `services/panel_extraction.py` (panel_extract/panel_keep_score/classify/apply_cap),
+`services/panel_judges.py`, `halacha_extractor._extract_via_panel`, `db.store_panel_principles`,
+`scripts/cull_principles.py`, `services/principles.py` (תווית). config `HALACHA_PANEL_*`.
+החלטת-יו"ר 2026-06-19; מקור-אמת: [`../legal-principles-redesign.md`](../legal-principles-redesign.md).
+**הפרה ידועה:** — (חדש)
+
 ---
 
 ## 4. הג'ובים המתוזמנים (תמיכת-תשתית ללולאה)
diff --git a/mcp-server/src/legal_mcp/services/db.py b/mcp-server/src/legal_mcp/services/db.py
index df5e2cc..75ad0c1 100644
--- a/mcp-server/src/legal_mcp/services/db.py
+++ b/mcp-server/src/legal_mcp/services/db.py
@@ -14,7 +14,7 @@ import asyncpg
 from pgvector.asyncpg import register_vector
 
 from legal_mcp import config
-from legal_mcp.services import court_citation, halacha_quality
+from legal_mcp.services import court_citation, halacha_quality, principles
 
 logger = logging.getLogger(__name__)
 
@@ -6156,7 +6156,8 @@ async def get_canonical_halacha(canonical_id: "UUID") -> "dict | None":
         "SELECT ch.id::text, ch.canonical_statement, ch.rule_type, "
         "       ch.practice_areas, ch.subject_tags, ch.review_status, "
         "       ch.instance_count, ch.created_at, ch.updated_at, "
-        "       cl.case_number AS first_established_case "
+        "       cl.case_number AS first_established_case, "
+        "       cl.source_kind, cl.is_binding "
         "FROM canonical_halachot ch "
         "LEFT JOIN case_law cl ON cl.id = ch.first_established_in "
         "WHERE ch.id = $1",
@@ -6172,10 +6173,12 @@ async def get_canonical_halacha(canonical_id: "UUID") -> "dict | None":
         "WHERE h.canonical_id = $1 ORDER BY h.instance_type, cl.case_number",
         canonical_id,
     )
-    return {
-        **dict(row),
-        "instances": [dict(i) for i in instances],
-    }
+    out = dict(row)
+    # #152: source-derived class + Hebrew label (הלכה / כלל פרשני / עיקרון).
+    out["principle_class"] = principles.principle_class(out.get("source_kind"), out.get("is_binding"))
+    out["principle_label"] = principles.label_for_class(out["principle_class"])
+    out["instances"] = [dict(i) for i in instances]
+    return out
 
 
 async def list_canonical_halachot(
@@ -6190,24 +6193,33 @@ async def list_canonical_halachot(
     params: list = []
     idx = 1
     if practice_area:
-        conditions.append(f"${ idx} = ANY(practice_areas)")
+        conditions.append(f"${ idx} = ANY(ch.practice_areas)")
         params.append(practice_area)
         idx += 1
     if review_status:
-        conditions.append(f"review_status = ${idx}")
+        conditions.append(f"ch.review_status = ${idx}")
         params.append(review_status)
         idx += 1
     params += [limit, offset]
     rows = await pool.fetch(
-        f"SELECT id::text, canonical_statement, rule_type, practice_areas, "
-        f"       subject_tags, review_status, instance_count, created_at, updated_at "
-        f"FROM canonical_halachot "
+        f"SELECT ch.id::text, ch.canonical_statement, ch.rule_type, ch.practice_areas, "
+        f"       ch.subject_tags, ch.review_status, ch.instance_count, "
+        f"       ch.created_at, ch.updated_at, cl.source_kind, cl.is_binding "
+        f"FROM canonical_halachot ch "
+        f"LEFT JOIN case_law cl ON cl.id = ch.first_established_in "
         f"WHERE {' AND '.join(conditions)} "
-        f"ORDER BY instance_count DESC, created_at DESC "
+        f"ORDER BY ch.instance_count DESC, ch.created_at DESC "
         f"LIMIT ${idx} OFFSET ${idx + 1}",
         *params,
     )
-    return [dict(r) for r in rows]
+    out = []
+    for r in rows:
+        d = dict(r)
+        cls = principles.principle_class(d.pop("source_kind", None), d.pop("is_binding", None))
+        d["principle_class"] = cls
+        d["principle_label"] = principles.label_for_class(cls)
+        out.append(d)
+    return out
 
 
 async def update_canonical_statement(
diff --git a/mcp-server/src/legal_mcp/services/panel_extraction.py b/mcp-server/src/legal_mcp/services/panel_extraction.py
index e93fa74..e18a659 100644
--- a/mcp-server/src/legal_mcp/services/panel_extraction.py
+++ b/mcp-server/src/legal_mcp/services/panel_extraction.py
@@ -126,6 +126,29 @@ def classify(votes: int, score: float) -> str:
     return "rejected"
 
 
+def apply_cap(judged: list[dict], max_new: int | None = None) -> list[dict]:
+    """Per-decision cap for the retroactive cull (#152, Phase C).
