release: prepare v1.3.0

2026-07-17 15:48:43 +03:00
parent 1d47408157
commit 4a2e9d658f
35 changed files with 3053 additions and 48 deletions
@@ -7,7 +7,47 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0

 ## [Unreleased]

-_Next: closed-loop refiner (critique→repair→re-critique with transaction-stack isolation), pre-plan clarification pass, ISO 286 fit-table lookup (H7/g6 → deviations), and a local-VLM visual judge._
+_Next: capability-group schema routing, paper-space/layout production depth, and ISO 286 fit-table lookup._
+
+## [1.3.0] — 2026-07-17
+
+Closed-loop quality, production annotation, and auditable delivery. **415 tests,
+Ruff lint-clean. Tool count: 116 → 122.**
+
+### Added
+
+- **`drawing_preflight`** normalizes production requirements, reports missing or
+  conflicting facts before geometry starts, and emits a deterministic SHA-256
+  spec hash that `drawing_plan` can enforce.
+- **`drawing_refine`** runs a bounded `critique → repair → re-critique` loop.
+  Each repair round is isolated in its own transaction and rolls back if score
+  regresses or hard-error count increases. Construction, duplicate, layer
+  color/lineweight, untrimmed endpoint, and dimension-overlap repairs are
+  supported; undefined GD&T datums remain explicitly manual.
+- **TABLE and MLEADER semantics** via `entity_create_table` and
+  `leader_create_mleader`: native ActiveX entities on COM and deterministic,
+  portable LINE/LWPOLYLINE/MTEXT composites on ezdxf. Composite creation
+  returns child handles and a logical ID; after DXF reopen, inspection uses the
+  persisted standard child entities rather than a native TABLE/MLEADER object.
+- **`system_capabilities`** and shared typed backend capability maps distinguish
+  native, rendered, composite, snapshot, shared, and unsupported features.
+- **`drawing_deliver`** creates DXF/PDF/PNG bundles, runs validator + critique,
+  re-opens the canonical DXF for entity/type/layer/bounds parity, hashes every
+  artifact, and writes `manifest.json` plus `validation.json`. Failed gates keep
+  artifacts for diagnosis but never report delivery success.
+- **Benchmark v2**: ten fixed vendor-neutral tasks, an adapter interface, an
+  ezdxf/COM reference adapter, cooperative per-task timeout reporting,
+  fixed-matrix coverage-aware scoring, machine-readable statuses,
+  runtime/environment/capability metadata, and artifact SHA-256.
+
+### Changed
+
+- ezdxf transaction snapshots are isolated from user-facing undo history, so a
+  refiner rollback cannot consume an unrelated undo entry.
+- Package metadata is the canonical version source used by `system_about` and
+  delivery manifests.
+- `benchmarks/correctness_suite.py` now shows usage for argless/`--help` calls
+  instead of raising a traceback.

 ### Fixed
 - **Install from source** — `pip install -e ".[full]"` failed at
@@ -13,7 +13,7 @@ query exact geometry, apply engineering standards, refine quality issues inside
 transactions, and deliver hashed artifacts with validation evidence.

 > **v1.3 release snapshot:** 122 tools · 6 resources · 5 prompt templates ·
-> 406 collected tests. Runtime discovery through `system_about` is authoritative.
+> 415 collected tests. Runtime discovery through `system_about` is authoritative.

 ## Start in 60 seconds

@@ -71,6 +71,32 @@ class DrawingInfo:
    backend: str = ""


+@dataclass(frozen=True)
+class FeatureCapability:
+    supported: bool
+    mode: str | None = None
+    reason: str | None = None
+
+    def to_dict(self) -> dict:
+        return {
+            "supported": self.supported,
+            "mode": self.mode,
+            "reason": self.reason,
+        }
+
+
+@dataclass(frozen=True)
+class CapabilityMap:
+    backend: str
+    features: dict[str, FeatureCapability]
+
+    def to_dict(self) -> dict:
+        return {
+            "backend": self.backend,
+            "features": {name: value.to_dict() for name, value in self.features.items()},
+        }
+
+
 # ---------------------------------------------------------------------------
 # Shared helpers
 # ---------------------------------------------------------------------------
@@ -126,6 +152,9 @@ class AutoCADBackend(ABC):
    @abstractmethod
    async def disconnect(self) -> None: ...

+    @abstractmethod
+    def capabilities(self) -> CapabilityMap: ...
+
    # ── drawing management ───────────────────────────────────────────────────
    @abstractmethod
    async def drawing_info(self) -> DrawingInfo: ...
@@ -205,6 +234,31 @@ class AutoCADBackend(ABC):
        layer: str | None = None, color: int | None = None,
    ) -> EntityInfo: ...

+    @abstractmethod
+    async def entity_create_table(
+        self,
+        x: float,
+        y: float,
+        rows: list[list[str]],
+        headers: list[str] | None = None,
+        column_widths: list[float] | None = None,
+        row_height: float = 7.0,
+        text_height: float = 2.5,
+        title: str | None = None,
+        layer: str = "TEXT",
+    ) -> EntityInfo: ...
+
+    @abstractmethod
+    async def leader_create_mleader(
+        self,
+        points: list[list[float]],
+        text: str,
+        text_height: float = 2.5,
+        landing_gap: float = 1.0,
+        arrow_size: float = 2.5,
+        layer: str = "DIM",
+    ) -> EntityInfo: ...
+
    @abstractmethod
    async def entity_create_hatch(
        self, pattern: str, boundary_points: list[list[float]],
@@ -594,6 +648,46 @@ class AutoCADBackend(ABC):
    # has no cross-backend equivalent among the standard entity_create_* tools.

    _plan_spec: PlanSpec | None = None
+    _preflight_result: Any = None
+
+    def _reset_document_state(self) -> None:
+        """Clear metadata that is valid only for the current drawing."""
+        self._plan_spec = None
+        self._preflight_result = None
+        self._gdt_datums_defined = set()
+        self._gdt_datums_referenced = set()
+
+    async def _ensure_document_state(self) -> None:
+        """Allow live backends to detect an out-of-band document switch."""
+        return None
+
+    async def drawing_preflight(
+        self,
+        intent: str,
+        requirements: dict | None = None,
+        sheet_size: SheetSize = "A3",
+        scale: float = 1.0,
+        layer_set_id: LayerSetId = "mech",
+        view_count: int = 1,
+        dim_style: DimStyle = "chain",
+        allow_assumptions: bool = False,
+    ):
+        from engineering.preflight import preflight_drawing
+
+        await self._ensure_document_state()
+
+        result = preflight_drawing(
+            intent,
+            requirements,
+            sheet_size,
+            scale,
+            layer_set_id,
+            view_count,
+            dim_style,
+            allow_assumptions=allow_assumptions,
+        )
+        self._preflight_result = result
+        return result

    async def drawing_plan(
        self, intent: str,
@@ -603,6 +697,8 @@ class AutoCADBackend(ABC):
        view_count: int = 1,
        dim_style: DimStyle = "chain",
        notes: list[str] | None = None,
+        requirements: dict | None = None,
+        spec_hash: str | None = None,
    ) -> PlanSpec:
        """Commit a drawing intent before any geometry is created.
        Returns a PlanSpec and stores it on the backend; read it back with
@@ -610,6 +706,47 @@ class AutoCADBackend(ABC):
        replays this plan was untrue — critique runs geometric checks, not a
        plan diff.)"""
        from engineering.plan_spec import PlanSpec
+
+        await self._ensure_document_state()
+        preflight_status = "skipped"
+        normalized_plan: dict[str, Any] | None = None
+        if spec_hash is not None:
+            from engineering.preflight import preflight_drawing
+
+            preflight = getattr(self, "_preflight_result", None)
+            if preflight is None or preflight.spec_hash != spec_hash:
+                raise RuntimeError("drawing_plan: spec_hash does not match the latest preflight.")
+            if not preflight.ready:
+                raise RuntimeError(
+                    "drawing_plan: preflight is not ready; resolve questions/conflicts first."
+                )
+            candidate = preflight_drawing(
+                intent,
+                requirements
+                if requirements is not None
+                else preflight.normalized_spec["requirements"],
+                sheet_size,
+                scale,
+                layer_set_id,
+                view_count,
+                dim_style,
+            )
+            if candidate.spec_hash != spec_hash:
+                raise RuntimeError(
+                    "drawing_plan: supplied fields do not match the hashed preflight spec."
+                )
+            normalized_plan = preflight.normalized_spec
+            preflight_status = "ready"
+
+        if normalized_plan is not None:
+            intent = normalized_plan["intent"]
+            sheet_size = normalized_plan["sheet_size"]
+            scale = normalized_plan["scale"]
+            layer_set_id = normalized_plan["layer_set_id"]
+            view_count = normalized_plan["view_count"]
+            dim_style = normalized_plan["dim_style"]
+            requirements = normalized_plan["requirements"]
+
        plan = PlanSpec(
            intent=str(intent),
            sheet_size=sheet_size,
@@ -618,6 +755,9 @@ class AutoCADBackend(ABC):
            view_count=int(view_count),
            dim_style=dim_style,
            notes=list(notes) if notes else [],
+            requirements=dict(requirements or {}),
+            spec_hash=spec_hash,
+            preflight_status=preflight_status,
        )
        self._plan_spec = plan
        return plan
@@ -629,6 +769,8 @@ class AutoCADBackend(ABC):
        `drawing_finalize`. `focus=None` runs all checks. (N8: this does NOT
        replay the committed PlanSpec — it inspects the live drawing geometry.)"""
        from engineering.critique import run_critique
+
+        await self._ensure_document_state()
        return await run_critique(self, focus)

    def get_plan_spec(self) -> dict | None:
@@ -21,8 +21,10 @@ import config
 from .base import (
    AutoCADBackend,
    BlockInfo,
+    CapabilityMap,
    DrawingInfo,
    EntityInfo,
+    FeatureCapability,
    LayerInfo,
    deg2rad,
    normalize_lineweight,
@@ -431,6 +433,7 @@ class ComBackend(AutoCADBackend):
        self._executor: ThreadPoolExecutor | None = None
        self._connected = False
        self._transaction_active = False
+        self._document_scope_key: tuple[str, str] | None = None

    @property
    def name(self) -> str:
@@ -440,6 +443,26 @@ class ComBackend(AutoCADBackend):
    def is_connected(self) -> bool:
        return self._connected

+    def capabilities(self) -> CapabilityMap:
+        return CapabilityMap(
+            backend="com",
+            features={
+                "drawing_2d": FeatureCapability(True, "native"),
+                "dxf": FeatureCapability(True, "native"),
+                "pdf": FeatureCapability(True, "native"),
+                "png": FeatureCapability(True, "native"),
+                "transactions": FeatureCapability(True, "undo_mark"),
+                "table": FeatureCapability(True, "native"),
+                "mleader": FeatureCapability(True, "native"),
+                "preflight": FeatureCapability(True, "shared"),
+                "refiner": FeatureCapability(True, "shared"),
+                "delivery": FeatureCapability(True, "shared"),
+                "paper_space": FeatureCapability(False, reason="planned_1.4"),
+                "solid_3d": FeatureCapability(False, reason="planned_1.5"),
+                "lisp": FeatureCapability(True, "sanitized"),
+            },
+        )
+
    async def connect(self) -> None:
        if not _COM_IMPORTS_OK:
            raise RuntimeError(
@@ -460,8 +483,28 @@ class ComBackend(AutoCADBackend):
                pass
            self._executor.shutdown(wait=False)
            self._executor = None
+        self._document_scope_key = None
+        self._transaction_active = False
+        self._reset_document_state()
        self._connected = False

+    async def _ensure_document_state(self) -> None:
+        """Reset drawing-scoped metadata when AutoCAD's active document changes."""
+
+        def _sync():
+            app = _acad_app()
+            if app.Documents.Count == 0:
+                key = None
+            else:
+                doc = app.ActiveDocument
+                key = (str(doc.Name), str(doc.FullName))
+            if key != self._document_scope_key:
+                self._document_scope_key = key
+                self._transaction_active = False
+                self._reset_document_state()
+
+        await self._run(_sync)
+
    def _sync_test_connection(self):
        """Verify AutoCAD is reachable (runs in COM thread)."""
        app = _acad_app()
@@ -572,14 +615,18 @@ class ComBackend(AutoCADBackend):
            else:
                doc = app.Documents.Add()
            return {"ok": True, "name": doc.Name}
-        return await self._run(_sync)
+        result = await self._run(_sync)
+        await self._ensure_document_state()
+        return result

    async def drawing_open(self, path: str) -> dict:
        def _sync():
            app = _acad_app()
            doc = app.Documents.Open(path)
            return {"ok": True, "name": doc.Name, "path": doc.FullName}
-        return await self._run(_sync)
+        result = await self._run(_sync)
+        await self._ensure_document_state()
+        return result

    async def drawing_save(self, path: str | None = None) -> dict:
        def _sync():
@@ -633,7 +680,9 @@ class ComBackend(AutoCADBackend):
            doc = _acad_doc()
            doc.Close(save)
            return {"ok": True}
-        return await self._run(_sync)
+        result = await self._run(_sync)
+        await self._ensure_document_state()
+        return result

    async def drawing_undo(self) -> dict:
        def _sync():
@@ -729,6 +778,113 @@ class ComBackend(AutoCADBackend):
            return _entity_info(mt)
        return await self._run(_sync)

