diff --git a/docs/changelog.d/rag-facets-calibration.md b/docs/changelog.d/rag-facets-calibration.md
new file mode 100644
index 000000000..813661265
--- /dev/null
+++ b/docs/changelog.d/rag-facets-calibration.md
@@ -0,0 +1,13 @@
+# Governed RAG facets calibration
+
+## Added
+
+- Added the governed RAG facets calibration adapter.
+- The adapter reuses the existing MFRM rater-severity/threshold calibration
+  machinery (Linacre, 1989; Eckes, 2015; Bock & Aitkin, 1981; Andrich, 1978)
+  for RAG evaluation executions. It does not introduce a new psychometric
+  estimator or rely on legacy evaluation package implementations (e.g.
+  RAGAS-style tooling) as the source of psychometric validity; all
+  likelihood/threshold arithmetic is delegated to the existing Rust-backed MFRM
+  fit grounded in the primary many-facet Rasch measurement literature. Full
+  citations are in `docs/scoring_facets_calibration_handoff.md`.
diff --git a/python/fast_mlsirm/scoring/rag_calibration.py b/python/fast_mlsirm/scoring/rag_calibration.py
new file mode 100644
index 000000000..56786d5fa
--- /dev/null
+++ b/python/fast_mlsirm/scoring/rag_calibration.py
@@ -0,0 +1,111 @@
+"""Project governed RAG scoring executions into shared many-facet designs.
+
+This adapter replays the package-managed RAG request provenance before handing
+criterion observations to the existing scoring calibration contracts. Python
+only validates identities and marshals records. All many-facet likelihood,
+threshold, estimation, optimization, and uncertainty arithmetic remains owned
+by the existing Rust-backed calibration path.
+"""
+
+from __future__ import annotations
+
+from collections.abc import Iterable
+
+from ._contract_safety import bounded_values
+from ._validation import assessment_error
+from .calibration import (
+    MAX_SCORING_FACETS_RATINGS,
+    ScoringFacetsCalibrationBundle,
+    ScoringFacetsRatingRecord,
+    build_scoring_facets_calibration_bundle,
+    build_scoring_facets_rating_records,
+)
+from .execution import EngineDescriptor, ScoringRequest, ScoringResult
+from .rag import _canonical_rag_request
+
+MAX_RAG_CALIBRATION_EXECUTIONS = MAX_SCORING_FACETS_RATINGS
+"""Maximum governed RAG executions accepted by one facets bundle assembly."""
+
+
+def build_rag_facets_rating_records(
+    *,
+    request: ScoringRequest,
+    result: ScoringResult,
+    engine: EngineDescriptor,
+) -> tuple[ScoringFacetsRatingRecord, ...]:
+    """Project one provenance-complete RAG execution into shared facets records.
+
+    The RAG request must contain the package-managed evidence-regime, candidate
+    visibility, system-configuration, retrieval-run, and query-revision
+    provenance produced by :func:`fast_mlsirm.scoring.rag.build_rag_scoring_request`.
+    The shared projector remains authoritative for exact request/result/engine
+    binding, terminal observation states, criterion scales, and rater identity.
+
+    Passing this boundary establishes replayable measurement provenance only.
+    It does not establish factual truth, retrieval recall, model adequacy,
+    validity, fairness, invariance, or a buyer-facing quality conclusion.
+    """
+    normalized_request = _canonical_rag_request(request, "rag_scoring_request")
+    return build_scoring_facets_rating_records(
+        request=normalized_request,
+        result=result,
+        engine=engine,
+    )
+
+
+def build_rag_facets_calibration_bundle(
+    executions: Iterable[tuple[ScoringRequest, ScoringResult, EngineDescriptor]],
+    *,
+    require_connected: bool = True,
+) -> ScoringFacetsCalibrationBundle:
+    """Assemble governed RAG executions into the existing many-facet bundle.
+
+    Each execution is an exact three-value tuple containing a canonical RAG
+    scoring request, its bound scoring result, and the producing engine
+    descriptor. Every request is replay-validated as RAG provenance before its
+    observations are projected into the shared criterion-separated design.
+
+    The returned value is the existing :class:`ScoringFacetsCalibrationBundle`;
+    this module introduces no parallel psychometric result hierarchy and no new
+    numerical estimator. System-run identity occupies the respondent axis,
+    query revision occupies the task-revision axis, and engine fingerprint
+    occupies the rater axis, preserving the hierarchy already encoded by the
+    canonical RAG scoring request.
+    """
+    values = bounded_values(
+        executions,
+        "executions",
+        minimum=1,
+        maximum=MAX_RAG_CALIBRATION_EXECUTIONS,
+    )
+
+    def validated_records() -> Iterable[ScoringFacetsRatingRecord]:
+        """Yield replay-validated RAG records for bounded shared assembly."""