+
+    ``judged`` = a decision's principles, each with a panel ``verdict`` + ``score``.
+    Survivors (approved/pending_review) are ranked by score; those beyond ``max_new``
+    are downgraded to 'rejected' (over-cap). Already-rejected stay rejected. Returns
+    a new list with ``final_verdict`` set on each (order preserved). Pure.
+    """
+    max_new = config.HALACHA_PANEL_MAX_NEW if max_new is None else max_new
+    survivors = [j for j in judged if j.get("verdict") in ("approved", "pending_review")]
+    survivors.sort(key=lambda j: j.get("score", 0.0), reverse=True)
+    keep_ids = {id(j) for j in survivors[:max_new]}
+    out = []
+    for j in judged:
+        v = j.get("verdict")
+        if v in ("approved", "pending_review") and id(j) not in keep_ids:
+            final = "rejected"          # over the cap
+        else:
+            final = v
+        out.append({**j, "final_verdict": final})
+    return out
+
+
 def cluster_candidates(
     per_model: dict[str, list[dict]], embs: dict[int, list[float]],
 ) -> list[dict]:
@@ -195,6 +218,63 @@ def cluster_candidates(
     return out
 
 
+def _keep_score_system(source_kind: str, is_binding: bool) -> str:
+    if source_kind == "internal_committee":
+        nature = ("המקור הוא החלטת ועדת-ערר (מיישמת דין, אינה יוצרת הלכה). ראוי-לשמירה = "
+                  "כלל פרשני חדש ובר-הכללה שהוועדה גיבשה; לא-ראוי = יישום תלוי-עובדות, "
+                  "חזרה על דין מוכר, אמרת-אגב, או חזרה מילולית על הציטוט.")
+    else:
+        nature = ("ראוי-לשמירה = עיקרון משפטי בר-הכללה והסתמכות (הלכה/פרשנות/כלל-פרוצדורלי); "
+                  "לא-ראוי = החלה תלוית-עובדות, אמרת-אגב, או חזרה מילולית על הציטוט.")
+    return (
+        "אתה משפטן בכיר בוועדת ערר לתכנון ובנייה. הוכרע אם עיקרון שחולץ מפסיקה ראוי "
+        f"להישמר כתקדים בר-ציטוט. {nature}\n"
+        "תן גם ציון-ביטחון 0-1 לכך שזהו עיקרון בר-הסתמכות אמיתי.\n"
+        'החזר JSON בלבד: {"keep": true/false, "score": 0.0-1.0, "reason": "<משפט קצר>"}. ללא markdown.'
+    )
+
+
+async def panel_keep_score(
+    rule_statement: str,
+    supporting_quote: str,
+    reasoning_summary: str = "",
+    *,
+    source_kind: str = "external_upload",
+    is_binding: bool = True,
+) -> dict:
+    """Run the 3-judge panel on ONE existing principle (Phase C cull, #152).
+
+    Each judge votes keep + score; votes = # keepers, score = mean of the keepers'
+    scores (chaim: "ממוצע המצביעים"), verdict via the shared :func:`classify`.
+    Returns {votes, score, verdict, voters, per_judge} — per_judge keeps raw
+    replies for the active-learning round (FU-1). Used by the retroactive cull;
+    the extractor uses :func:`panel_extract` instead.