+    async def entity_create_table(
+        self,
+        x,
+        y,
+        rows,
+        headers=None,
+        column_widths=None,
+        row_height=7.0,
+        text_height=2.5,
+        title=None,
+        layer="TEXT",
+    ) -> EntityInfo:
+        from engineering.annotation import prepare_table_layout
+
+        layout = prepare_table_layout(
+            rows, headers, column_widths, row_height, text_height, title
+        )
+
+        def _sync():
+            mspace = _msp()
+            table = mspace.AddTable(
+                _apoint(x, y),
+                layout.row_count,
+                layout.column_count,
+                layout.row_height,
+                layout.column_widths[0],
+            )
+            for column, width in enumerate(layout.column_widths):
+                table.SetColumnWidth(column, width)
+            for row_index, row in enumerate(layout.cells):
+                table.SetRowHeight(row_index, layout.row_height)
+                for column_index, value in enumerate(row):
+                    table.SetText(row_index, column_index, value)
+            try:
+                table.TextHeight = layout.text_height
+            except Exception as exc:
+                log.debug("setting table TextHeight failed: %s", exc)
+            _apply_entity_attrs(table, layer, None, None)
+            _regen()
+            info = _entity_info(table)
+            info.type = "TABLE"
+            info.layer = layer
+            info.properties.update(
+                {
+                    "representation": "native",
+                    "logical_group_id": f"table:{uuid.uuid4().hex}",
+                    "child_handles": [info.handle],
+                    "rows": layout.row_count,
+                    "columns": layout.column_count,
+                    "bounds": {
+                        "min": [float(x), float(y) - layout.height],
+                        "max": [float(x) + layout.width, float(y)],
+                    },
+                }
+            )
+            return info
+
+        return await self._run(_sync)
+
+    async def leader_create_mleader(
+        self,
+        points,
+        text,
+        text_height=2.5,
+        landing_gap=1.0,
+        arrow_size=2.5,
+        layer="DIM",
+    ) -> EntityInfo:
+        from engineering.annotation import validate_mleader
+
+        normalized = validate_mleader(points, text)
+        if text_height <= 0 or landing_gap < 0 or arrow_size <= 0:
+            raise RuntimeError(
+                "leader_create_mleader: text_height/arrow_size must be positive and landing_gap non-negative"
+            )
+
+        def _sync():
+            flat = []
+            for px, py in normalized:
+                flat.extend([px, py, 0.0])
+            leader = _msp().AddMLeader(_av(flat), 0)
+            leader.TextString = str(text)
+            leader.TextHeight = float(text_height)
+            leader.LandingGap = float(landing_gap)
+            leader.ArrowheadSize = float(arrow_size)
+            _apply_entity_attrs(leader, layer, None, None)
+            _regen()
+            info = _entity_info(leader)
+            info.type = "MLEADER"
+            info.layer = layer
+            info.properties.update(
+                {
+                    "representation": "native",
+                    "logical_group_id": f"mleader:{uuid.uuid4().hex}",
+                    "child_handles": [info.handle],
+                    "points": [list(point) for point in normalized],
+                    "text": str(text),
+                    "bounds": {
+                        "min": [min(p[0] for p in normalized), min(p[1] for p in normalized)],
+                        "max": [max(p[0] for p in normalized), max(p[1] for p in normalized)],
+                    },
+                }
+            )
+            return info
+
+        return await self._run(_sync)
+
    async def entity_create_hatch(
        self, pattern, boundary_points, scale=1.0, angle=0.0,
        layer=None, color=None,
@@ -1649,6 +1805,9 @@ class ComBackend(AutoCADBackend):
    # ── transactions ──────────────────────────────────────────────────────────

    async def transaction_begin(self) -> dict:
+        if self._transaction_active:
+            return {"ok": False, "error": "A transaction is already active"}
+
        def _sync():
            doc = _acad_doc()
            doc.StartUndoMark()
@@ -13,6 +13,8 @@ import logging
 import math
 import os
 import tempfile
+import uuid
+from importlib.util import find_spec
 from pathlib import Path
 from typing import Any

@@ -21,8 +23,10 @@ import config
 from .base import (
    AutoCADBackend,
    BlockInfo,
+    CapabilityMap,
    DrawingInfo,
    EntityInfo,
+    FeatureCapability,
    LayerInfo,
    normalize_lineweight,
    shoelace_area,
@@ -253,7 +257,8 @@ class EzdxfBackend(AutoCADBackend):
        self._doc_path: str | None = None
        self._dirty: bool = False
        self._current_layer: str = "0"
-        self._undo_stack: list[Path] = []  # temp-file paths to DXF snapshots
+        self._undo_stack: list[Path] = []  # user-facing undo history
+        self._transaction_stack: list[Path] = []  # isolated transaction snapshots
        self._connected = False
        self._lock = asyncio.Lock()

@@ -265,21 +270,49 @@ class EzdxfBackend(AutoCADBackend):
    def is_connected(self) -> bool:
        return self._connected

+    def capabilities(self) -> CapabilityMap:
+        render_available = find_spec("matplotlib") is not None
+        render_capability = (
+            FeatureCapability(True, "rendered")
+            if render_available
+            else FeatureCapability(False, reason="optional_dependency_missing:matplotlib")
+        )
+        return CapabilityMap(
+            backend="ezdxf",
+            features={
+                "drawing_2d": FeatureCapability(True, "native"),
+                "dxf": FeatureCapability(True, "native"),
+                "pdf": render_capability,
+                "png": render_capability,
+                "transactions": FeatureCapability(True, "snapshot"),
+                "table": FeatureCapability(True, "composite"),
+                "mleader": FeatureCapability(True, "composite"),
+                "preflight": FeatureCapability(True, "shared"),
+                "refiner": FeatureCapability(True, "shared"),
+                "delivery": FeatureCapability(True, "shared"),
+                "paper_space": FeatureCapability(False, reason="planned_1.4"),
+                "solid_3d": FeatureCapability(False, reason="planned_1.5"),
+                "lisp": FeatureCapability(False, reason="live_com_only"),
+            },
+        )
+
    async def connect(self) -> None:
        self._connected = True
        log.info("ezdxf backend ready")

    async def disconnect(self) -> None:
        self._cleanup_undo_stack()
+        self._reset_document_state()
        self._connected = False

    def _cleanup_undo_stack(self):
-        while self._undo_stack:
-            p = self._undo_stack.pop()
-            try:
-                p.unlink()
-            except OSError:
-                pass
+        for stack in (self._undo_stack, self._transaction_stack):
+            while stack:
+                p = stack.pop()
+                try:
+                    p.unlink()
+                except OSError:
+                    pass

    def _require_doc(self):
        if self._doc is None:
@@ -347,6 +380,7 @@ class EzdxfBackend(AutoCADBackend):
            self._doc_path = None
            self._dirty = False
            self._current_layer = "0"
+            self._reset_document_state()
            return {"ok": True, "name": "untitled.dxf"}
        return await self._async(_sync)

@@ -371,6 +405,7 @@ class EzdxfBackend(AutoCADBackend):
            except Exception as exc:
                log.debug("reading $CLAYER from header: %s", exc)
                self._current_layer = "0"
+            self._reset_document_state()
            return {"ok": True, "name": Path(path).name, "path": path}
        return await self._async(_sync)

@@ -511,6 +546,7 @@ class EzdxfBackend(AutoCADBackend):
            self._doc = None
            self._doc_path = None
            self._dirty = False
+            self._reset_document_state()
            return {"ok": True}
        return await self._async(_sync)

@@ -645,6 +681,160 @@ class EzdxfBackend(AutoCADBackend):
            return _entity_info_dxf(ent)
        return await self._async(_sync)