+        for index, execution in enumerate(values):
+            if type(execution) is not tuple or len(execution) != 3:
+                raise assessment_error(
+                    "invalid_rag_calibration_execution",
+                    f"$.executions[{index}]",
+                    (
+                        "each execution must be an exact three-value tuple of "
+                        "request, result, and engine"
+                    ),
+                )
+            request, result, engine = execution
+            yield from build_rag_facets_rating_records(
+                request=request,
+                result=result,
+                engine=engine,
+            )
+
+    return build_scoring_facets_calibration_bundle(
+        validated_records(),
+        require_connected=require_connected,
+    )
+
+
+__all__ = [
+    "MAX_RAG_CALIBRATION_EXECUTIONS",
+    "build_rag_facets_calibration_bundle",
+    "build_rag_facets_rating_records",
+]
diff --git a/tests/test_scoring_rag_facets_calibration.py b/tests/test_scoring_rag_facets_calibration.py
new file mode 100644
index 000000000..3b1f76deb
--- /dev/null
+++ b/tests/test_scoring_rag_facets_calibration.py
@@ -0,0 +1,207 @@
+"""Many-facet orchestration contracts for governed RAG scoring executions."""
+
+from __future__ import annotations
+
+import hashlib
+import inspect
+from pathlib import Path
+import runpy
+
+import pytest
+
+from fast_mlsirm.scoring import (
+    AssessmentSpecError,
+    FixtureOutcome,
+    ObservationStatus,
+    ScoringFacetsCalibrationBundle,
+    ScoringFacetsRatingRecord,
+    StaticFixtureEngine,
+)
+from fast_mlsirm.scoring.rag import build_rag_scoring_request
+from fast_mlsirm.scoring.rag_calibration import (
+    build_rag_facets_calibration_bundle,
+    build_rag_facets_rating_records,
+)
+
+_FIXTURES = runpy.run_path(
+    str(Path(__file__).with_name("scoring_execution_fixtures.py"))
+)
+assessment = _FIXTURES["assessment"]
+rubric = _FIXTURES["rubric"]
+automated_engine = _FIXTURES["automated_engine"]
+criterion_request = _FIXTURES["criterion_request"]
+
+QUERY_REVISION_FPS = tuple(
+    hashlib.sha256(f"rag-calibration-query-revision-{index:03d}".encode()).hexdigest()
+    for index in (1, 2)
+)
+SYSTEM_FP = hashlib.sha256(b"rag-calibration-system").hexdigest()
+
+
+def _execution(
+    respondent_index: int = 1,
+    task_index: int = 1,
+    engine_id: str = "fixture_engine",
+):
+    """Return one deterministic governed RAG request/result/engine execution."""
+    respondent_suffix = f"{respondent_index:03d}"
+    task_suffix = f"{task_index:03d}"
+    scenario_index = (respondent_index - 1) * 2 + (task_index - 1)
+    grounded_scores = (2, 1, 0, 2)
+    relevance_scores = (1, 2, 0, 1)
+    request = build_rag_scoring_request(
+        request_id=(
+            f"rag_calibration_request_{respondent_suffix}_{task_suffix}"
+        ),
+        assessment=assessment(),
+        rubric=rubric(),
+        query_id="refund_policy_query",
+        query_revision_fingerprint=QUERY_REVISION_FPS[task_index - 1],
+        query_testlet_id="evidence_review",
+        evidence_regime="retrieved_context",
+        candidate_visibility="candidate_blind",
+        system_configuration_id="retrieval_stack_a",
+        system_configuration_fingerprint=SYSTEM_FP,
+        system_run_id=f"retrieval_stack_a_run_{respondent_suffix}",
+        response_id=f"generated_response_{respondent_suffix}_{task_suffix}",
+        retrieval_run_fingerprint=hashlib.sha256(
+            f"rag-calibration-retrieval-{respondent_suffix}-{task_suffix}".encode()
+        ).hexdigest(),
+        response_content_fingerprint=hashlib.sha256(
+            f"rag-calibration-response-{respondent_suffix}-{task_suffix}".encode()
+        ).hexdigest(),
+        occasion_id="evaluation_wave_001",
+        criterion_ids=("grounded_generation", "answer_relevance"),
+        response_character_count=412 + scenario_index,
+        response_unit_count=7,
+        metadata={"evaluation_split": "offline_holdout"},
+    )
+    engine = automated_engine(engine_id=engine_id)
+    fixture = StaticFixtureEngine(
+        descriptor=engine,
+        outcomes=(
+            FixtureOutcome(
+                criterion_id="grounded_generation",
+                status=ObservationStatus.SCORED,
+                score_category=grounded_scores[scenario_index],
+            ),
+            FixtureOutcome(
+                criterion_id="answer_relevance",
+                status=ObservationStatus.SCORED,
+                score_category=relevance_scores[scenario_index],
+            ),
+        ),
+    )
+    return request, fixture.score(request), engine
+
+
+def test_rag_facets_projection_reuses_shared_calibration_contracts() -> None:
+    """RAG provenance projects into the shared facets record hierarchy only."""