+    """
+    import asyncio
+    system = _keep_score_system(source_kind, is_binding)
+    user = (f"ניסוח העיקרון:\n{rule_statement}\n\n"
+            f"היגיון:\n{reasoning_summary}\n\nציטוט תומך:\n{supporting_quote}")
+    async with httpx.AsyncClient() as client:
+        c, ds, gm = await asyncio.gather(
+            panel_judges.judge_claude(system, user, max_tokens=300),
+            panel_judges.judge_deepseek(client, system, user, max_tokens=300),
+            panel_judges.judge_gemini(client, system, user, max_tokens=2000),
+        )
+    raw = {"claude": c, "deepseek": ds, "gemini": gm}
+    keepers, scores = [], []
+    for name, reply in raw.items():
+        if panel_judges.to_bool(reply, "keep"):
+            keepers.append(name)
+            try:
+                scores.append(max(0.0, min(1.0, float(reply.get("score", 0.0)))))
+            except (TypeError, ValueError):
+                scores.append(0.0)
+    votes = len(keepers)
+    score = round(sum(scores) / votes, 4) if votes else 0.0
+    return {"votes": votes, "score": score, "verdict": classify(votes, score),
+            "voters": sorted(keepers), "per_judge": raw}
+
+
 async def _run_three(system: str, user: str, max_tokens: int) -> dict[str, object]:
     async with httpx.AsyncClient() as client:
         import asyncio
diff --git a/mcp-server/src/legal_mcp/services/principles.py b/mcp-server/src/legal_mcp/services/principles.py
new file mode 100644
index 0000000..a75367b
--- /dev/null
+++ b/mcp-server/src/legal_mcp/services/principles.py
@@ -0,0 +1,45 @@
+"""Legal-principles terminology — the single source for what a principle is CALLED (#152).
+
+chaim 2026-06-19: "הלכה" was the wrong umbrella. The corpus holds **עקרונות
+משפטיים** (legal principles); the term for one depends on its SOURCE:
+
+  • binding higher court (מחוזי/עליון)      → "הלכה"          (binding precedent)
+  • appeals committee (internal_committee)  → "כלל פרשני"     (interpretive rule —
+                                              the committee applies law, never makes it)
+  • non-binding external (persuasive)       → "עיקרון"        (persuasive principle)
+
+The class is derived from where a principle was FIRST established
+(canonical_halachot.first_established_in → case_law.source_kind/is_binding), so no
+new column is needed. UI/tools call :func:`label` instead of hardcoding "הלכה".
+"""
+from __future__ import annotations
+
+UMBRELLA = "עקרונות משפטיים"
+
+CLASS_HALACHA = "halacha"
+CLASS_INTERPRETIVE_RULE = "interpretive_rule"
+CLASS_PRINCIPLE = "principle"
+
+_LABEL = {
+    CLASS_HALACHA: "הלכה",
+    CLASS_INTERPRETIVE_RULE: "כלל פרשני",
+    CLASS_PRINCIPLE: "עיקרון",
+}
+
+
+def principle_class(source_kind: str | None, is_binding: bool | None) -> str:
+    """Map a source to its principle class (stable key, not display text)."""
+    if source_kind == "internal_committee":
+        return CLASS_INTERPRETIVE_RULE
+    if is_binding:
+        return CLASS_HALACHA
+    return CLASS_PRINCIPLE
+
+
+def label(source_kind: str | None, is_binding: bool | None) -> str:
+    """Hebrew display term for a principle from this source (#152)."""
+    return _LABEL[principle_class(source_kind, is_binding)]
+
+
+def label_for_class(cls: str) -> str:
+    return _LABEL.get(cls, _LABEL[CLASS_PRINCIPLE])
diff --git a/mcp-server/src/legal_mcp/tools/precedent_library.py b/mcp-server/src/legal_mcp/tools/precedent_library.py
index 5722378..4c3d87c 100644
--- a/mcp-server/src/legal_mcp/tools/precedent_library.py
+++ b/mcp-server/src/legal_mcp/tools/precedent_library.py
@@ -407,7 +407,10 @@ async def canonical_halacha_list(
     limit: int = 50,
     offset: int = 0,
 ) -> str:
-    """רשימת עקרונות קנוניים (canonical_halachot) — שאילתת נוחות לסוכני-הכתיבה.
+    """רשימת עקרונות משפטיים קנוניים — שאילתת נוחות לסוכני-הכתיבה.