+    async def entity_create_table(
+        self,
+        x,
+        y,
+        rows,
+        headers=None,
+        column_widths=None,
+        row_height=7.0,
+        text_height=2.5,
+        title=None,
+        layer="TEXT",
+    ) -> EntityInfo:
+        from engineering.annotation import prepare_table_layout
+
+        layout = prepare_table_layout(
+            rows, headers, column_widths, row_height, text_height, title
+        )
+
+        def _sync():
+            msp = self._msp()
+            children = []
+            x0, y0 = float(x), float(y)
+            x_positions = [x0]
+            for width in layout.column_widths:
+                x_positions.append(x_positions[-1] + width)
+
+            for row_index in range(layout.row_count + 1):
+                yy = y0 - row_index * layout.row_height
+                line = msp.add_line((x0, yy), (x0 + layout.width, yy))
+                self._apply_attrs(line, layer, None)
+                children.append(line)
+            for xx in x_positions:
+                line = msp.add_line((xx, y0), (xx, y0 - layout.height))
+                self._apply_attrs(line, layer, None)
+                children.append(line)
+
+            padding = min(1.0, layout.row_height * 0.15)
+            for row_index, row in enumerate(layout.cells):
+                for column_index, value in enumerate(row):
+                    if not value:
+                        continue
+                    text_entity = msp.add_mtext(
+                        value,
+                        dxfattribs={
+                            "char_height": layout.text_height,
+                            "width": max(1.0, layout.column_widths[column_index] - 2 * padding),
+                        },
+                    )
+                    text_entity.dxf.insert = (
+                        x_positions[column_index] + padding,
+                        y0 - row_index * layout.row_height - padding,
+                        0.0,
+                    )
+                    self._apply_attrs(text_entity, layer, None)
+                    children.append(text_entity)
+
+            self._mark_dirty()
+            child_handles = [str(entity.dxf.handle) for entity in children]
+            group_id = f"table:{uuid.uuid4().hex}"
+            return EntityInfo(
+                handle=child_handles[0],
+                type="TABLE",
+                layer=layer,
+                color=256,
+                linetype="ByLayer",
+                visible=True,
+                properties={
+                    "representation": "composite",
+                    "logical_group_id": group_id,
+                    "child_handles": child_handles,
+                    "rows": layout.row_count,
+                    "columns": layout.column_count,
+                    "bounds": {
+                        "min": [x0, y0 - layout.height],
+                        "max": [x0 + layout.width, y0],
+                    },
+                },
+            )
+
+        return await self._async(_sync)
+
+    async def leader_create_mleader(
+        self,
+        points,
+        text,
+        text_height=2.5,
+        landing_gap=1.0,
+        arrow_size=2.5,
+        layer="DIM",
+    ) -> EntityInfo:
+        from engineering.annotation import validate_mleader
+
+        normalized = validate_mleader(points, text)
+        if text_height <= 0 or landing_gap < 0 or arrow_size <= 0:
+            raise RuntimeError(
+                "leader_create_mleader: text_height/arrow_size must be positive and landing_gap non-negative"
+            )
+
+        def _sync():
+            msp = self._msp()
+            path = msp.add_lwpolyline(normalized)
+            self._apply_attrs(path, layer, None)
+
+            (x0, y0), (x1, y1) = normalized[0], normalized[1]
+            dx, dy = x1 - x0, y1 - y0
+            length = math.hypot(dx, dy)
+            if length <= 1e-12:
+                raise RuntimeError("leader_create_mleader: first two points must be distinct")
+            ux, uy = dx / length, dy / length
+            px, py = -uy, ux
+            size = float(arrow_size)
+            arrow = msp.add_lwpolyline(
+                [
+                    (x0, y0),
+                    (x0 + ux * size + px * size * 0.35, y0 + uy * size + py * size * 0.35),
+                    (x0 + ux * size - px * size * 0.35, y0 + uy * size - py * size * 0.35),
+                ],
+                close=True,
+            )
+            self._apply_attrs(arrow, layer, None)
+
+            tx, ty = normalized[-1]
+            label = msp.add_mtext(
+                str(text),
+                dxfattribs={"char_height": float(text_height), "width": 100.0},
+            )
+            label.dxf.insert = (tx + float(landing_gap), ty, 0.0)
+            self._apply_attrs(label, layer, None)
+            self._mark_dirty()
+
+            children = [path, arrow, label]
+            child_handles = [str(entity.dxf.handle) for entity in children]
+            return EntityInfo(
+                handle=child_handles[0],
+                type="MLEADER",
+                layer=layer,
+                color=256,
+                linetype="ByLayer",
+                visible=True,
+                properties={
+                    "representation": "composite",
+                    "logical_group_id": f"mleader:{uuid.uuid4().hex}",
+                    "child_handles": child_handles,
+                    "points": [list(point) for point in normalized],
+                    "text": str(text),
+                    "bounds": {
+                        "min": [min(point[0] for point in normalized), min(point[1] for point in normalized)],
+                        "max": [max(point[0] for point in normalized), max(point[1] for point in normalized)],
+                    },
+                },
+            )
+
+        return await self._async(_sync)
+
    async def entity_create_hatch(
        self, pattern, boundary_points, scale=1.0, angle=0.0,
        layer=None, color=None,
@@ -1632,10 +1822,10 @@ class EzdxfBackend(AutoCADBackend):
                except OSError:
                    pass
                raise
-            self._undo_stack.append(Path(tmp_path))
+            self._transaction_stack.append(Path(tmp_path))
            max_stack = config.settings.max_undo_stack
-            while len(self._undo_stack) > max_stack:
-                old = self._undo_stack.pop(0)
+            while len(self._transaction_stack) > max_stack:
+                old = self._transaction_stack.pop(0)
                try:
                    old.unlink()
                except OSError:
@@ -1647,9 +1837,9 @@ class EzdxfBackend(AutoCADBackend):
        # NEW-undo-1: guard the _undo_stack read+pop with self._lock (via _async)
        # instead of touching it outside any lock.
        def _sync():
-            if not self._undo_stack:
+            if not self._transaction_stack:
                return {"ok": False, "error": "No active transaction"}
-            p = self._undo_stack.pop()
+            p = self._transaction_stack.pop()
            try:
                p.unlink()
            except OSError:
@@ -1661,11 +1851,12 @@ class EzdxfBackend(AutoCADBackend):
        # NEW-undo-1: read+mutate _undo_stack under self._lock (via _async) so the
        # empty-check and the pop/readfile are serialized with other sync work.
        def _sync():
-            if not self._undo_stack:
+            if not self._transaction_stack:
                return {"ok": False, "error": "No active transaction to rollback"}
-            p = self._undo_stack.pop()
+            p = self._transaction_stack.pop()
            try:
                self._doc = ezdxf.readfile(str(p))
+                self._current_layer = str(self._doc.header.get("$CLAYER", "0"))
                self._dirty = True
            finally:
                try:
@@ -1685,7 +1876,7 @@ class EzdxfBackend(AutoCADBackend):
            "has_document": has_doc,
            "document_path": self._doc_path,
            "unsaved_changes": self._dirty,
-            "transaction_depth": len(self._undo_stack),
+            "transaction_depth": len(self._transaction_stack),
            "capabilities": [
                "file_read_write", "dxf_export", "pdf_export",
                "entity_creation", "layer_management", "blocks",
@@ -29,8 +29,12 @@ dates, evidence grades, project URLs, and the boundary are stored with the data.
 The v2 runner separates runtime evidence from the source-review rubric. Every
 adapter receives the same ten tasks and produces the same closed result enum:
 `pass`, `partial`, `unsupported`, `fail`, `timeout`, or `not_run`. A timeout is
-isolated to its task. Reports include commit SHA, Python/platform details,
-backend capability claims, durations, and hashes for returned artifacts.
+reported per task. Execution is in-process, so adapters that delegate blocking
+work must provide cancellation-safe task implementations or be wrapped by an
+external process supervisor. Reports include commit SHA, Python/platform
+details, backend capability claims, durations, coverage, and hashes for returned
+artifacts. Unsupported tasks remain in the fixed-matrix denominator with score
+zero, preventing partial implementations from receiving an inflated score.

 ```bash
 python -m benchmarks.run_competitors --list
@@ -0,0 +1 @@
+"""Benchmark suites for AutoCAD MCP Pro."""
@@ -0,0 +1,5 @@
+"""Benchmark adapters."""
+
+from .base import BenchmarkAdapter, TaskResult
+
+__all__ = ["BenchmarkAdapter", "TaskResult"]
@@ -0,0 +1,151 @@
+"""Reference v2 adapter for this repository."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+from benchmarks.adapters.base import BenchmarkAdapter, TaskResult
+from benchmarks.tasks_v2 import TaskSpec
+from engineering.delivery import deliver_drawing
+from engineering.layers import ensure_engineering_layers, ensure_standard_linetypes
+from engineering.refiner import refine_drawing
+
+
+class AutoCADMCPProAdapter(BenchmarkAdapter):
+    name = "autocad-mcp-pro"
+
+    def __init__(self, *, backend: str = "ezdxf"):
+        self.backend_name = backend
+        self.backend = None
+        self.artifact_dir = Path()
+
+    async def setup(self, artifact_dir: Path) -> None:
+        if self.backend_name == "ezdxf":
+            from backends.ezdxf_backend import EzdxfBackend
+
+            self.backend = EzdxfBackend()
+        elif self.backend_name == "com":
+            from backends.com_backend import ComBackend
+
+            self.backend = ComBackend()
+        else:
+            raise ValueError(f"Unknown backend: {self.backend_name}")
+        self.artifact_dir = artifact_dir
+        await self.backend.connect()
+
+    async def cleanup(self) -> None:
+        if self.backend is not None:
+            await self.backend.disconnect()
+
+    def metadata(self) -> dict:
+        return {
+            "backend": self.backend_name,
+            "capabilities": self.backend.capabilities().to_dict() if self.backend else None,
+        }
+
+    async def _reset(self) -> None:
+        await self.backend.drawing_new()
+
+    async def run_task(self, task: TaskSpec) -> TaskResult:
+        await self._reset()
+        handler = getattr(self, f"_task_{task.task_id}", None)
+        if handler is None:
+            return TaskResult(task.task_id, "unsupported", 0.0, "No adapter implementation")
+        passed, metrics, artifacts = await handler()
+        return TaskResult(
+            task.task_id,
+            "pass" if passed else "fail",
+            100.0 if passed else 0.0,
+            metrics=metrics,
+            artifacts=artifacts,
+        )
+
+    async def _task_core_geometry(self):
+        line = await self.backend.entity_create_line(0, 0, 3, 4)
+        circle = await self.backend.entity_create_circle(10, 10, 2)
+        info = await self.backend.entity_get(line.handle)
+        passed = abs(float(info.properties["length"]) - 5.0) < 1e-6
+        return passed and bool(circle.handle), {"line_length": info.properties["length"]}, []
+
+    async def _task_modify_query(self):
+        line = await self.backend.entity_create_line(0, 0, 10, 0)
+        await self.backend.entity_move(line.handle, 5, 2)
+        moved = await self.backend.entity_get(line.handle)
+        return moved.properties["start"][:2] == [5.0, 2.0], {}, []
+
+    async def _task_layers_linetypes(self):
+        await ensure_standard_linetypes(self.backend)
+        await ensure_engineering_layers(self.backend)
+        layers = {item.name for item in await self.backend.layer_list()}
+        linetypes = {item.upper() for item in await self.backend.linetype_list()}
+        return "GEOMETRY" in layers and "CENTER" in linetypes, {}, []
+
+    async def _task_dimensions(self):
+        await ensure_engineering_layers(self.backend)
+        dimension = await self.backend.dimension_linear(0, 0, 50, 0, 25, 10, layer="DIM")
+        return "DIM" in dimension.type.upper(), {"entity_type": dimension.type}, []
+
+    async def _task_table_mleader(self):
+        await ensure_engineering_layers(self.backend)
+        table = await self.backend.entity_create_table(
+            0, 20, [["A", "1"]], headers=["ITEM", "QTY"], layer="TEXT"
+        )
+        leader = await self.backend.leader_create_mleader(
+            [[0, 0], [10, 10]], "NOTE", layer="DIM"
+        )
+        representations = [table.properties.get("representation"), leader.properties.get("representation")]
+        return all(representations), {"representations": representations}, []
+
+    async def _task_transactions(self):
+        await self.backend.transaction_begin()
+        await self.backend.entity_create_circle(0, 0, 1)
+        await self.backend.transaction_rollback()
+        entities = await self.backend.entity_list(limit=100)
+        return len(entities) == 0, {"entity_count": len(entities)}, []
+
+    async def _task_preflight(self):
+        result = await self.backend.drawing_preflight(
+            "Benchmark plate",
+            {
+                "units": "mm",
+                "part_type": "plate",
+                "dimensions": {"width": 50, "height": 25},
+                "tolerance_policy": "ISO 2768-m",
+            },
+        )
+        return result.ready and result.spec_hash.startswith("sha256:"), {}, []
+
+    async def _task_quality_refiner(self):
+        await ensure_engineering_layers(self.backend)
+        await self.backend.entity_create_line(0, 0, 10, 0, layer="GEOMETRY")
+        await self.backend.entity_create_line(0, 0, 10, 0, layer="GEOMETRY")
+        result = await refine_drawing(self.backend, focus=["duplicate_entities"])
+        return result.final_score > result.initial_score, result.to_dict(), []
+
+    async def _task_dxf_roundtrip(self):
+        await self.backend.entity_create_line(0, 0, 10, 10)
+        path = self.artifact_dir / "roundtrip.dxf"
+        await self.backend.drawing_export_dxf(str(path))
+        from backends.ezdxf_backend import EzdxfBackend
+
+        reopened = EzdxfBackend()
+        await reopened.connect()
+        try:
+            await reopened.drawing_open(str(path))
+            count = (await reopened.drawing_info()).entity_count
+        finally:
+            await reopened.disconnect()
+        return count == 1, {"entity_count": count}, [str(path)]
+
+    async def _task_auditable_delivery(self):
+        await ensure_engineering_layers(self.backend)
+        await self.backend.entity_create_line(0, 0, 10, 0, layer="GEOMETRY")
+        destination = self.artifact_dir / "delivery"
+        result = await deliver_drawing(
+            self.backend,
+            destination,
+            formats=["dxf"],
+            min_score=0,
+            strict_critique=False,
+        )
+        return result.status == "success", {"score": result.score}, [result.manifest_path]
@@ -0,0 +1,58 @@
+"""Common, vendor-neutral benchmark adapter contract."""
+
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import TYPE_CHECKING, Any, Literal
+
+if TYPE_CHECKING:
+    from benchmarks.tasks_v2 import TaskSpec
+
+TaskStatus = Literal["pass", "partial", "unsupported", "fail", "timeout", "not_run"]
+
+
+@dataclass
+class TaskResult:
+    task_id: str
+    status: TaskStatus
+    score: float
+    message: str = ""
+    duration_ms: float = 0.0
+    artifacts: list[str | dict[str, Any]] = field(default_factory=list)
+    metrics: dict = field(default_factory=dict)
+    stdout_summary: str = ""
+    stderr_summary: str = ""
+
+    def to_dict(self) -> dict:
+        return {
+            "task_id": self.task_id,
+            "status": self.status,
+            "score": float(self.score),
+            "message": self.message,
+            "duration_ms": float(self.duration_ms),
+            "artifacts": list(self.artifacts),
+            "metrics": dict(self.metrics),
+            "stdout_summary": self.stdout_summary,
+            "stderr_summary": self.stderr_summary,
+        }
+
+
+class BenchmarkAdapter(ABC):
+    """Minimum interface every competing MCP adapter must implement."""
+
+    name: str
+