+    request, result, engine = _execution()
+
+    records = build_rag_facets_rating_records(
+        request=request,
+        result=result,
+        engine=engine,
+    )
+
+    assert all(type(record) is ScoringFacetsRatingRecord for record in records)
+    assert {record.criterion_id for record in records} == {
+        "grounded_generation",
+        "answer_relevance",
+    }
+    assert {record.respondent_id for record in records} == {request.respondent_id}
+    assert {record.task_id for record in records} == {request.task_id}
+    assert {record.task_revision_fingerprint for record in records} == {
+        request.task_revision_fingerprint
+    }
+    assert {record.engine_fingerprint for record in records} == {
+        engine.engine_fingerprint
+    }
+
+
+def test_rag_facets_bundle_delegates_to_shared_many_facet_design() -> None:
+    """RAG orchestration returns one connected shared many-facet design."""
+    executions = (
+        _execution(1, 1, "fixture_engine"),
+        _execution(1, 2, "alternate_engine"),
+        _execution(2, 1, "alternate_engine"),
+        _execution(2, 2, "fixture_engine"),
+    )
+
+    bundle = build_rag_facets_calibration_bundle(executions)
+
+    assert type(bundle) is ScoringFacetsCalibrationBundle
+    assert bundle.criterion_ids == ("answer_relevance", "grounded_generation")
+    expected_revisions = tuple(sorted(QUERY_REVISION_FPS))
+    for design in bundle.designs:
+        assert design.respondent_ids == (
+            "retrieval_stack_a_run_001",
+            "retrieval_stack_a_run_002",
+        )
+        assert design.task_revision_fingerprints == expected_revisions
+        assert design.task_ids == (
+            "refund_policy_query",
+            "refund_policy_query",
+        )
+        assert set(design.rater_engine_ids) == {
+            "alternate_engine",
+            "fixture_engine",
+        }
+        assert design.respondent_task_connected
+        assert design.task_rater_connected
+        assert design.connected
+        assert {
+            record.score_category
+            for record in design.rating_records
+            if record.status is ObservationStatus.SCORED
+        } == {0, 1, 2}
+
+
+def test_rag_facets_projection_rejects_non_rag_request() -> None:
+    """A generic scoring execution cannot masquerade as governed RAG evidence."""
+    generic_request = criterion_request()
+    engine = automated_engine()
+    result = StaticFixtureEngine(
+        descriptor=engine,
+        outcomes=(
+            FixtureOutcome(
+                criterion_id="claim_support",
+                status=ObservationStatus.SCORED,
+                score_category=2,
+            ),
+            FixtureOutcome(
+                criterion_id="source_alignment",
+                status=ObservationStatus.SCORED,
+                score_category=1,
+            ),
+        ),
+    ).score(generic_request)
+
+    with pytest.raises(AssessmentSpecError) as caught:
+        build_rag_facets_rating_records(
+            request=generic_request,
+            result=result,
+            engine=engine,
+        )
+
+    assert caught.value.code == "invalid_rag_scoring_request"
+
+
+def test_rag_facets_surface_accepts_no_raw_rag_content() -> None:
+    """Calibration orchestration remains identity-only at the public boundary."""
+    forbidden = {
+        "query_text",
+        "question_text",
+        "context_text",
+        "retrieved_text",
+        "answer_text",
+        "response_text",
+        "source_text",
+    }
+    for function in (
+        build_rag_facets_rating_records,
+        build_rag_facets_calibration_bundle,
+    ):
+        assert not forbidden.intersection(inspect.signature(function).parameters)
+        assert function.__doc__
diff --git a/tests/test_scoring_rag_facets_calibration_bounds.py b/tests/test_scoring_rag_facets_calibration_bounds.py
new file mode 100644
index 000000000..07fc3c232
--- /dev/null
+++ b/tests/test_scoring_rag_facets_calibration_bounds.py
@@ -0,0 +1,16 @@
+"""Boundary coverage for governed RAG calibration assembly."""
+
+from __future__ import annotations
+
+import pytest
+
+from fast_mlsirm.scoring import AssessmentSpecError
+from fast_mlsirm.scoring.rag_calibration import build_rag_facets_calibration_bundle
+
+
+def test_rag_facets_bundle_rejects_non_triple_execution() -> None:
+    """Every calibration execution must preserve request, result, and engine."""
+    with pytest.raises(AssessmentSpecError) as caught:
+        build_rag_facets_calibration_bundle(((object(), object()),))
+
+    assert caught.value.code == "invalid_rag_calibration_execution"