+
+    כל פריט כולל principle_label לפי מקורו (#152): 'הלכה' (פס"ד מחוזי/עליון מחייב),
+    'כלל פרשני' (החלטת ועדת-ערר), או 'עיקרון' (פסיקה משכנעת).
 
     Args:
         practice_area: סינון לפי תחום עיסוק (ריק = הכל).
diff --git a/mcp-server/tests/test_panel_extraction.py b/mcp-server/tests/test_panel_extraction.py
index ae070b0..30e387f 100644
--- a/mcp-server/tests/test_panel_extraction.py
+++ b/mcp-server/tests/test_panel_extraction.py
@@ -106,6 +106,25 @@ def test_cluster_same_model_twice_counts_one_vote_keeps_best_score():
     assert cl["rule_statement"] == "X"
 
 
+def test_apply_cap_downgrades_over_cap_survivors_by_score():
+    judged = [
+        {"verdict": "approved", "score": 0.9},
+        {"verdict": "approved", "score": 0.7},
+        {"verdict": "pending_review", "score": 0.8},
+        {"verdict": "rejected", "score": 0.95},   # already rejected stays
+    ]
+    out = pe.apply_cap(judged, max_new=2)
+    fv = [j["final_verdict"] for j in out]
+    # top-2 survivors by score = 0.9(approved) + 0.8(pending); 0.7 → over cap → rejected
+    assert fv == ["approved", "rejected", "pending_review", "rejected"]
+
+
+def test_apply_cap_keeps_all_when_under_cap():
+    judged = [{"verdict": "approved", "score": 0.9}, {"verdict": "pending_review", "score": 0.5}]
+    out = pe.apply_cap(judged, max_new=5)
+    assert [j["final_verdict"] for j in out] == ["approved", "pending_review"]
+
+
 def test_cluster_sorted_strongest_first():
     a = _c("X", 0.9)   # 1 vote
     b, c = _c("Y", 0.9), _c("Y", 0.9)  # 2 votes
diff --git a/mcp-server/tests/test_principles_terminology.py b/mcp-server/tests/test_principles_terminology.py
new file mode 100644
index 0000000..652619c
--- /dev/null
+++ b/mcp-server/tests/test_principles_terminology.py
@@ -0,0 +1,27 @@
+"""Terminology mapping — הלכה / כלל פרשני / עיקרון by source (#152, Phase D)."""
+from __future__ import annotations
+
+from legal_mcp.services import principles as pr
+
+
+def test_binding_higher_court_is_halacha():
+    assert pr.label("external_upload", True) == "הלכה"
+    assert pr.principle_class("external_upload", True) == pr.CLASS_HALACHA
+
+
+def test_committee_is_interpretive_rule():
+    # the appeals committee applies law — never makes a הלכה
+    assert pr.label("internal_committee", True) == "כלל פרשני"
+    assert pr.label("internal_committee", False) == "כלל פרשני"
+    assert pr.principle_class("internal_committee", False) == pr.CLASS_INTERPRETIVE_RULE
+
+
+def test_non_binding_external_is_principle():
+    assert pr.label("external_upload", False) == "עיקרון"
+    assert pr.label(None, None) == "עיקרון"
+
+
+def test_label_for_class_roundtrip():
+    for sk, binding in [("external_upload", True), ("internal_committee", False), (None, False)]:
+        cls = pr.principle_class(sk, binding)
+        assert pr.label_for_class(cls) == pr.label(sk, binding)
diff --git a/scripts/SCRIPTS.md b/scripts/SCRIPTS.md
index a945c7b..26549df 100644
--- a/scripts/SCRIPTS.md
+++ b/scripts/SCRIPTS.md
@@ -65,6 +65,7 @@