+    def metadata(self) -> dict:
+        """Runtime/backend claims captured alongside every benchmark report."""
+        return {}
+
+    @abstractmethod
+    async def setup(self, artifact_dir: Path) -> None: ...
+
+    @abstractmethod
+    async def run_task(self, task: TaskSpec) -> TaskResult: ...
+
+    @abstractmethod
+    async def cleanup(self) -> None: ...
@@ -0,0 +1,10 @@
+{
+  "schema_version": "2.0",
+  "competitors": [
+    {
+      "id": "autocad-mcp-pro",
+      "adapter": "benchmarks.adapters.autocad_mcp_pro:AutoCADMCPProAdapter",
+      "backends": ["ezdxf", "com"]
+    }
+  ]
+}
@@ -224,11 +224,18 @@ CHECKS = {


 def main():
+    if len(sys.argv) == 1 or sys.argv[1] in {"-h", "--help"}:
+        print("usage: correctness_suite.py <check_name> | --list")
+        print("Run one deterministic correctness check, or list available checks.")
+        return
    if len(sys.argv) == 2 and sys.argv[1] == "--list":
        for k, (_fn, cat) in CHECKS.items():
            print(f"{k}\t{cat}")
        return
    name = sys.argv[1]
+    if name not in CHECKS:
+        print(f"Unknown check: {name}", file=sys.stderr)
+        raise SystemExit(2)
    fn, _cat = CHECKS[name]
    ok = asyncio.run(fn())
    print("PASS" if ok else "FAIL")
@@ -0,0 +1 @@
+
@@ -0,0 +1,156 @@
+"""Command line runner for the fixed-task benchmark v2 adapter contract."""
+
+from __future__ import annotations
+
+import argparse
+import asyncio
+import hashlib
+import json
+import platform
+import subprocess
+import sys
+import time
+from datetime import UTC, datetime
+from pathlib import Path
+
+from benchmarks.adapters.base import BenchmarkAdapter, TaskResult
+from benchmarks.tasks_v2 import TASKS_V2, TaskSpec, task_by_id
+
+
+def _git_sha() -> str | None:
+    try:
+        result = subprocess.run(
+            ["git", "rev-parse", "HEAD"],
+            check=True,
+            capture_output=True,
+            text=True,
+            timeout=2,
+        )
+        return result.stdout.strip() or None
+    except (OSError, subprocess.SubprocessError):
+        return None
+
+
+def _hash_file(path: Path) -> dict:
+    digest = hashlib.sha256()
+    with path.open("rb") as stream:
+        for chunk in iter(lambda: stream.read(1024 * 1024), b""):
+            digest.update(chunk)
+    return {
+        "path": str(path),
+        "filename": path.name,
+        "size_bytes": path.stat().st_size,
+        "sha256": digest.hexdigest(),
+    }
+
+
+def _normalize_artifacts(result: TaskResult) -> None:
+    normalized: list[dict] = []
+    for artifact in result.artifacts:
+        if isinstance(artifact, dict):
+            normalized.append(dict(artifact))
+            continue
+        path = Path(artifact).resolve()
+        if path.is_file():
+            normalized.append(_hash_file(path))
+        else:
+            normalized.append({"path": str(path), "missing": True})
+    result.artifacts = normalized
+
+
+async def run_tasks(
+    adapter: BenchmarkAdapter,
+    tasks: list[TaskSpec] | tuple[TaskSpec, ...],
+    artifact_dir: str | Path,
+    *,
+    timeout: float = 30.0,
+) -> dict:
+    destination = Path(artifact_dir).resolve()
+    destination.mkdir(parents=True, exist_ok=True)
+    results: list[TaskResult] = []
+    await adapter.setup(destination)
+    try:
+        for task in tasks:
+            started = time.perf_counter()
+            try:
+                result = await asyncio.wait_for(adapter.run_task(task), timeout=timeout)
+            except TimeoutError:
+                result = TaskResult(task.task_id, "timeout", 0.0, f"Exceeded {timeout}s")
+            except Exception as exc:
+                result = TaskResult(task.task_id, "fail", 0.0, str(exc), stderr_summary=str(exc))
+            result.duration_ms = round((time.perf_counter() - started) * 1000, 2)
+            _normalize_artifacts(result)
+            results.append(result)
+    finally:
+        await adapter.cleanup()
+
+    attempted = len(results)
+    passed = sum(item.status == "pass" for item in results)
+    supported = [item for item in results if item.status != "unsupported"]
+    weight_by_id = {task.task_id: task.weight for task in tasks}
+    total_weight = sum(weight_by_id.values())
+    supported_weight = sum(weight_by_id[item.task_id] for item in supported)
+    weighted_score = sum(
+        weight_by_id[item.task_id] * item.score
+        for item in results
+        if item.status != "unsupported"
+    )
+    return {
+        "schema_version": "2.0",
+        "adapter": adapter.name,
+        "adapter_metadata": adapter.metadata(),
+        "git_sha": _git_sha(),
+        "generated_at": datetime.now(UTC).isoformat(),
+        "environment": {
+            "python": sys.version.split()[0],
+            "implementation": platform.python_implementation(),
+            "platform": platform.platform(),
+            "machine": platform.machine(),
+        },
+        "timeout_seconds": timeout,
+        "summary": {
+            "attempted": attempted,
+            "passed": passed,
+            "supported": len(supported),
+            "unsupported": sum(item.status == "unsupported" for item in results),
+            "coverage_percent": (
+                round(supported_weight / total_weight * 100, 2) if total_weight else 0.0
+            ),
+            "score": round(weighted_score / total_weight, 2) if total_weight else 0.0,
+        },
+        "results": [item.to_dict() for item in results],
+    }
+
+
+def _parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--list", action="store_true", help="List the fixed benchmark tasks")
+    parser.add_argument("--server", default="autocad-mcp-pro", choices=["autocad-mcp-pro"])
+    parser.add_argument("--backend", default="ezdxf", choices=["ezdxf", "com"])
+    parser.add_argument("--task", action="append", help="Run only this task id (repeatable)")
+    parser.add_argument("--artifact-dir", default="benchmarks/results/latest")
+    parser.add_argument("--timeout", type=float, default=30.0)
+    parser.add_argument("--json", action="store_true", help="Print machine-readable JSON")
+    return parser
+
+
+def main() -> None:
+    args = _parser().parse_args()
+    if args.list:
+        for task in TASKS_V2:
+            print(f"{task.task_id}\t{task.category}\t{task.description}")
+        return
+
+    from benchmarks.adapters.autocad_mcp_pro import AutoCADMCPProAdapter
+
+    tasks = [task_by_id(task_id) for task_id in args.task] if args.task else list(TASKS_V2)
+    adapter = AutoCADMCPProAdapter(backend=args.backend)
+    report = asyncio.run(
+        run_tasks(adapter, tasks, args.artifact_dir, timeout=args.timeout)
+    )
+    rendered = json.dumps(report, ensure_ascii=False, indent=2, sort_keys=True)
+    print(rendered)
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,34 @@
+"""Fixed v2 task matrix shared by all competitor adapters."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class TaskSpec:
+    task_id: str
+    category: str
+    description: str
+    weight: float
+
+
+TASKS_V2: tuple[TaskSpec, ...] = (
+    TaskSpec("core_geometry", "creation", "Create and inspect deterministic geometry", 1.0),
+    TaskSpec("modify_query", "editing", "Modify geometry and verify the result", 1.0),
+    TaskSpec("layers_linetypes", "standards", "Apply standard layers and linetypes", 1.0),
+    TaskSpec("dimensions", "annotation", "Create real associative dimensions", 1.0),
+    TaskSpec("table_mleader", "annotation", "Create TABLE and MLEADER semantics", 1.0),
+    TaskSpec("transactions", "reliability", "Rollback a bounded change", 1.0),
+    TaskSpec("preflight", "planning", "Resolve and hash drawing requirements", 1.0),
+    TaskSpec("quality_refiner", "quality", "Detect and repair a quality defect", 1.0),
+    TaskSpec("dxf_roundtrip", "interoperability", "Save and re-open with structural parity", 1.0),
+    TaskSpec("auditable_delivery", "delivery", "Produce a hashed validation manifest", 1.0),
+)
+
+
+def task_by_id(task_id: str) -> TaskSpec:
+    for task in TASKS_V2:
+        if task.task_id == task_id:
+            return task
+    raise KeyError(f"Unknown benchmark task: {task_id}")
@@ -1,6 +1,6 @@
 # Repository Cleanup, README, and Benchmark Design

-**Date:** 2026-07-17  
+**Date:** 2026-07-17
 **Status:** Approved for implementation

 ## Objective
@@ -1,5 +1,12 @@
 """Deterministic engineering CAD layer for AutoCAD MCP Pro."""

+from .annotation import TableLayout, prepare_table_layout, validate_mleader
+from .delivery import (
+    DeliveryResult,
+    compare_drawing_snapshots,
+    deliver_drawing,
+    drawing_snapshot,
+)
 from .gear import (
    draw_gear_section_aa,
    draw_helical_gear_front_view,
@@ -31,11 +38,25 @@ from .plan_spec import (
    SheetSize,
    SnapType,
 )
+from .preflight import (
+    PreflightConflict,
+    PreflightQuestion,
+    PreflightResult,
+    preflight_drawing,
+)
+from .refiner import RefineResult, RefineRound, RepairAction, refine_drawing
 from .titleblock import TitleBlockMetadata, apply_iso_a3_titleblock
 from .validator import DrawingValidator, ValidationFinding, ValidationResult

 __all__ = [
    "draw_helical_gear_front_view",
+    "DeliveryResult",
+    "compare_drawing_snapshots",
+    "deliver_drawing",
+    "drawing_snapshot",
+    "TableLayout",
+    "prepare_table_layout",
+    "validate_mleader",
    "draw_spur_gear_front_view",
    "draw_gear_section_aa",
    "generate_full_gear_outline",
@@ -63,4 +84,12 @@ __all__ = [
    "SnapType",
    "ALL_CRITIQUE_FOCUSES",
    "ISO_128_LINEWEIGHTS_MM",
+    "PreflightConflict",
+    "PreflightQuestion",
+    "PreflightResult",
+    "preflight_drawing",
+    "RefineResult",
+    "RefineRound",
+    "RepairAction",
+    "refine_drawing",
 ]
@@ -0,0 +1,87 @@
+"""Shared validation and layout math for tables and multileaders."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class TableLayout:
+    cells: list[list[str]]
+    column_widths: list[float]
+    row_height: float
+    text_height: float
+
+    @property
+    def row_count(self) -> int:
+        return len(self.cells)
+
+    @property
+    def column_count(self) -> int:
+        return len(self.column_widths)
+
+    @property
+    def width(self) -> float:
+        return sum(self.column_widths)
+
+    @property
+    def height(self) -> float:
+        return self.row_count * self.row_height
+
+
+def prepare_table_layout(
+    rows: list[list[str]],
+    headers: list[str] | None = None,
+    column_widths: list[float] | None = None,
+    row_height: float = 7.0,
+    text_height: float = 2.5,
+    title: str | None = None,
+) -> TableLayout:
+    """Validate table input and calculate deterministic cross-backend sizes."""
+    if not rows:
+        raise RuntimeError("entity_create_table: at least one data row is required")
+    if len(rows) > 200:
+        raise RuntimeError("entity_create_table: maximum is 200 rows")
+    if row_height <= 0 or text_height <= 0:
+        raise RuntimeError("entity_create_table: row_height and text_height must be positive")
+
+    column_count = len(headers) if headers is not None else len(rows[0])
+    if column_count < 1 or column_count > 30:
+        raise RuntimeError("entity_create_table: table must contain between 1 and 30 columns")
+    if any(len(row) != column_count for row in rows):
+        raise RuntimeError("entity_create_table: all rows must have the same number of columns")
+    if headers is not None and len(headers) != column_count:
+        raise RuntimeError("entity_create_table: headers must match the data column count")
+
+    data = [[str(cell) for cell in row] for row in rows]
+    if headers is not None:
+        data.insert(0, [str(cell) for cell in headers])
+    if title:
+        data.insert(0, [str(title)] + [""] * (column_count - 1))
+
+    if column_widths is None:
+        widths = []
+        for column in range(column_count):
+            longest = max(len(row[column]) for row in data)
+            widths.append(max(10.0, longest * float(text_height) * 0.7))
+    else:
+        if len(column_widths) != column_count:
+            raise RuntimeError("entity_create_table: column_widths must match column count")
+        widths = [float(width) for width in column_widths]
+        if any(width <= 0 for width in widths):
+            raise RuntimeError("entity_create_table: column widths must be positive")
+
+    return TableLayout(data, widths, float(row_height), float(text_height))
+
+
+def validate_mleader(points: list[list[float]], text: str) -> list[tuple[float, float]]:
+    if len(points) < 2:
+        raise RuntimeError("leader_create_mleader: at least two points are required")
+    if not str(text).strip():
+        raise RuntimeError("leader_create_mleader: text must not be empty")
+    normalized = []
+    for point in points:
+        if len(point) < 2:
+            raise RuntimeError("leader_create_mleader: each point must contain x and y")
+        normalized.append((float(point[0]), float(point[1])))
+    return normalized
@@ -0,0 +1,320 @@
+"""Auditable drawing delivery with artifact hashes and DXF reopen parity."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import subprocess
+import time
+from collections import Counter
+from dataclasses import dataclass
+from datetime import UTC, datetime
+from pathlib import Path
+from typing import TYPE_CHECKING, Any
+
+from version import __version__
+
+from .scoring import combine
+from .validator import DrawingValidator
+
+if TYPE_CHECKING:
+    from backends.base import AutoCADBackend
+
+
+_FORMAT_FILENAMES = {
+    "dxf": "drawing.dxf",
+    "pdf": "drawing.pdf",
+    "png": "preview.png",
+}
+
+
+@dataclass
+class DeliveryResult:
+    status: str
+    output_dir: str
+    manifest_path: str
+    artifacts: list[dict]
+    score: dict
+    parity: dict
+
+    def to_dict(self) -> dict:
+        return {
+            "status": self.status,
+            "output_dir": self.output_dir,
+            "manifest_path": self.manifest_path,
+            "artifacts": list(self.artifacts),
+            "score": dict(self.score),
+            "parity": dict(self.parity),
+        }
+
+
+def _sha256(path: Path) -> str:
+    digest = hashlib.sha256()
+    with path.open("rb") as stream:
+        for chunk in iter(lambda: stream.read(1024 * 1024), b""):
+            digest.update(chunk)
+    return digest.hexdigest()
+
+
+def _git_sha() -> str | None:
+    try:
+        result = subprocess.run(
+            ["git", "rev-parse", "HEAD"],
+            check=True,
+            capture_output=True,
+            text=True,
+            timeout=2,
+        )
+        return result.stdout.strip() or None
+    except (OSError, subprocess.SubprocessError):
+        return None
+
+
+async def drawing_snapshot(backend: AutoCADBackend) -> dict[str, Any]:
+    """Return the stable structural fields used for save/reopen comparison."""
+    await backend._ensure_document_state()
+    entities = await backend.entity_list(limit=100_000)
+    info = await backend.drawing_info()
+    listed_entity_count = len(entities)
+    reported_entity_count = int(info.entity_count)
+    return {
+        "entity_count": reported_entity_count,
+        "listed_entity_count": listed_entity_count,
+        "inventory_complete": listed_entity_count == reported_entity_count,
+        "types": dict(sorted(Counter(item.type for item in entities).items())),
+        "layers": dict(sorted(Counter(item.layer for item in entities).items())),
+        "extents_min": [float(value) for value in info.extents_min],
+        "extents_max": [float(value) for value in info.extents_max],
+    }
+
+
+def compare_drawing_snapshots(
+    source: dict[str, Any],
+    reopened: dict[str, Any],
+    *,
+    tolerance: float = 1e-6,
+) -> dict:
+    """Compare counts exactly and drawing extents within ``tolerance``."""
+    differences: list[dict] = []
+    for label, snapshot in (("source", source), ("reopened", reopened)):
+        if snapshot.get("inventory_complete") is False:
+            differences.append(
+                {
+                    "field": f"{label}_inventory_complete",
+                    "entity_count": snapshot.get("entity_count"),
+                    "listed_entity_count": snapshot.get("listed_entity_count"),
+                }
+            )
+    for field in ("entity_count", "types", "layers"):
+        if source.get(field) != reopened.get(field):
+            differences.append(
+                {"field": field, "source": source.get(field), "reopened": reopened.get(field)}
+            )
+    for field in ("extents_min", "extents_max"):
+        left = source.get(field, [])
+        right = reopened.get(field, [])
+        equal = len(left) == len(right) and all(
+            abs(float(a) - float(b)) <= tolerance for a, b in zip(left, right, strict=True)
+        )
+        if not equal:
+            differences.append({"field": field, "source": left, "reopened": right})
+    return {
+        "ok": not differences,
+        "tolerance": tolerance,
+        "source": source,
+        "reopened": reopened,
+        "differences": differences,
+    }
+