 | `halacha_panel_calibrate.py` | python | **כיול + מדידת הפאנל** (Trust-or-Escalate, ICLR 2025). `--source live` (ברירת-מחדל): מריץ את שאלת-ה-KEEP על מדגם-הזהב ומודד מול `is_holding` precision+coverage+**split-rate** לכל מדיניות + false-keep/false-drop (מייבא שופטים מ-`halacha_panel_approve`, **חובה מקומי**). **#133/FU-5** — `--source captured`: **אפס-עלות** (בלי re-vote/LLM) — מצליב סבבים שמורים (FU-1) מול הכרעות-יו"ר (FU-2) דרך `db.panel_rounds_vs_chair` ומדווח split-rate+auto-precision **לכל סבב** (מגמת הלולאה: ככל שהרובריקה משתפרת precision נשמר ו-split יורד); משתף את `analyze_pairs` של FU-4 (מקור-יחיד). שתי המדידות מדווחות **anon-stability** (מבחן-אנונימיזציה #81.7) כמטריקת-בריאות נגד echo-chamber. `--batch`/`--limit`/`--concurrency`. | ידני — לפני חיווט `--apply` (live) / תקופתי — מעקב-לולאה (captured) |
 | `halacha_rubric_distill.py` | python | **#133/FU-4 — זיקוק-רובריקה PROPOSE-ONLY.** מצליב `halacha_panel_rounds` (FU-1, הצבעות+נימוקים) מול הכרעות-היו"ר (FU-2, seeds ב-`halacha_goldset` batch `chair-live`) דרך `db.panel_rounds_vs_chair` (read-only), מנתח דטרמיניסטית **כשלים שיטתיים** (false-keep/false-drop, פיצולים-שהוכרעו, שיעור-מחלוקת-עם-היו"ר לכל שופט), ומציע `KEEP_SYSTEM` v2 + exemplars מופשטים (claude_session מקומי, אפס עלות) כ**דוח-diff** ל-`data/learning/rubric-proposal-<ts>.md`. **לעולם לא auto-apply** — אימוץ v2 = עריכה אנושית של הקבוע דרך PR (INV-LRN1); exemplars מופשטים בלבד (INV-LRN5); הסיגנל היחיד = הכרעת-יו"ר, לא הצבעות-פאנל (anti-echo). מתחת ל-12 זוגות → "אין מספיק נתונים". `--no-llm` (סטטיסטיקה בלבד) / `--limit N`. **חובה מקומי**. | תקופתי — אחרי שהצטברו הכרעות-יו"ר על מחלוקות-פאנל |
 | `backfill_canonical_halachot.py` | python | **V41 — הקמת מודל ההלכות הקנוניות (חד-פעמי + idempotent).** (1) בונה רכיבים-קשורים (connected components) מ-`equivalent_halachot` (transitive closure — union-find). (2) לכל אשכול: בוחר נציג-קנוני (הכי הרבה corroboration → confidence → earliest), יוצר שורת `canonical_halachot`, ומעדכן `canonical_id` + `instance_type` לכל חברי האשכול. (3) לסינגלטונים (ללא קישורי-שוויון): 1:1 canonical. (4) מאכלס `halacha_citation_corroboration.canonical_id` מ-`halachot.canonical_id`. `--dry-run` (ברירת-מחדל, מחשב ומדווח בלבד) / `--apply` (כותב) / `--verbose`. לאחר הרצה: `canonical_statement` = ניסוח-נציג (pending_synthesis); עוקב: `backfill_canonical_synthesis.py` (Phase 4) יסנתז ניסוח-רחב דרך LLM. הרץ: `mcp-server/.venv/bin/python scripts/backfill_canonical_halachot.py --apply`. | **חד-פעמי** (לאחר deploy V41) / idempotent לפי צורך |
+| `cull_principles.py` | python | **#152 Phase C — סינון רטרואקטיבי של קורפוס-העקרונות דרך פאנל-3 (הפיך).** מריץ על כל עיקרון 'original' קיים את אותו משטר שה-extractor משתמש בו להבא (`services/panel_extraction.panel_keep_score`, G2): 3 שופטים (Claude מקומי + DeepSeek + Gemini) מצביעים keep+score → כלל-האישור (3 קולות→שורד · 2 וציון≥0.85→שורד · 2 ו<0.85→יו"ר · ≤1→נדחה) → תקרת `HALACHA_PANEL_MAX_NEW`=5 לכל החלטה לפי ציון (`apply_cap`). נדחה → `halachot.review_status='rejected'` + ה-canonical שלו `rejected` (הפיך, גיבוי-CSV ב-`data/audit/` לפני כל כתיבה). מרוסן ב-`usage_limits` (עוצר-רך בתקרת-שימוש, resumable). `--dry-run` (ברירת-מחדל) / `--apply` / `--sample N` (החלטות אקראיות) / `--limit N` / `--no-throttle` / `--verbose`. **חובה מקומי** (3 שופטים). הרץ: `cd mcp-server && HOME=/home/chaim .venv/bin/python ../scripts/cull_principles.py --apply`. | **חד-פעמי** (סינון ראשוני) + ניתן-לחזרה |
 | `backfill_canonical_synthesis.py` | python | **V41 Phase 4 — סינתזת-LLM ל-`canonical_statement` (idempotent + resumable).** עובר על canonicals ב-`review_status='pending_synthesis'` (רב-instance ראשונים) ומזקק לכל אחד ניסוח אחד כללי ומעוגן בציטוטי-המופעים (INV-AH) דרך `services/canonical_synthesis.py` (מסלול-יחיד, G2). שערים: עיגון/הימנעות, **drift-floor** (cosine מול המקור, ברירת-מחדל 0.80 — סטייה גדולה→נשמר המקור), ואיסור ציטוטי-תיק חדשים. בכל מקרה הסטטוס מתקדם ל-`pending_review` לשער-היו"ר (G10/INV-LRN6). מודל Opus (`HALACHA_CANONICAL_SYNTH_MODEL`). מרוסן ע"י `usage_limits` (עוצר-רך בתקרת-שימוש, resumable). `--dry-run` (ברירת-מחדל) / `--apply` / `--sample N` (מדגם אקראי לבדיקה) / `--limit N` / `--no-throttle` / `--verbose`. CSV-audit ל-`data/audit/canonical-synthesis-*.csv`. **חובה מקומי** (claude_session). הרץ: `cd mcp-server && HOME=/home/chaim .venv/bin/python ../scripts/backfill_canonical_synthesis.py --apply`. שוטף: כלי-MCP `canonical_synthesize_pending`. | **חד-פעמי** (המסה הראשונית) + idempotent לחדשים |
 | `halacha_batch_reconcile.py` | python | **#82.7** — dedup חוצה-פסקים offline (שמרני, **dry-run בלבד**). dedup-on-insert משווה רק תוך-פסק; כאן סף מחמיר (cosine ≥0.95, `--cosine`) ולא-הרסני: מאתר זוגות הלכות near-duplicate בין פסקים שונים (pgvector `<=>` exact) עם איתות לקסיקלי (Jaccard/Levenshtein) ומדווח ל-CSV ב-`data/audit/` לסקירת היו"ר. לא מדלג/ממזג/מוחק. `--include-pending`. **`--link`** רושם את הזוגות שנמצאו כ-`equivalent_halachot` (parallel authority, #84.2 — **deprecated post-V41** — השתמש ב-`backfill_canonical_halachot.py --apply` במקום). רץ עם venv של mcp-server. | **deprecated** — הוחלף ב-`backfill_canonical_halachot.py` (V41). נשמר לצורכי audit |
 | `calibrate_halacha_dedup.py` | python | **#82.1** — כיול ספי ה-dedup הלקסיקלי (#82.3) מול gold-set הניקוי. קורא `halacha-cleanup-manifest-*.csv` (זוגות duplicate↔survivor מתויגי-אדם), טוען טקסט-survivor מה-DB, ו-sweep של (jaccard_min × levenshtein_min) עם P/R/F1, מסמן את נקודת-העבודה המוגדרת. אימת ש-(0.55, 0.70) → **precision 1.0** (אפס false-merge), recall 0.30 — מתאים לאיתות-משני שחוסם auto-approve. `--manifest <path>`. רץ עם venv של mcp-server | חד-פעמי — כיול (בוצע 2026-06-06) |
diff --git a/scripts/cull_principles.py b/scripts/cull_principles.py
new file mode 100644
index 0000000..0b13ddc
--- /dev/null
+++ b/scripts/cull_principles.py
@@ -0,0 +1,204 @@
+#!/usr/bin/env python3
+"""Retroactive cull of the legal-principles corpus via the 3-model panel (#152, Phase C).
+
+The corpus grew to ~5,243 principles (18.8/decision) under the old single-model
+auto-approve. This re-adjudicates EVERY existing 'original' principle with the
+SAME regime the extractor now uses going forward (chaim 2026-06-19):
+
+  • 3 judges (Claude local + DeepSeek + Gemini) vote keep + score on each principle.
+  • Approval rule: 3 votes→survive · 2 & score≥0.85→survive · 2 & <0.85→chair
+    (pending_review) · ≤1→reject.