+
+def _artifact_record(fmt: str, path: Path, *, status: str, error: str | None = None) -> dict:
+    record = {
+        "format": fmt,
+        "filename": path.name,
+        "status": status,
+        "size_bytes": path.stat().st_size if path.is_file() else 0,
+        "sha256": _sha256(path) if path.is_file() else None,
+    }
+    if error:
+        record["error"] = error
+    return record
+
+
+async def _reopen_parity(source: dict, dxf_path: Path) -> dict:
+    from backends.ezdxf_backend import EzdxfBackend
+
+    reopened_backend = EzdxfBackend()
+    await reopened_backend.connect()
+    try:
+        await reopened_backend.drawing_open(str(dxf_path))
+        reopened = await drawing_snapshot(reopened_backend)
+    finally:
+        await reopened_backend.disconnect()
+    return compare_drawing_snapshots(source, reopened)
+
+
+async def deliver_drawing(
+    backend: AutoCADBackend,
+    output_dir: str | Path,
+    *,
+    formats: list[str] | None = None,
+    min_score: float = 95.0,
+    strict_critique: bool = True,
+    expected: dict | None = None,
+) -> DeliveryResult:
+    """Export, validate, hash and re-open a release-ready drawing bundle."""
+    requested = list(dict.fromkeys(item.lower().strip() for item in (formats or ["dxf", "pdf", "png"])))
+    unknown = [item for item in requested if item not in _FORMAT_FILENAMES]
+    if unknown:
+        raise ValueError(f"Unsupported delivery format(s): {', '.join(unknown)}")
+    if not 0 <= float(min_score) <= 100:
+        raise ValueError("min_score must be between 0 and 100")
+
+    destination = Path(output_dir).resolve()
+    destination.mkdir(parents=True, exist_ok=True)
+    if not destination.is_dir():
+        raise ValueError(f"Delivery output is not a directory: {destination}")
+
+    started = time.perf_counter()
+    timings: dict[str, float] = {}
+    artifacts: list[dict] = []
+    capabilities = backend.capabilities().to_dict()
+    feature_caps = capabilities["features"]
+    source = await drawing_snapshot(backend)
+
+    # DXF is always emitted: it is the canonical, re-openable parity artifact.
+    export_formats = ["dxf", *(item for item in requested if item != "dxf")]
+    for fmt in export_formats:
+        artifact_path = destination / _FORMAT_FILENAMES[fmt]
+        capability = feature_caps.get(fmt, {"supported": False, "reason": "not_declared"})
+        export_started = time.perf_counter()
+        if not capability.get("supported"):
+            artifacts.append(
+                _artifact_record(
+                    fmt,
+                    artifact_path,
+                    status="unsupported",
+                    error=str(capability.get("reason") or "backend does not support format"),
+                )
+            )
+            continue
+        try:
+            if fmt == "dxf":
+                await backend.drawing_export_dxf(str(artifact_path))
+            elif fmt == "pdf":
+                await backend.drawing_export_pdf(str(artifact_path))
+            else:
+                png = await backend.view_screenshot()
+                if not png:
+                    raise RuntimeError("backend returned no PNG bytes")
+                artifact_path.write_bytes(png)
+            artifacts.append(_artifact_record(fmt, artifact_path, status="created"))
+        except Exception as exc:
+            artifacts.append(
+                _artifact_record(fmt, artifact_path, status="failed", error=str(exc))
+            )
+        finally:
+            timings[f"export_{fmt}_ms"] = round((time.perf_counter() - export_started) * 1000, 2)
+
+    validation_started = time.perf_counter()
+    validation = await DrawingValidator().run(backend, expected=expected or {})
+    critique_issues = await backend.drawing_critique(focus=None)
+    critique_summary = {severity: 0 for severity in ("error", "warning", "info")}
+    for issue in critique_issues:
+        critique_summary[issue.severity] += 1
+
+    dxf_record = next(item for item in artifacts if item["format"] == "dxf")
+    if dxf_record["status"] == "created":
+        try:
+            parity = await _reopen_parity(source, destination / dxf_record["filename"])
+        except Exception as exc:
+            parity = {
+                "ok": False,
+                "tolerance": 1e-6,
+                "source": source,
+                "reopened": None,
+                "differences": [{"field": "reopen", "error": str(exc)}],
+            }
+    else:
+        parity = {
+            "ok": False,
+            "tolerance": 1e-6,
+            "source": source,
+            "reopened": None,
+            "differences": [{"field": "dxf", "error": "canonical DXF was not created"}],
+        }
+
+    score = combine(
+        validation.summary,
+        critique_summary,
+        {"error": 0 if parity["ok"] else 1},
+    )
+    timings["validation_ms"] = round((time.perf_counter() - validation_started) * 1000, 2)
+
+    failed_export = any(item["status"] != "created" for item in artifacts)
+    critique_blocked = any(
+        issue.severity == "error" or (strict_critique and issue.severity != "info")
+        for issue in critique_issues
+    )
+    validation_blocked = (
+        not validation.ok
+        or not parity["ok"]
+        or score["score"] < float(min_score)
+        or critique_blocked
+    )
+    if failed_export:
+        status = "failed_export"
+    elif validation_blocked:
+        status = "failed_validation"
+    else:
+        status = "success"
+
+    validation_payload = {
+        "validator": validation.to_dict(),
+        "critique": [issue.to_dict() for issue in critique_issues],
+        "critique_summary": critique_summary,
+        "score": score,
+        "parity": parity,
+    }
+    validation_path = destination / "validation.json"
+    validation_path.write_text(
+        json.dumps(validation_payload, ensure_ascii=False, indent=2, sort_keys=True),
+        encoding="utf-8",
+    )
+
+    plan = backend.get_plan_spec()
+    timings["total_ms"] = round((time.perf_counter() - started) * 1000, 2)
+    manifest = {
+        "schema_version": "1.0",
+        "status": status,
+        "version": __version__,
+        "git_sha": _git_sha(),
+        "generated_at": datetime.now(UTC).isoformat(),
+        "backend": backend.name,
+        "capabilities": capabilities,
+        "spec_hash": plan.get("spec_hash") if plan else None,
+        "requested_formats": requested,
+        "min_score": float(min_score),
+        "strict_critique": strict_critique,
+        "score": score,
+        "validator": validation.to_dict(),
+        "critique": validation_payload["critique"],
+        "critique_summary": critique_summary,
+        "parity": parity,
+        "artifacts": artifacts,
+        "validation_artifact": _artifact_record("json", validation_path, status="created"),
+        "timings_ms": timings,
+    }
+    manifest_path = destination / "manifest.json"
+    manifest_path.write_text(
+        json.dumps(manifest, ensure_ascii=False, indent=2, sort_keys=True),
+        encoding="utf-8",
+    )
+    return DeliveryResult(
+        status=status,
+        output_dir=str(destination),
+        manifest_path=str(manifest_path),
+        artifacts=artifacts,
+        score=score,
+        parity=parity,
+    )
@@ -50,6 +50,9 @@ class PlanSpec:
    view_count: int = 1
    dim_style: DimStyle = "chain"
    notes: list[str] = field(default_factory=list)
+    requirements: dict = field(default_factory=dict)
+    spec_hash: str | None = None
+    preflight_status: Literal["ready", "skipped"] = "skipped"

    def to_dict(self) -> dict:
        return {
@@ -60,6 +63,9 @@ class PlanSpec:
            "view_count": self.view_count,
            "dim_style": self.dim_style,
            "notes": list(self.notes),
+            "requirements": dict(self.requirements),
+            "spec_hash": self.spec_hash,
+            "preflight_status": self.preflight_status,
        }


@@ -0,0 +1,218 @@
+"""Pure, deterministic drawing-requirement preflight validation."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import math
+from dataclasses import dataclass, field
+from numbers import Real
+from typing import Any
+
+
+@dataclass(frozen=True)
+class PreflightQuestion:
+    code: str
+    field: str
+    question: str
+    choices: list[str] = field(default_factory=list)
+
+    def to_dict(self) -> dict:
+        return {
+            "code": self.code,
+            "field": self.field,
+            "question": self.question,
+            "choices": list(self.choices),
+        }
+
+
+@dataclass(frozen=True)
+class PreflightConflict:
+    code: str
+    field: str
+    values: list[Any]
+    message: str
+
+    def to_dict(self) -> dict:
+        return {
+            "code": self.code,
+            "field": self.field,
+            "values": list(self.values),
+            "message": self.message,
+        }
+
+
+@dataclass
+class PreflightResult:
+    ready: bool
+    normalized_spec: dict
+    questions: list[PreflightQuestion]
+    conflicts: list[PreflightConflict]
+    assumptions: list[dict]
+    spec_hash: str
+
+    def to_dict(self) -> dict:
+        return {
+            "ready": self.ready,
+            "normalized_spec": self.normalized_spec,
+            "questions": [item.to_dict() for item in self.questions],
+            "conflicts": [item.to_dict() for item in self.conflicts],
+            "assumptions": list(self.assumptions),
+            "spec_hash": self.spec_hash,
+        }
+
+
+def _normalize(value: Any) -> Any:
+    if isinstance(value, dict):
+        return {str(key).strip(): _normalize(item) for key, item in sorted(value.items())}
+    if isinstance(value, list):
+        return [_normalize(item) for item in value]
+    if isinstance(value, str):
+        return value.strip()
+    if isinstance(value, Real) and not isinstance(value, bool):
+        return float(value)
+    return value
+
+
+def _spec_hash(spec: dict) -> str:
+    encoded = json.dumps(spec, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
+    return "sha256:" + hashlib.sha256(encoded.encode("utf-8")).hexdigest()
+
+
+def preflight_drawing(
+    intent: str,
+    requirements: dict | None = None,
+    sheet_size: str = "A3",
+    scale: float = 1.0,
+    layer_set_id: str = "mech",
+    view_count: int = 1,
+    dim_style: str = "chain",
+    *,
+    allow_assumptions: bool = False,
+) -> PreflightResult:
+    """Normalize a drawing request and surface missing or conflicting facts."""
+    req = _normalize(dict(requirements or {}))
+    assumptions: list[dict] = []
+
+    if allow_assumptions and not req.get("units"):
+        req["units"] = "mm"
+        assumptions.append({"field": "units", "value": "mm", "reason": "default"})
+    if allow_assumptions and not req.get("tolerance_policy"):
+        req["tolerance_policy"] = "general"
+        assumptions.append(
+            {"field": "tolerance_policy", "value": "general", "reason": "default"}
+        )
+
+    if isinstance(req.get("units"), str):
+        req["units"] = req["units"].lower()
+    if isinstance(req.get("part_type"), str):
+        req["part_type"] = req["part_type"].lower()
+
+    questions: list[PreflightQuestion] = []
+    required = (
+        ("units", "MISSING_UNITS", "Çizim birimi nedir?", ["mm", "inch"]),
+        ("part_type", "MISSING_PART_TYPE", "Üretilecek parça/çizim türü nedir?", []),
+        ("dimensions", "MISSING_DIMENSIONS", "Ana üretim ölçüleri nelerdir?", []),
+        (
+            "tolerance_policy",
+            "MISSING_TOLERANCE_POLICY",
+            "Hangi tolerans politikası kullanılmalı?",
+            ["ISO 2768-m", "explicit", "none"],
+        ),
+    )
+    for field_name, code, question, choices in required:
+        value = req.get(field_name)
+        if value is None or value == "" or value == {} or value == []:
+            questions.append(PreflightQuestion(code, field_name, question, choices))
+
+    conflicts: list[PreflightConflict] = []
+    seen_constraints: dict[str, Any] = {}
+    for constraint in req.get("constraints", []) or []:
+        if not isinstance(constraint, dict):
+            continue
+        field_name = str(constraint.get("field", "")).strip()
+        if not field_name or "value" not in constraint:
+            continue
+        value = _normalize(constraint["value"])
+        if field_name in seen_constraints and seen_constraints[field_name] != value:
+            conflicts.append(
+                PreflightConflict(
+                    "CONFLICTING_REQUIREMENT",
+                    field_name,
+                    [seen_constraints[field_name], value],
+                    f"{field_name} için birbiriyle çelişen değerler verildi.",
+                )
+            )
+        else:
+            seen_constraints[field_name] = value
+
+    normalized_sheet = str(sheet_size).strip().upper()
+    normalized_layer_set = str(layer_set_id).strip().lower()
+    normalized_dim_style = str(dim_style).strip().lower()
+    try:
+        normalized_scale = float(scale)
+    except (TypeError, ValueError):
+        normalized_scale = float("nan")
+    try:
+        normalized_view_count = int(view_count)
+    except (TypeError, ValueError):
+        normalized_view_count = 0
+
+    parameter_rules = (
+        (
+            normalized_sheet in {"A4", "A3", "A2", "A1", "A0"},
+            "INVALID_SHEET_SIZE",
+            "sheet_size",
+            sheet_size,
+            "sheet_size must be one of A4, A3, A2, A1, A0.",
+        ),
+        (
+            math.isfinite(normalized_scale) and normalized_scale > 0,
+            "INVALID_SCALE",
+            "scale",
+            scale,
+            "scale must be a finite number greater than zero.",
+        ),
+        (
+            1 <= normalized_view_count <= 20,
+            "INVALID_VIEW_COUNT",
+            "view_count",
+            view_count,
+            "view_count must be between 1 and 20.",
+        ),
+        (
+            normalized_dim_style in {"chain", "baseline", "ordinate", "mixed"},
+            "INVALID_DIM_STYLE",
+            "dim_style",
+            dim_style,
+            "dim_style must be chain, baseline, ordinate, or mixed.",
+        ),
+        (
+            normalized_layer_set in {"iso13567", "mech", "pid"},
+            "INVALID_LAYER_SET",
+            "layer_set_id",
+            layer_set_id,
+            "layer_set_id must be iso13567, mech, or pid.",
+        ),
+    )
+    for valid, code, field_name, value, message in parameter_rules:
+        if not valid:
+            conflicts.append(PreflightConflict(code, field_name, [value], message))
+
+    normalized_spec = {
+        "intent": str(intent).strip(),
+        "requirements": req,
+        "sheet_size": normalized_sheet,
+        "scale": normalized_scale if math.isfinite(normalized_scale) else str(scale).strip(),
+        "layer_set_id": normalized_layer_set,
+        "view_count": normalized_view_count,
+        "dim_style": normalized_dim_style,
+    }
+    return PreflightResult(
+        ready=not questions and not conflicts,
+        normalized_spec=normalized_spec,
+        questions=questions,
+        conflicts=conflicts,
+        assumptions=assumptions,
+        spec_hash=_spec_hash(normalized_spec),
+    )
@@ -0,0 +1,271 @@
+"""Bounded critique -> repair -> re-critique orchestration."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+from collections.abc import Awaitable, Callable
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING
+
+from .plan_spec import Issue
+from .repairs import (
+    repair_construction_left,
+    repair_dim_overlap,
+    repair_duplicate_entities,
+    repair_iso128,
+    repair_layer_color,
+    repair_untrimmed_corner,
+)
+from .scoring import score_findings
+
+if TYPE_CHECKING:
+    from backends.base import AutoCADBackend
+
+RepairHandler = Callable[["AutoCADBackend", Issue], Awaitable[dict]]
+
+
+@dataclass
+class RepairAction:
+    focus: str
+    fingerprint: str
+    handles: list[str]
+    status: str
+    detail: dict = field(default_factory=dict)
+
+    def to_dict(self) -> dict:
+        return {
+            "focus": self.focus,
+            "fingerprint": self.fingerprint,
+            "handles": list(self.handles),
+            "status": self.status,
+            "detail": self.detail,
+        }
+
+
+@dataclass
+class RefineRound:
+    number: int
+    score_before: float
+    score_after: float
+    error_count_before: int
+    error_count_after: int
+    transaction: str
+    actions: list[RepairAction]
+
+    def to_dict(self) -> dict:
+        return {
+            "number": self.number,
+            "score_before": self.score_before,
+            "score_after": self.score_after,
+            "error_count_before": self.error_count_before,
+            "error_count_after": self.error_count_after,
+            "transaction": self.transaction,
+            "actions": [action.to_dict() for action in self.actions],
+        }
+
+
+@dataclass
+class RefineResult:
+    status: str
+    initial_score: float
+    final_score: float
+    rounds: list[RefineRound]
+    remaining_issues: list[dict]
+
+    def to_dict(self) -> dict:
+        return {
+            "status": self.status,
+            "initial_score": self.initial_score,
+            "final_score": self.final_score,
+            "rounds": [item.to_dict() for item in self.rounds],
+            "remaining_issues": list(self.remaining_issues),
+        }
+
+
+class RepairRegistry:
+    def __init__(self, *, defaults: bool = False):
+        self._handlers: dict[str, RepairHandler] = {}
+        if defaults:
+            self.register("construction_left", repair_construction_left)
+            self.register("duplicate_entities", repair_duplicate_entities)
+            self.register("layer_color", repair_layer_color)
+            self.register("iso128", repair_iso128)
+            self.register("untrimmed_corner", repair_untrimmed_corner)
+            self.register("dim_overlap", repair_dim_overlap)
+
+    def register(self, focus: str, handler: RepairHandler) -> None:
+        self._handlers[focus] = handler
+
+    def get(self, focus: str) -> RepairHandler | None:
+        return self._handlers.get(focus)
+
+
+def issue_fingerprint(issue: Issue) -> str:
+    payload = {
+        "focus": issue.focus,
+        "handles": sorted(issue.handles),
+        "detail": issue.detail,
+    }
+    encoded = json.dumps(payload, sort_keys=True, separators=(",", ":"), default=str)
+    return hashlib.sha256(encoded.encode("utf-8")).hexdigest()
+
+
+def _metrics(issues: list[Issue]) -> tuple[float, int]:
+    errors = sum(issue.severity == "error" for issue in issues)
+    warnings = sum(issue.severity == "warning" for issue in issues)
+    info = sum(issue.severity == "info" for issue in issues)
+    return float(score_findings(errors, warnings, info)["score"]), errors
+
+
+async def refine_drawing(
+    backend: AutoCADBackend,
+    max_rounds: int = 3,
+    min_score: float = 95.0,
+    focus: list[str] | None = None,
+    allowed_repairs: list[str] | None = None,
+    dry_run: bool = False,
+    *,
+    registry: RepairRegistry | None = None,
+) -> RefineResult:
+    """Apply deterministic repairs without allowing quality to regress."""
+    max_rounds = max(1, min(3, int(max_rounds)))
+    min_score = max(0.0, min(100.0, float(min_score)))
+    registry = registry or RepairRegistry(defaults=True)
+    allowed = set(allowed_repairs) if allowed_repairs is not None else None
+
+    issues = await backend.drawing_critique(focus)
+    initial_score, _initial_errors = _metrics(issues)
+    rounds: list[RefineRound] = []
+
+    if initial_score >= min_score:
+        return RefineResult(
+            "threshold_met", initial_score, initial_score, [], [item.to_dict() for item in issues]
+        )
+
+    if dry_run:
+        actions = []
+        for issue in issues:
+            handler = registry.get(issue.focus)
+            permitted = allowed is None or issue.focus in allowed
+            actions.append(
+                RepairAction(
+                    issue.focus,
+                    issue_fingerprint(issue),
+                    list(issue.handles),
+                    "planned" if handler is not None and permitted else "manual_required",
+                )
+            )
+        score, errors = _metrics(issues)
+        rounds.append(RefineRound(1, score, score, errors, errors, "not_started", actions))
+        return RefineResult("dry_run", initial_score, score, rounds, [i.to_dict() for i in issues])
+
+    status = "max_rounds"
+    for number in range(1, max_rounds + 1):
+        score_before, errors_before = _metrics(issues)
+        candidates: list[tuple[Issue, RepairHandler]] = []
+        for issue in issues:
+            handler = registry.get(issue.focus)
+            if handler is not None and (allowed is None or issue.focus in allowed):
+                candidates.append((issue, handler))
+        if not candidates:
+            status = "manual_required"
+            break
+
+        transaction = await backend.transaction_begin()
+        actions: list[RepairAction] = []
+        if not isinstance(transaction, dict) or not transaction.get("ok"):
+            detail = transaction if isinstance(transaction, dict) else {"result": transaction}
+            actions.append(RepairAction("internal", "", [], "failed", detail))
+            rounds.append(
+                RefineRound(
+                    number,
+                    score_before,
+                    score_before,
+                    errors_before,
+                    errors_before,
+                    "not_started",
+                    actions,
+                )
+            )
+            status = "transaction_unavailable"
+            break
+        try:
+            for issue, handler in candidates:
+                detail = await handler(backend, issue)
+                actions.append(
+                    RepairAction(
+                        issue.focus,
+                        issue_fingerprint(issue),
+                        list(issue.handles),
+                        "applied",
+                        detail,
+                    )
+                )
+            after = await backend.drawing_critique(focus)
+            score_after, errors_after = _metrics(after)
+        except Exception as exc:
+            await backend.transaction_rollback()
+            actions.append(RepairAction("internal", "", [], "failed", {"error": str(exc)}))
+            rounds.append(
+                RefineRound(
+                    number,
+                    score_before,
+                    score_before,
+                    errors_before,
+                    errors_before,
+                    "rolled_back",
+                    actions,
+                )
+            )
+            status = "repair_failed"
+            issues = await backend.drawing_critique(focus)
+            break
+
+        if score_after < score_before or errors_after > errors_before:
+            await backend.transaction_rollback()
+            rounds.append(
+                RefineRound(
+                    number,
+                    score_before,
+                    score_before,
+                    errors_before,
+                    errors_before,
+                    "rolled_back",
+                    actions,
+                )
+            )
+            status = "rolled_back"
+            issues = await backend.drawing_critique(focus)
+            break
+
+        await backend.transaction_commit()
+        rounds.append(
+            RefineRound(
+                number,
+                score_before,
+                score_after,
+                errors_before,
+                errors_after,
+                "committed",
+                actions,
+            )
+        )
+        before_fingerprints = {issue_fingerprint(item) for item in issues}
+        after_fingerprints = {issue_fingerprint(item) for item in after}
+        issues = after
+        if score_after >= min_score:
+            status = "threshold_met"
+            break
+        if after_fingerprints == before_fingerprints:
+            status = "no_progress"
+            break
+
+    final_score, _final_errors = _metrics(issues)
+    return RefineResult(
+        status,
+        initial_score,
+        final_score,
+        rounds,
+        [item.to_dict() for item in issues],
+    )
@@ -0,0 +1,81 @@
+"""Deterministic repair handlers for the bounded drawing refiner."""
+
+from __future__ import annotations
+
+import math
+from typing import TYPE_CHECKING
+
+from .plan_spec import ISO_128_LINEWEIGHTS_MM, Issue
+
+if TYPE_CHECKING:
+    from backends.base import AutoCADBackend
+
+
+async def repair_construction_left(backend: AutoCADBackend, issue: Issue) -> dict:
+    deleted = []
+    for handle in issue.handles:
+        await backend.entity_delete(handle)
+        deleted.append(handle)
+    return {"deleted_handles": deleted}
+
+
+async def repair_duplicate_entities(backend: AutoCADBackend, issue: Issue) -> dict:
+    if len(issue.handles) < 2:
+        raise RuntimeError("duplicate_entities repair requires two handles")
+    handle = issue.handles[-1]
+    await backend.entity_delete(handle)
+    return {"deleted_handle": handle, "kept_handle": issue.handles[0]}
+
+
+async def repair_layer_color(backend: AutoCADBackend, issue: Issue) -> dict:
+    layer = str(issue.detail["layer"])
+    expected = int(issue.detail["expected"])
+    await backend.layer_modify(layer, color=expected)
+    return {"layer": layer, "color": expected}
+
+
+async def repair_iso128(backend: AutoCADBackend, issue: Issue) -> dict:
+    layer = str(issue.detail["layer"])
+    raw = float(issue.detail["lineweight"])
+    millimetres = raw / 100.0 if raw > 2.05 else raw
+    nearest = min(ISO_128_LINEWEIGHTS_MM, key=lambda item: abs(item - millimetres))
+    await backend.layer_modify(layer, lineweight=nearest)
+    return {"layer": layer, "lineweight_mm": nearest}
+
+
+def _endpoint_kwargs(entity, old_point: list[float], new_point: tuple[float, float]) -> dict:
+    start = entity.properties.get("start")
+    end = entity.properties.get("end")
+    if not start or not end:
+        raise RuntimeError(f"LINE {entity.handle} is missing endpoints")
+    old = (float(old_point[0]), float(old_point[1]))
+    if math.dist((float(start[0]), float(start[1])), old) <= math.dist(
+        (float(end[0]), float(end[1])), old
+    ):
+        return {"x1": new_point[0], "y1": new_point[1]}
+    return {"x2": new_point[0], "y2": new_point[1]}
+
+
+async def repair_untrimmed_corner(backend: AutoCADBackend, issue: Issue) -> dict:
+    if len(issue.handles) != 2 or "p1" not in issue.detail or "p2" not in issue.detail:
+        raise RuntimeError("untrimmed_corner repair requires two handles and endpoint details")
+    p1 = issue.detail["p1"]
+    p2 = issue.detail["p2"]
+    midpoint = ((float(p1[0]) + float(p2[0])) / 2, (float(p1[1]) + float(p2[1])) / 2)
+    first = await backend.entity_get(issue.handles[0])
+    second = await backend.entity_get(issue.handles[1])
+    await backend.entity_edit_geometry(
+        issue.handles[0], **_endpoint_kwargs(first, p1, midpoint)
+    )
+    await backend.entity_edit_geometry(
+        issue.handles[1], **_endpoint_kwargs(second, p2, midpoint)
+    )
+    return {"handles": list(issue.handles), "corner": list(midpoint)}
+
+
+async def repair_dim_overlap(backend: AutoCADBackend, issue: Issue) -> dict:
+    if len(issue.handles) < 2:
+        raise RuntimeError("dim_overlap repair requires two handles")
+    handle = issue.handles[-1]
+    await backend.entity_move(handle, 0.0, 6.0)
+    return {"moved_handle": handle, "dy": 6.0}
@@ -1,6 +1,6 @@
 [project]
 name = "autocad-mcp-pro"
-version = "1.2.0"
+version = "1.3.0"
 description = "Production-grade AutoCAD MCP server — COM API + ezdxf dual engine"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -25,7 +25,7 @@ full = [
 ]

 [project.scripts]
-autocad-mcp = "server:mcp"
+autocad-mcp = "server:main"

 [tool.uv]
 dev-dependencies = [
@@ -42,7 +42,7 @@ build-backend = "hatchling.build"
 [tool.hatch.build.targets.wheel]
 # Flat layout: no directory matches the project name, so hatchling needs an
 # explicit file selection (GH issue #2). server.py imports config + security.
-only-include = ["server.py", "config.py", "security.py", "backends", "engineering"]
+only-include = ["server.py", "config.py", "security.py", "version.py", "backends", "engineering"]

 [tool.ruff]
 line-length = 100
@@ -10,6 +10,7 @@ Usage:

 from __future__ import annotations

+import argparse
 import json
 import logging
 import math
@@ -31,6 +32,7 @@ from pydantic import Field

 import config
 from security import sanitize_command, sanitize_lisp, validate_path
+from version import __version__

 # ---------------------------------------------------------------------------
 # Logging
@@ -698,6 +700,49 @@ async def entity_create_mtext(
    return _dc(result)


+@mcp.tool(
+    annotations={"title": "Create Table", "readOnlyHint": False},
+    tags={"entity", "create", "annotation", "table"},
+)
+async def entity_create_table(
+    x: Annotated[float, "Top-left X coordinate"],
+    y: Annotated[float, "Top-left Y coordinate"],
+    rows: Annotated[list[list[str]], "Data rows; every row must have the same length"],
+    headers: Annotated[list[str] | None, "Optional column header row"] = None,
+    column_widths: Annotated[list[float] | None, "Optional explicit widths per column"] = None,
+    row_height: Annotated[float, Field(default=7.0, gt=0)] = 7.0,
+    text_height: Annotated[float, Field(default=2.5, gt=0)] = 2.5,
+    title: Annotated[str | None, "Optional title row"] = None,
+    layer: Annotated[str, "Target layer"] = "TEXT",
+    ctx: Context = None,
+) -> dict:
+    """Create a native COM table or an explicitly-labelled ezdxf composite table."""
+    result = await _backend(ctx).entity_create_table(
+        x, y, rows, headers, column_widths, row_height, text_height, title, layer
+    )
+    return _dc(result)
+
+
+@mcp.tool(
+    annotations={"title": "Create Multileader", "readOnlyHint": False},
+    tags={"entity", "create", "annotation", "leader"},
+)
+async def leader_create_mleader(
+    points: Annotated[list[list[float]], "Leader vertices as [x, y] pairs"],
+    text: Annotated[str, "Leader annotation text"],
+    text_height: Annotated[float, Field(default=2.5, gt=0)] = 2.5,
+    landing_gap: Annotated[float, Field(default=1.0, ge=0)] = 1.0,
+    arrow_size: Annotated[float, Field(default=2.5, gt=0)] = 2.5,
+    layer: Annotated[str, "Target layer"] = "DIM",
+    ctx: Context = None,
+) -> dict:
+    """Create a native COM MLeader or an explicitly-labelled ezdxf composite."""
+    result = await _backend(ctx).leader_create_mleader(
+        points, text, text_height, landing_gap, arrow_size, layer
+    )
+    return _dc(result)
+
+
@mcp.tool(
    annotations={"title": "Create Hatch", "readOnlyHint": False},
    tags={"entity", "create"},
@@ -2230,6 +2275,15 @@ async def system_status(ctx: Context = None) -> dict:
    return status


+@mcp.tool(
+    annotations={"title": "Backend Capabilities", "readOnlyHint": True},
+    tags={"system", "query"},
+)
+async def system_capabilities(ctx: Context = None) -> dict:
+    """Return machine-readable support modes for the active backend."""
+    return _backend(ctx).capabilities().to_dict()
+
+
@mcp.tool(
    annotations={"title": "Get System Variable", "readOnlyHint": True},
    tags={"system"},
@@ -2335,11 +2389,12 @@ async def system_about(ctx: Context = None) -> dict:
    tool_count = await _registered_tool_count()
    out = {
        "name": "AutoCAD MCP Pro",
-        "version": "1.0.0",
+        "version": __version__,
        "description": "Production-grade AutoCAD MCP server with dual COM+ezdxf engine",
        "active_backend": backend_name,
        "tool_groups": await _tool_groups(),
        "unsafe_mode": config.settings.dangerous_commands_enabled,
+        "capabilities": b.capabilities().to_dict()["features"] if b else {},
    }
    # R20: omit total_tools when unknown rather than reporting a fake -1.
    if tool_count is not None:
@@ -2601,6 +2656,55 @@ async def drawing_finalize(
    return payload


+@mcp.tool(
+    annotations={
+        "title": "Drawing: Deliver Auditable Bundle",
+        "destructiveHint": True,
+    },
+    tags={"engineering", "drawing", "validation", "delivery"},
+)
+async def drawing_deliver(
+    output_dir: Annotated[
+        str,
+        Field(description="Output directory for drawing artifacts and manifest.json."),
+    ],
+    formats: Annotated[
+        list[str] | None,
+        Field(default=None, description="Requested formats: dxf, pdf, png."),
+    ] = None,
+    min_score: Annotated[
+        float,
+        Field(default=95.0, ge=0.0, le=100.0, description="Minimum accepted quality score."),
+    ] = 95.0,
+    strict_critique: Annotated[
+        bool,
+        Field(default=True, description="Block delivery on every non-info critique issue."),
+    ] = True,
+    expected: Annotated[
+        dict | None,
+        Field(default=None, description="Optional validator expectations for the drawing."),
+    ] = None,
+    ctx: Context = None,
+) -> dict:
+    """Create a hashed, validated delivery bundle and verify DXF save/reopen parity.
+
+    The result status is ``success``, ``failed_validation`` or ``failed_export``.
+    Failure intentionally keeps all generated artifacts for diagnosis.
+    """
+    from engineering import deliver_drawing
+
+    destination = validate_path(output_dir, allow_write=True)
+    result = await deliver_drawing(
+        _backend(ctx),
+        destination,
+        formats=formats,
+        min_score=min_score,
+        strict_critique=strict_critique,
+        expected=expected,
+    )
+    return result.to_dict()
+
+
 # ---------------------------------------------------------------------------
 # ── SECTION 13: Premium Drafting Meta-Tools (5x quality multiplier) ─────────
 # ---------------------------------------------------------------------------
@@ -2608,6 +2712,45 @@ async def drawing_finalize(