+  • Per DECISION, survivors are capped to HALACHA_PANEL_MAX_NEW (=5) by score; the
+    rest are rejected (over-cap).
+
+All logic is shared with the extractor via services/panel_extraction (G2). The
+cull is REVERSIBLE: a CSV backup of every (id, old_status) is written before any
+write, and a rejected principle's canonical is also set 'rejected' (recoverable).
+Throttled by usage_limits (stops gracefully at the soft ceiling, resumable).
+
+  cd ~/legal-ai/mcp-server
+  HOME=/home/chaim .venv/bin/python ../scripts/cull_principles.py --sample 5     # dry-run, 5 decisions
+  HOME=/home/chaim .venv/bin/python ../scripts/cull_principles.py --dry-run      # all, dry-run
+  HOME=/home/chaim .venv/bin/python ../scripts/cull_principles.py --apply        # full, throttled
+"""
+from __future__ import annotations
+
+import argparse
+import asyncio
+import csv
+import os
+import random
+import sys
+from collections import Counter
+from datetime import datetime, timezone
+from uuid import UUID
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "mcp-server", "src"))
+
+from legal_mcp import config  # noqa: E402
+from legal_mcp.services import db, panel_extraction as pe  # noqa: E402
+
+try:
+    from legal_mcp.services import usage_limits
+except Exception:  # pragma: no cover
+    usage_limits = None
+
+AUDIT_DIR = os.path.join(os.path.dirname(__file__), "..", "data", "audit")
+_JUDGE_CONCURRENCY = 4
+
+
+async def _decisions(limit, sample):
+    """case_law ids that have 'original' principles, with source metadata."""
+    pool = await db.get_pool()
+    rows = await pool.fetch(
+        "SELECT cl.id, cl.case_number, cl.source_kind, cl.is_binding, "
+        "       count(*) AS n "
+        "FROM halachot h JOIN case_law cl ON cl.id = h.case_law_id "
+        "WHERE h.instance_type = 'original' AND h.review_status <> 'rejected' "
+        "GROUP BY cl.id, cl.case_number, cl.source_kind, cl.is_binding "
+        "ORDER BY n DESC",
+    )
+    items = [dict(r) for r in rows]
+    if sample and sample < len(items):
+        items = random.sample(items, sample)
+    if limit:
+        items = items[:limit]
+    return items
+
+
+async def _principles(case_law_id):
+    pool = await db.get_pool()
+    rows = await pool.fetch(
+        "SELECT id, rule_statement, supporting_quote, reasoning_summary, "
+        "       canonical_id, review_status "
+        "FROM halachot WHERE case_law_id = $1 AND instance_type = 'original' "
+        "AND review_status <> 'rejected' ORDER BY halacha_index",
+        case_law_id,
+    )
+    return [dict(r) for r in rows]
+
+
+def _throttled():
+    if usage_limits is None:
+        return False, "no usage_limits"
+    u = usage_limits.subscription_usage()
+    if u is None:
+        return False, "usage read failed"
+    over, _r, detail = usage_limits.ceiling_status(u)
+    return over, detail
+
+
+async def _judge_decision(dec, sem):
+    principles = await _principles(dec["id"])
+    if not principles:
+        return []
+
+    async def one(p):
+        async with sem:
+            v = await pe.panel_keep_score(
+                p["rule_statement"], p["supporting_quote"], p.get("reasoning_summary") or "",
+                source_kind=dec["source_kind"] or "external_upload",
+                is_binding=bool(dec["is_binding"]),
+            )
+        return {**p, **v}
+
+    judged = await asyncio.gather(*[one(p) for p in principles])
+    return pe.apply_cap(list(judged))
+
+
+async def _apply_decision(judged, reviewer):
+    pool = await db.get_pool()
+    async with pool.acquire() as conn:
+        async with conn.transaction():
+            for j in judged:
+                fv = j["final_verdict"]
+                if fv == "approved":
+                    await conn.execute(
+                        "UPDATE halachot SET review_status='approved', reviewed_at=now(), "
+                        "reviewer=$2, updated_at=now() WHERE id=$1", j["id"], reviewer)
+                elif fv == "pending_review":
+                    await conn.execute(
+                        "UPDATE halachot SET review_status='pending_review', reviewer=$2, "
+                        "updated_at=now() WHERE id=$1", j["id"], reviewer)
+                else:  # rejected — also reject its canonical (reversible)
+                    await conn.execute(
+                        "UPDATE halachot SET review_status='rejected', reviewed_at=now(), "
+                        "reviewer=$2, updated_at=now() WHERE id=$1", j["id"], reviewer)
+                    if j.get("canonical_id"):
+                        await conn.execute(
+                            "UPDATE canonical_halachot SET review_status='rejected', "
+                            "updated_at=now() WHERE id=$1", j["canonical_id"])
+
+
+async def _run(apply, limit, sample, throttle, verbose):
+    decisions = await _decisions(limit, sample)
+    mode = "APPLY" if apply else "DRY-RUN"
+    print(f"[{mode}] {len(decisions)} decisions with principles "
+          f"(throttle={'on' if throttle else 'off'})\n", flush=True)
+    if not decisions:
+        print("nothing to do.")