 #   plan → snap-aware geometry → critique → finalize.
 # See `.claude/skills/autocad-mcp-premium/` for the full discipline.

+@mcp.tool(
+    annotations={"title": "Drawing: Preflight", "readOnlyHint": True},
+    tags={"premium", "planning", "validation"},
+)
+async def drawing_preflight(
+    intent: Annotated[str, "One-line description of what this drawing represents."],
+    requirements: Annotated[
+        dict | None,
+        "Units, part_type, dimensions, tolerance_policy and optional constraints.",
+    ] = None,
+    sheet_size: Annotated[
+        str, Field(default="A3", description="A4 / A3 / A2 / A1 / A0.")
+    ] = "A3",
+    scale: Annotated[float, Field(default=1.0, gt=0)] = 1.0,
+    layer_set_id: Annotated[
+        str, Field(default="mech", description="mech / pid / iso13567.")
+    ] = "mech",
+    view_count: Annotated[int, Field(default=1, ge=1)] = 1,
+    dim_style: Annotated[
+        str, Field(default="chain", description="chain / baseline / ordinate / mixed.")
+    ] = "chain",
+    allow_assumptions: Annotated[
+        bool, "Allow documented defaults for units and tolerance policy."
+    ] = False,
+    ctx: Context = None,
+) -> dict:
+    """Validate and normalize requirements before committing a drawing plan."""
+    result = await _backend(ctx).drawing_preflight(
+        intent,
+        requirements,
+        sheet_size,
+        scale,
+        layer_set_id,
+        view_count,
+        dim_style,
+        allow_assumptions,
+    )
+    return result.to_dict()
+
@mcp.tool(
    annotations={"title": "Drawing: Plan (commit intent before drawing)",
                 "destructiveHint": False},
@@ -2625,6 +2768,10 @@ async def drawing_plan(
    dim_style: Annotated[str, Field(default="chain",
        description="Default dimensioning style: chain / baseline / ordinate / mixed.")] = "chain",
    notes: Annotated[list[str] | None, "Free-form constraint notes."] = None,
+    requirements: Annotated[dict | None, "Normalized preflight requirements."] = None,
+    spec_hash: Annotated[
+        str | None, "Hash returned by the latest ready drawing_preflight."
+    ] = None,
    ctx: Context = None,
 ) -> dict:
    """Commit a PlanSpec before any geometry is created.
@@ -2635,6 +2782,7 @@ async def drawing_plan(
    """
    plan = await _backend(ctx).drawing_plan(
        intent, sheet_size, scale, layer_set_id, view_count, dim_style, notes,
+        requirements, spec_hash,
    )
    return _dc(plan)

@@ -2658,6 +2806,34 @@ async def drawing_critique(
    return [_dc(i) for i in issues]


+@mcp.tool(
+    annotations={"title": "Drawing: Refine", "destructiveHint": True},
+    tags={"premium", "validation", "modify"},
+)
+async def drawing_refine(
+    max_rounds: Annotated[int, Field(default=3, ge=1, le=3)] = 3,
+    min_score: Annotated[float, Field(default=95.0, ge=0, le=100)] = 95.0,
+    focus: Annotated[list[str] | None, "Critique focuses to repair; None runs all."] = None,
+    allowed_repairs: Annotated[
+        list[str] | None, "Optional allowlist of repair focus names."
+    ] = None,
+    dry_run: Annotated[bool, "Return the repair plan without modifying the drawing."] = False,
+    ctx: Context = None,
+) -> dict:
+    """Run a bounded, transaction-safe critique/repair/re-critique loop."""
+    from engineering.refiner import refine_drawing
+
+    result = await refine_drawing(
+        _backend(ctx),
+        max_rounds=max_rounds,
+        min_score=min_score,
+        focus=focus,
+        allowed_repairs=allowed_repairs,
+        dry_run=dry_run,
+    )
+    return result.to_dict()
+
+
@mcp.tool(
    annotations={"title": "Point: Snap (deterministic OSNAP)", "readOnlyHint": True},
    tags={"premium", "snap"},
@@ -3260,24 +3436,24 @@ async def _guarded_run_async(transport=None, *args, **kwargs):
 mcp.run_async = _guarded_run_async


-if __name__ == "__main__":
-    transport = "stdio"
-    host = "127.0.0.1"
-    port = 8000
+def main() -> None:
+    """Run the MCP server from the installed ``autocad-mcp`` command."""
+    parser = argparse.ArgumentParser(description="Run the AutoCAD MCP server")
+    parser.add_argument("--transport", default="stdio")
+    parser.add_argument("--host", default="127.0.0.1")
+    parser.add_argument("--port", type=int, default=8000)
+    args = parser.parse_args()

-    args = sys.argv[1:]
-    if "--transport" in args:
-        idx = args.index("--transport")
-        transport = args[idx + 1]
-    if "--port" in args:
-        idx = args.index("--port")
-        port = int(args[idx + 1])
-    if "--host" in args:
-        idx = args.index("--host")
-        host = args[idx + 1]
-
-    if transport == "stdio":
+    if args.transport == "stdio":
        mcp.run()
    else:
-        _validate_http_bind(host)
-        mcp.run(transport=transport, host=host, port=port)
+        _validate_http_bind(args.host)
+        mcp.run(
+            transport=args.transport,
+            host=args.host,
+            port=args.port,
+        )
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,108 @@
+"""Fair benchmark v2 runner contract."""
+
+from __future__ import annotations
+
+import asyncio
+import subprocess
+import sys
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from benchmarks.adapters.autocad_mcp_pro import AutoCADMCPProAdapter
+from benchmarks.adapters.base import BenchmarkAdapter, TaskResult
+from benchmarks.run_competitors import run_tasks
+from benchmarks.tasks_v2 import TaskSpec
+
+
+class FakeAdapter(BenchmarkAdapter):
+    name = "fake"
+
+    async def setup(self, artifact_dir):
+        self.artifact_dir = artifact_dir
+
+    async def run_task(self, task):
+        if task.task_id == "slow":
+            await asyncio.sleep(0.2)
+        if task.task_id == "unsupported":
+            return TaskResult(task_id=task.task_id, status="unsupported", score=0.0)
+        return TaskResult(task_id=task.task_id, status="pass", score=100.0)
+
+    async def cleanup(self):
+        return None
+
+
+@pytest.mark.asyncio
+async def test_benchmark_timeout_does_not_stop_later_tasks(tmp_path):
+    tasks = [
+        TaskSpec("slow", "reliability", "Times out", 1.0),
+        TaskSpec("next", "reliability", "Still runs", 1.0),
+    ]
+
+    report = await run_tasks(FakeAdapter(), tasks, tmp_path, timeout=0.01)
+
+    assert [item["status"] for item in report["results"]] == ["timeout", "pass"]
+    assert report["summary"]["attempted"] == 2
+    assert report["summary"]["passed"] == 1
+    assert report["git_sha"]
+    assert report["environment"]["python"]
+
+
+@pytest.mark.asyncio
+async def test_unsupported_tasks_count_as_zero_in_fixed_matrix_score(tmp_path):
+    tasks = [
+        TaskSpec("supported", "coverage", "Implemented", 1.0),
+        TaskSpec("unsupported", "coverage", "Not implemented", 9.0),
+    ]
+
+    report = await run_tasks(FakeAdapter(), tasks, tmp_path)
+
+    assert report["summary"]["score"] == 10.0
+    assert report["summary"]["supported"] == 1
+    assert report["summary"]["coverage_percent"] == 10.0
+
+
+def test_task_result_has_machine_readable_status():
+    result = TaskResult(
+        task_id="table",
+        status="unsupported",
+        score=0.0,
+        message="TABLE is unavailable",
+    )
+
+    assert result.to_dict() == {
+        "task_id": "table",
+        "status": "unsupported",
+        "score": 0.0,
+        "message": "TABLE is unavailable",
+        "duration_ms": 0.0,
+        "artifacts": [],
+        "metrics": {},
+        "stdout_summary": "",
+        "stderr_summary": "",
+    }
+
+
+@pytest.mark.asyncio
+async def test_reference_adapter_selects_real_com_backend_class(tmp_path):
+    with patch("backends.com_backend.ComBackend") as backend_class:
+        backend_class.return_value.connect = AsyncMock()
+        adapter = AutoCADMCPProAdapter(backend="com")
+
+        await adapter.setup(tmp_path)
+
+    backend_class.assert_called_once_with()
+    assert adapter.backend is backend_class.return_value
+
+
+@pytest.mark.parametrize("args", [[], ["--help"]])
+def test_correctness_suite_help_is_not_a_traceback(args):
+    completed = subprocess.run(
+        [sys.executable, "benchmarks/correctness_suite.py", *args],
+        capture_output=True,
+        text=True,
+        timeout=10,
+    )
+
+    assert completed.returncode == 0
+    assert "usage" in completed.stdout.lower()
@@ -0,0 +1,74 @@
+from __future__ import annotations
+
+from unittest.mock import patch
+
+import pytest
+
+from backends.base import CapabilityMap, FeatureCapability
+
+
+def test_capability_types_serialize_stably():
+    caps = CapabilityMap(
+        backend="demo",
+        features={"table": FeatureCapability(True, "native")},
+    )
+
+    assert caps.to_dict() == {
+        "backend": "demo",
+        "features": {
+            "table": {"supported": True, "mode": "native", "reason": None},
+        },
+    }
+
+
+@pytest.mark.asyncio
+async def test_ezdxf_capabilities_are_machine_readable(backend):
+    caps = backend.capabilities().to_dict()
+
+    assert caps["backend"] == "ezdxf"
+    assert caps["features"]["table"] == {
+        "supported": True,
+        "mode": "composite",
+        "reason": None,
+    }
+    assert caps["features"]["mleader"]["mode"] == "composite"
+    assert caps["features"]["paper_space"]["supported"] is False
+    assert caps["features"]["solid_3d"]["supported"] is False
+
+
+@pytest.mark.asyncio
+async def test_ezdxf_render_capabilities_follow_optional_dependency(backend):
+    with patch("backends.ezdxf_backend.find_spec", return_value=None):
+        caps = backend.capabilities().to_dict()
+
+    assert caps["features"]["pdf"] == {
+        "supported": False,
+        "mode": None,
+        "reason": "optional_dependency_missing:matplotlib",
+    }
+    assert caps["features"]["png"]["supported"] is False
+
+
+@pytest.mark.asyncio
+async def test_system_capabilities_tool_uses_backend_map(backend):
+    import server
+
+    class Ctx:
+        lifespan_context = {"backend": backend}
+
+    result = await server.system_capabilities(ctx=Ctx())
+
+    assert result == backend.capabilities().to_dict()
+
+
+@pytest.mark.asyncio
+async def test_system_about_uses_package_version_and_capability_summary(backend):
+    import server
+
+    class Ctx:
+        lifespan_context = {"backend": backend}
+
+    result = await server.system_about(ctx=Ctx())
+
+    assert result["version"] == "1.3.0"
+    assert result["capabilities"]["table"]["mode"] == "composite"
@@ -11,6 +11,7 @@ Strategy:
 from __future__ import annotations

 import sys
+from types import SimpleNamespace
 from typing import Any
 from unittest.mock import MagicMock, patch

@@ -19,6 +20,34 @@ import pytest
 from backends.base import normalize_lineweight
 from backends.com_backend import ComBackend

+
+@pytest.mark.asyncio
+async def test_active_document_switch_clears_com_document_state(monkeypatch):
+    backend = ComBackend()
+    documents = SimpleNamespace(Count=1)
+    app = SimpleNamespace(
+        Documents=documents,
+        ActiveDocument=SimpleNamespace(Name="First.dwg", FullName="C:/First.dwg"),
+    )
+
+    async def run_inline(function, *args, **kwargs):
+        return function(*args, **kwargs)
+
+    monkeypatch.setattr(backend, "_run", run_inline)
+    monkeypatch.setattr("backends.com_backend._acad_app", lambda: app)
+
+    await backend._ensure_document_state()
+    backend._plan_spec = object()
+    backend._preflight_result = object()
+    backend._gdt_datums_defined = {"A"}
+    app.ActiveDocument = SimpleNamespace(Name="Second.dwg", FullName="C:/Second.dwg")
+
+    await backend._ensure_document_state()
+
+    assert backend._plan_spec is None
+    assert backend._preflight_result is None
+    assert backend._gdt_datums_defined == set()
+
 # ---------------------------------------------------------------------------
 # normalize_lineweight — parametrized table
 # ---------------------------------------------------------------------------
@@ -0,0 +1,139 @@
+"""Release delivery contract: artifacts, hashes, validation and DXF parity."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+
+import pytest
+
+import server
+from engineering.delivery import compare_drawing_snapshots, deliver_drawing, drawing_snapshot
+from engineering.layers import ensure_engineering_layers, ensure_standard_linetypes
+
+
+async def _clean_drawing(backend) -> None:
+    await ensure_standard_linetypes(backend)
+    await ensure_engineering_layers(backend)
+    await backend.entity_create_line(0, 0, 40, 0, layer="GEOMETRY")
+
+
+@pytest.mark.asyncio
+async def test_delivery_writes_hashed_manifest_and_reopen_parity(backend, tmp_path):
+    await _clean_drawing(backend)
+
+    result = await deliver_drawing(
+        backend,
+        tmp_path,
+        formats=["dxf"],
+        min_score=0,
+        strict_critique=False,
+    )
+
+    assert result.status == "success"
+    manifest_path = tmp_path / "manifest.json"
+    validation_path = tmp_path / "validation.json"
+    assert manifest_path.is_file()
+    assert validation_path.is_file()
+
+    manifest = json.loads(manifest_path.read_text(encoding="utf-8"))
+    dxf = next(item for item in manifest["artifacts"] if item["format"] == "dxf")
+    artifact_path = tmp_path / dxf["filename"]
+    assert dxf["status"] == "created"
+    assert dxf["sha256"] == hashlib.sha256(artifact_path.read_bytes()).hexdigest()
+    assert manifest["parity"]["ok"] is True
+    assert manifest["version"] == "1.3.0"
+    assert manifest["capabilities"]["backend"] == "ezdxf"
+
+
+@pytest.mark.asyncio
+async def test_delivery_keeps_artifacts_when_validation_fails(backend, tmp_path):
+    await _clean_drawing(backend)
+    await backend.construction_xline(0, 0, 45)
+
+    result = await deliver_drawing(
+        backend,
+        tmp_path,
+        formats=["dxf"],
+        min_score=100,
+        strict_critique=True,
+    )
+
+    assert result.status == "failed_validation"
+    assert (tmp_path / "drawing.dxf").is_file()
+    manifest = json.loads((tmp_path / "manifest.json").read_text(encoding="utf-8"))
+    assert manifest["status"] == "failed_validation"
+    assert any(issue["focus"] == "construction_left" for issue in manifest["critique"])
+
+
+def test_snapshot_comparison_reports_type_and_bounds_drift():
+    source = {
+        "entity_count": 2,
+        "types": {"LINE": 2},
+        "layers": {"GEOMETRY": 2},
+        "extents_min": [0.0, 0.0],
+        "extents_max": [10.0, 5.0],
+    }
+    reopened = {
+        "entity_count": 2,
+        "types": {"LINE": 1, "CIRCLE": 1},
+        "layers": {"GEOMETRY": 2},
+        "extents_min": [0.0, 0.0],
+        "extents_max": [10.5, 5.0],
+    }
+
+    result = compare_drawing_snapshots(source, reopened)
+
+    assert result["ok"] is False
+    assert {item["field"] for item in result["differences"]} == {"types", "extents_max"}
+
+
+@pytest.mark.asyncio
+async def test_snapshot_parity_rejects_incomplete_entity_inventory(backend, monkeypatch):
+    await backend.entity_create_line(0, 0, 10, 0)
+    await backend.entity_create_circle(5, 5, 2)
+    original_entity_list = backend.entity_list
+
+    async def incomplete_entity_list(*args, **kwargs):
+        return (await original_entity_list(*args, **kwargs))[:1]
+
+    monkeypatch.setattr(backend, "entity_list", incomplete_entity_list)
+    source = await drawing_snapshot(backend)
+    reopened = dict(source, inventory_complete=True, listed_entity_count=2)
+
+    result = compare_drawing_snapshots(source, reopened)
+
+    assert source["entity_count"] == 2
+    assert source["listed_entity_count"] == 1
+    assert source["inventory_complete"] is False
+    assert result["ok"] is False
+    assert "source_inventory_complete" in {
+        item["field"] for item in result["differences"]
+    }
+
+
+@pytest.mark.asyncio
+async def test_delivery_rejects_unknown_format(backend, tmp_path):
+    await _clean_drawing(backend)
+
+    with pytest.raises(ValueError, match="Unsupported delivery format"):
+        await deliver_drawing(backend, tmp_path, formats=["step"])
+
+
+@pytest.mark.asyncio
+async def test_server_drawing_deliver_exposes_manifest_tool(backend, tmp_path):
+    await _clean_drawing(backend)
+
+    class Ctx:
+        lifespan_context = {"backend": backend}
+
+    result = await server.drawing_deliver(
+        output_dir=str(tmp_path),
+        formats=["dxf"],
+        min_score=0,
+        strict_critique=False,
+        ctx=Ctx(),
+    )
+
+    assert result["status"] == "success"
+    assert result["manifest_path"].endswith("manifest.json")
@@ -0,0 +1,166 @@
+from __future__ import annotations
+
+import pytest
+
+from engineering.preflight import preflight_drawing
+
+
+def _complete_requirements() -> dict:
+    return {
+        "units": "mm",
+        "part_type": "bracket",
+        "dimensions": {"width": 80.0, "height": 50.0, "thickness": 8.0},
+        "tolerance_policy": "ISO 2768-m",
+    }
+
+
+def test_preflight_missing_production_fields_is_not_ready():
+    result = preflight_drawing("L bracket", requirements={})
+
+    assert result.ready is False
+    assert {q.code for q in result.questions} == {