+        return 0
+
+    stamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
+    os.makedirs(AUDIT_DIR, exist_ok=True)
+    audit = os.path.join(AUDIT_DIR, f"principle-cull-{'apply' if apply else 'dryrun'}-{stamp}.csv")
+    reviewer = f"cull:panel v{config.HALACHA_PANEL_SCORE_FLOOR} cap{config.HALACHA_PANEL_MAX_NEW}"
+    sem = asyncio.Semaphore(_JUDGE_CONCURRENCY)
+    tally = Counter()
+    n_in = n_out = 0
+    stopped = False
+
+    with open(audit, "w", newline="", encoding="utf-8") as fh:
+        w = csv.writer(fh)
+        w.writerow(["case_number", "halacha_id", "old_status", "final_verdict",
+                    "votes", "score", "canonical_id", "rule"])
+        for k, dec in enumerate(decisions, 1):
+            if throttle:
+                over, detail = _throttled()
+                if over:
+                    print(f"\n⏸  usage ceiling ({detail}) — stopping at {k-1}/{len(decisions)}. "
+                          f"Re-run to resume.", flush=True)
+                    stopped = True
+                    break
+            judged = await _judge_decision(dec, sem)
+            survivors = sum(1 for j in judged if j["final_verdict"] in ("approved", "pending_review"))
+            n_in += len(judged)
+            n_out += survivors
+            for j in judged:
+                tally[j["final_verdict"]] += 1
+                w.writerow([dec["case_number"], str(j["id"]), j["review_status"],
+                            j["final_verdict"], j["votes"], j["score"],
+                            str(j.get("canonical_id") or ""), (j["rule_statement"] or "")[:160]])
+            if apply and judged:
+                await _apply_decision(judged, reviewer)
+            print(f"[{k}/{len(decisions)}] {dec['case_number']:<16} "
+                  f"{len(judged)}→{survivors} survive", flush=True)
+            if verbose:
+                for j in judged:
+                    mark = {"approved": "✓", "pending_review": "→chair", "rejected": "✗"}[j["final_verdict"]]
+                    print(f"      {mark} v={j['votes']} s={j['score']} {(j['rule_statement'] or '')[:80]}")
+
+    print(f"\n── {mode} summary{' (stopped early)' if stopped else ''} ──")
+    print(f"   principles judged: {n_in} → survive: {n_out}  ({n_in - n_out} rejected)")
+    for v, c in tally.most_common():
+        print(f"   {v:<16} {c}")
+    print(f"\naudit CSV: {audit}")
+    if not apply:
+        print("dry-run — no DB writes. Re-run with --apply to commit (reversible).")
+    return 0
+
+
+def main():
+    p = argparse.ArgumentParser(description="Retroactive principle cull via 3-model panel (#152)")
+    p.add_argument("--apply", action="store_true", help="write verdicts (reversible, CSV-backed)")
+    p.add_argument("--dry-run", action="store_true", help="explicit dry-run (default)")
+    p.add_argument("--limit", type=int, default=None)
+    p.add_argument("--sample", type=int, default=None, help="random sample of N decisions")
+    p.add_argument("--no-throttle", action="store_true")
+    p.add_argument("--verbose", action="store_true")
+    a = p.parse_args()
+    return asyncio.run(_run(a.apply, a.limit, a.sample, not a.no_throttle, a.verbose))
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())