+        "MISSING_UNITS",
+        "MISSING_PART_TYPE",
+        "MISSING_DIMENSIONS",
+        "MISSING_TOLERANCE_POLICY",
+    }
+    assert result.spec_hash.startswith("sha256:")
+
+
+def test_preflight_normalizes_complete_spec_and_hash_is_deterministic():
+    first = preflight_drawing(
+        "  L bracket  ",
+        requirements=_complete_requirements(),
+        sheet_size="a3",
+        layer_set_id="MECH",
+    )
+    second = preflight_drawing(
+        "L bracket",
+        requirements={
+            "tolerance_policy": "ISO 2768-m",
+            "dimensions": {"thickness": 8, "height": 50, "width": 80},
+            "part_type": "bracket",
+            "units": "MM",
+        },
+        sheet_size="A3",
+        layer_set_id="mech",
+    )
+
+    assert first.ready is True
+    assert first.normalized_spec["requirements"]["units"] == "mm"
+    assert first.spec_hash == second.spec_hash
+
+
+def test_preflight_detects_conflicting_constraints():
+    requirements = _complete_requirements()
+    requirements["constraints"] = [
+        {"field": "dimensions.width", "value": 80},
+        {"field": "dimensions.width", "value": 90},
+    ]
+
+    result = preflight_drawing("L bracket", requirements=requirements)
+
+    assert result.ready is False
+    assert result.conflicts[0].code == "CONFLICTING_REQUIREMENT"
+    assert result.conflicts[0].field == "dimensions.width"
+
+
+@pytest.mark.parametrize(
+    ("kwargs", "code"),
+    [
+        ({"sheet_size": "LETTER"}, "INVALID_SHEET_SIZE"),
+        ({"scale": 0}, "INVALID_SCALE"),
+        ({"view_count": 0}, "INVALID_VIEW_COUNT"),
+        ({"dim_style": "guess"}, "INVALID_DIM_STYLE"),
+        ({"layer_set_id": "custom"}, "INVALID_LAYER_SET"),
+    ],
+)
+def test_preflight_rejects_invalid_planning_parameters(kwargs, code):
+    result = preflight_drawing(
+        "L bracket",
+        requirements=_complete_requirements(),
+        **kwargs,
+    )
+
+    assert result.ready is False
+    assert code in {conflict.code for conflict in result.conflicts}
+
+
+def test_preflight_does_not_invent_values_when_assumptions_disabled():
+    result = preflight_drawing(
+        "L bracket",
+        requirements={"part_type": "bracket"},
+        allow_assumptions=False,
+    )
+
+    assert result.assumptions == []
+    assert "units" not in result.normalized_spec["requirements"]
+
+
+@pytest.mark.asyncio
+async def test_backend_plan_accepts_matching_ready_preflight(backend):
+    preflight = await backend.drawing_preflight(
+        "L bracket", requirements=_complete_requirements(),
+    )
+
+    plan = await backend.drawing_plan(
+        "L bracket",
+        requirements=preflight.normalized_spec["requirements"],
+        spec_hash=preflight.spec_hash,
+    )
+
+    assert plan.preflight_status == "ready"
+    assert plan.spec_hash == preflight.spec_hash
+    assert plan.requirements["units"] == "mm"
+
+
+@pytest.mark.asyncio
+async def test_backend_plan_rejects_not_ready_preflight_hash(backend):
+    preflight = await backend.drawing_preflight("L bracket", requirements={})
+
+    with pytest.raises(RuntimeError, match="not ready"):
+        await backend.drawing_plan("L bracket", spec_hash=preflight.spec_hash)
+
+
+@pytest.mark.asyncio
+async def test_backend_plan_rejects_fields_that_do_not_match_preflight_hash(backend):
+    preflight = await backend.drawing_preflight(
+        "L bracket", requirements=_complete_requirements(),
+    )
+
+    with pytest.raises(RuntimeError, match="do not match"):
+        await backend.drawing_plan(
+            "Different part",
+            requirements=preflight.normalized_spec["requirements"],
+            spec_hash=preflight.spec_hash,
+        )
+
+
+@pytest.mark.asyncio
+async def test_legacy_plan_remains_compatible_and_marks_preflight_skipped(backend):
+    plan = await backend.drawing_plan("Legacy drawing")
+
+    assert plan.preflight_status == "skipped"
+    assert plan.spec_hash is None
+    assert plan.requirements == {}
+
+
+@pytest.mark.asyncio
+async def test_document_change_clears_plan_preflight_and_gdt_state(backend):
+    preflight = await backend.drawing_preflight(
+        "First drawing", requirements=_complete_requirements(),
+    )
+    await backend.drawing_plan(
+        "First drawing",
+        requirements=preflight.normalized_spec["requirements"],
+        spec_hash=preflight.spec_hash,
+    )
+    backend._gdt_datums_defined = {"A"}
+    backend._gdt_datums_referenced = {"A", "B"}
+
+    await backend.drawing_new()
+
+    assert backend.get_plan_spec() is None
+    assert backend._preflight_result is None
+    assert backend._gdt_datums_defined == set()
+    assert backend._gdt_datums_referenced == set()
@@ -0,0 +1,139 @@
+from __future__ import annotations
+
+import pytest
+
+from engineering.refiner import RepairRegistry, refine_drawing
+
+pytestmark = pytest.mark.asyncio
+
+
+async def test_transaction_stack_is_separate_from_drawing_undo(backend):
+    await backend.entity_create_line(0, 0, 10, 0)
+    await backend.transaction_begin()
+
+    undo = await backend.drawing_undo()
+    status = await backend.system_status()
+
+    assert undo["ok"] is False
+    assert status["transaction_depth"] == 1
+    await backend.transaction_rollback()
+
+
+async def test_transaction_rollback_restores_document(backend):
+    await backend.entity_create_line(0, 0, 10, 0)
+    await backend.transaction_begin()
+    await backend.entity_create_circle(5, 5, 2)
+
+    await backend.transaction_rollback()
+
+    entities = await backend.entity_list(limit=100)
+    assert [entity.type for entity in entities] == ["LINE"]
+
+
+async def test_refiner_does_not_mutate_when_transaction_cannot_start(backend, monkeypatch):
+    await backend.entity_create_line(0, 0, 20, 0)
+    await backend.entity_create_line(0, 0, 20, 0)
+
+    async def transaction_rejected():
+        return {"ok": False, "error": "A transaction is already active"}
+
+    async def unexpected_transaction_end():
+        raise AssertionError("refiner must not end a transaction it did not start")
+
+    monkeypatch.setattr(backend, "transaction_begin", transaction_rejected)
+    monkeypatch.setattr(backend, "transaction_commit", unexpected_transaction_end)
+    monkeypatch.setattr(backend, "transaction_rollback", unexpected_transaction_end)
+
+    result = await refine_drawing(backend, focus=["duplicate_entities"])
+
+    assert result.status == "transaction_unavailable"
+    assert result.rounds[0].transaction == "not_started"
+    assert len(await backend.entity_list(type_filter="LINE")) == 2
+
+
+async def test_refiner_repairs_construction_and_duplicates(backend):
+    await backend.entity_create_line(0, 0, 20, 0)
+    await backend.entity_create_line(0, 0, 20, 0)
+    await backend.construction_xline(0, 0, 0)
+
+    result = await refine_drawing(
+        backend,
+        max_rounds=3,
+        min_score=95,
+        focus=["duplicate_entities", "construction_left"],
+    )
+
+    assert result.status == "threshold_met"
+    assert result.final_score == 100.0
+    assert result.rounds[0].score_after > result.rounds[0].score_before
+    assert await backend.drawing_critique(
+        ["duplicate_entities", "construction_left"]
+    ) == []
+
+
+async def test_refiner_dry_run_does_not_mutate_drawing(backend):
+    await backend.entity_create_line(0, 0, 20, 0)
+    await backend.entity_create_line(0, 0, 20, 0)
+
+    result = await refine_drawing(
+        backend,
+        focus=["duplicate_entities"],
+        dry_run=True,
+    )
+
+    assert result.status == "dry_run"
+    assert len(await backend.entity_list(type_filter="LINE")) == 2
+    assert result.rounds[0].actions[0].status == "planned"
+
+
+async def test_refiner_rolls_back_when_custom_repair_lowers_score(backend):
+    await backend.entity_create_line(0, 0, 20, 0)
+    await backend.entity_create_line(0, 0, 20, 0)
+
+    registry = RepairRegistry()
+
+    async def harmful_repair(active_backend, issue):
+        await active_backend.construction_xline(0, 0, 0)
+        return {"created": "construction"}
+
+    registry.register("duplicate_entities", harmful_repair)
+    before = len(await backend.entity_list(limit=100))
+
+    result = await refine_drawing(
+        backend,
+        focus=["duplicate_entities", "construction_left"],
+        registry=registry,
+    )
+
+    assert result.status == "rolled_back"
+    assert result.rounds[0].transaction == "rolled_back"
+    assert len(await backend.entity_list(limit=100)) == before
+
+
+async def test_refiner_leaves_missing_gdt_datum_for_manual_resolution(backend):
+    backend._gdt_datums_referenced = {"A"}
+
+    result = await refine_drawing(backend, focus=["gdt"])
+
+    assert result.status == "manual_required"
+    assert result.remaining_issues[0]["focus"] == "gdt"
+
+
+async def test_server_drawing_refine_returns_structured_payload(backend):
+    import server
+
+    class Ctx:
+        lifespan_context = {"backend": backend}
+
+    await backend.entity_create_line(0, 0, 20, 0)
+    await backend.entity_create_line(0, 0, 20, 0)
+
+    result = await server.drawing_refine(
+        max_rounds=3,
+        min_score=95,
+        focus=["duplicate_entities"],
+        ctx=Ctx(),
+    )
+
+    assert result["status"] == "threshold_met"
+    assert result["final_score"] == 100.0
@@ -0,0 +1,41 @@
+from __future__ import annotations
+
+import sys
+import tomllib
+from pathlib import Path
+from unittest.mock import MagicMock
+
+import server
+
+
+def test_packaged_console_script_targets_callable_main():
+    metadata = tomllib.loads(Path("pyproject.toml").read_text(encoding="utf-8"))
+
+    assert metadata["project"]["scripts"]["autocad-mcp"] == "server:main"
+    assert callable(server.main)
+
+
+def test_main_starts_stdio_transport(monkeypatch):
+    run = MagicMock()
+    monkeypatch.setattr(server.mcp, "run", run)
+    monkeypatch.setattr(sys, "argv", ["autocad-mcp"])
+
+    server.main()
+
+    run.assert_called_once_with()
+
+
+def test_main_help_exits_without_starting_server(monkeypatch, capsys):
+    run = MagicMock()
+    monkeypatch.setattr(server.mcp, "run", run)
+    monkeypatch.setattr(sys, "argv", ["autocad-mcp", "--help"])
+
+    try:
+        server.main()
+    except SystemExit as exc:
+        assert exc.code == 0
+    else:
+        raise AssertionError("--help must exit without starting the server")
+
+    assert "usage:" in capsys.readouterr().out.lower()
+    run.assert_not_called()
@@ -113,11 +113,12 @@ async def test_rollback_without_transaction_returns_error(backend):
 async def test_transaction_begin_commit_roundtrip(backend):
    begin = await backend.transaction_begin()
    assert begin["ok"] is True
-    assert len(backend._undo_stack) == 1
+    assert len(backend._transaction_stack) == 1
+    assert len(backend._undo_stack) == 0

    commit = await backend.transaction_commit()
    assert commit["ok"] is True
-    assert len(backend._undo_stack) == 0
+    assert len(backend._transaction_stack) == 0


 async def test_transaction_rollback_restores_snapshot(backend):
@@ -129,7 +130,7 @@ async def test_transaction_rollback_restores_snapshot(backend):

    res = await backend.transaction_rollback()
    assert res["ok"] is True
-    assert len(backend._undo_stack) == 0
+    assert len(backend._transaction_stack) == 0

    # The line is gone after rollback.
    listed = await backend.entity_list(type_filter="LINE")
@@ -0,0 +1,141 @@
+from __future__ import annotations
+
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from backends.base import EntityInfo
+from backends.com_backend import ComBackend
+
+pytestmark = pytest.mark.asyncio
+
+
+async def test_ezdxf_table_returns_composite_contract(backend):
+    result = await backend.entity_create_table(
+        10,
+        100,
+        rows=[["P-01", "4"], ["P-02", "2"]],
+        headers=["Part", "Qty"],
+        title="BOM",
+        layer="TEXT",
+    )
+
+    assert result.type == "TABLE"
+    assert result.layer == "TEXT"
+    assert result.properties["representation"] == "composite"
+    assert result.properties["logical_group_id"].startswith("table:")
+    assert len(result.properties["child_handles"]) >= 10
+    assert result.properties["bounds"] == {"min": [10.0, 72.0], "max": [30.0, 100.0]}
+
+
+async def test_ezdxf_table_survives_save_and_reopen(backend, tmp_path):
+    result = await backend.entity_create_table(
+        0, 30, rows=[["A", "1"]], headers=["Name", "Qty"]
+    )
+    path = tmp_path / "table.dxf"
+    await backend.drawing_save_as(str(path), fmt="dxf")
+    await backend.drawing_open(str(path))
+
+    entities = await backend.entity_list(limit=100)
+    handles = {entity.handle for entity in entities}
+    texts = {
+        entity.properties.get("text")
+        for entity in entities
+        if entity.type in {"TEXT", "MTEXT"}
+    }
+    assert set(result.properties["child_handles"]).issubset(handles)
+    assert {"Name", "Qty", "A", "1"}.issubset(texts)
+
+
+async def test_table_rejects_ragged_rows_and_size_limit(backend):
+    with pytest.raises(RuntimeError, match="same number of columns"):
+        await backend.entity_create_table(0, 0, rows=[["A", "B"], ["C"]])
+
+    with pytest.raises(RuntimeError, match="200 rows"):
+        await backend.entity_create_table(0, 0, rows=[["A"]] * 201)
+
+
+async def test_ezdxf_mleader_returns_composite_contract(backend):
+    result = await backend.leader_create_mleader(
+        [[0, 0], [10, 5], [30, 5]],
+        "SURFACE A",
+        layer="DIM",
+    )
+
+    assert result.type == "MLEADER"
+    assert result.properties["representation"] == "composite"
+    assert result.properties["logical_group_id"].startswith("mleader:")
+    assert result.properties["text"] == "SURFACE A"
+    assert len(result.properties["child_handles"]) == 3
+
+
+async def test_mleader_requires_two_points(backend):
+    with pytest.raises(RuntimeError, match="at least two points"):
+        await backend.leader_create_mleader([[0, 0]], "note")
+
+
+async def test_com_table_uses_native_add_table():
+    backend = ComBackend()
+
+    async def run_inline(func, *args, **kwargs):
+        return func(*args, **kwargs)
+
+    backend._run = run_inline
+    mspace = MagicMock()
+    table = MagicMock()
+    table.Handle = "A1"
+    mspace.AddTable.return_value = table
+    info = EntityInfo("A1", "ACAD_TABLE", "TEXT", 256, "ByLayer", True, {})
+
+    with (
+        patch("backends.com_backend._msp", return_value=mspace),
+        patch("backends.com_backend._apoint", side_effect=lambda *p: tuple(p)),
+        patch("backends.com_backend._entity_info", return_value=info),
+        patch("backends.com_backend._apply_entity_attrs"),
+        patch("backends.com_backend._regen"),
+    ):
+        result = await backend.entity_create_table(
+            0, 50, [["A", "1"]], headers=["Name", "Qty"], title="BOM"
+        )
+
+    mspace.AddTable.assert_called_once()
+    assert mspace.AddTable.call_args[0][1:3] == (3, 2)
+    assert table.SetText.call_count == 6
+    assert result.type == "TABLE"
+    assert result.properties["representation"] == "native"
+
+
+async def test_com_mleader_uses_native_add_mleader():
+    backend = ComBackend()
+
+    async def run_inline(func, *args, **kwargs):
+        return func(*args, **kwargs)
+
+    backend._run = run_inline
+    mspace = MagicMock()
+    leader = MagicMock()
+    leader.Handle = "B1"
+    mspace.AddMLeader.return_value = leader
+    info = EntityInfo("B1", "ACAD_MLEADER", "DIM", 256, "ByLayer", True, {})
+
+    with (
+        patch("backends.com_backend._msp", return_value=mspace),
+        patch("backends.com_backend._av", side_effect=lambda values: values),
+        patch("backends.com_backend._entity_info", return_value=info),
+        patch("backends.com_backend._apply_entity_attrs"),
+        patch("backends.com_backend._regen"),
+    ):
+        result = await backend.leader_create_mleader([[0, 0], [20, 5]], "NOTE")
+
+    mspace.AddMLeader.assert_called_once_with([0.0, 0.0, 0.0, 20.0, 5.0, 0.0], 0)
+    assert leader.TextString == "NOTE"
+    assert result.type == "MLEADER"
+    assert result.properties["representation"] == "native"
+
+
+async def test_server_registers_table_and_mleader_tools():
+    import server
+
+    names = {tool.name for tool in await server._registered_tools()}
+    assert "entity_create_table" in names
+    assert "leader_create_mleader" in names
@@ -0,0 +1,20 @@
+"""Package version resolved from installed metadata or the source pyproject."""
+
+from __future__ import annotations
+
+import tomllib
+from importlib.metadata import PackageNotFoundError, version
+from pathlib import Path
+
+
+def package_version() -> str:
+    """Return the canonical project version without a duplicate constant."""
+    try:
+        return version("autocad-mcp-pro")
+    except PackageNotFoundError:
+        pyproject = Path(__file__).with_name("pyproject.toml")
+        with pyproject.open("rb") as stream:
+            return str(tomllib.load(stream)["project"]["version"])
+
+
+__version__ = package_version()
				`@@ -0,0 +1 @@`
				`"""Benchmark suites for AutoCAD MCP Pro."""`