Housekeeping: CHANGELOG + gated live integration tests (gap #7)

Add CHANGELOG.md documenting 0.0.1–0.0.19. Add tests/test_integration.py:
live COM tests (attach + list_trees + rollback-vs-MC_V_ cross-check),
skipped unless MODELCHOICE_LIVE=1 with a ModelChoice workbook open — a
regression harness for the GetActiveObject attach fix. 56 pass, 2 skipped;
ruff + mypy clean. 0.0.19.

Deferred: passing the sheet name from the build_control_panel bridge call —
the currently-installed add-in's command is 0-param (an extra arg would
break it); will wire once the AB#2615 build ships.

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

Author: vosesoft

CHANGELOG.md

@@ -0,0 +1,53 @@
+# Changelog
+
+All notable changes to `modelchoice-mcp`. Versions are tag-driven; pushing a
+`vX.Y.Z` tag publishes to PyPI via the release workflow.
+
+## 0.0.19
+- **Housekeeping** — CHANGELOG; gated live integration tests
+  (`MODELCHOICE_LIVE=1`) exercising the real COM attach/read/rollback path.
+
+## 0.0.18
+- **Knowledge resources** — `modelchoice://guide/*` curated decision-analysis
+  guidance (building trees, EVPI vs EVII, which analysis to use, pitfalls).
+
+## 0.0.17
+- **Import / export** — `export_tree_json` (a tree's raw ModelChoice JSON, to
+  save/share/version) and `import_tree_json` (validate raw JSON → write/render).
+
+## 0.0.16
+- `run_decision_report` gains **`force_to_outcome`** (what inputs would have to
+  change to force a chosen outcome).
+
+## 0.0.15
+- **`run_risk_profile`** — per-option outcome distribution (expected value, min,
+  max, std dev + the cumulative-probability table), not just the EV.
+
+## 0.0.14
+- **`run_scenarios`** — what-if comparison: named bundles of input edits, each
+  rolled back and compared to the baseline (EV, optimal decision, Δ, flips).
+
+## 0.0.13
+- **Excel attach fix** — bind to the running Excel via `GetActiveObject` (ROT)
+  instead of xlwings' window-handle walk, which failed (`0x800A01A8`) once the
+  ModelChoice add-in was loaded. Fixes timeouts/hangs from the MCP server.
+
+## 0.0.12
+- **`build_control_panel`** — lift a tree's inputs into a labelled control panel
+  at the top of its sheet, with the tree linked back to it.
+
+## 0.0.11
+- `edit_tree` gains structural ops: `add_option`, `add_branch`, `remove_branch`.
+
+## 0.0.10
+- **`run_evii`** — Expected Value of Imperfect Information for a specific test.
+
+## 0.0.9
+- **`run_decision_report`** — strategy table / policy / decision brief / MCDA.
+
+## 0.0.1 – 0.0.8
+- Phases 0–2: read + rollback engine (`list_trees`, `get_tree`, `roll_up`,
+  `verify_rollback`), build/edit (`build_tree`, `edit_tree`), and analysis
+  drivers (`run_evpi`, `run_robustness`, `run_sensitivity`, `run_analysis`,
+  `read_sheet`), plus the `/design-decision-tree` prompt. CI + tag-driven PyPI
+  release. First PyPI publish at 0.0.1.

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "modelchoice-mcp"
-version = "0.0.18"
+version = "0.0.19"
 description = "An open Model Context Protocol server for Vose Software's ModelChoice decision-tree add-in for Excel."
 readme = "README.md"
 requires-python = ">=3.11"

src/modelchoice_mcp/__init__.py

@@ -1,4 +1,4 @@
 """ModelChoice MCP — an open Model Context Protocol server for Vose
 Software's ModelChoice decision-tree add-in for Excel."""
 
-__version__ = "0.0.18"
+__version__ = "0.0.19"

tests/test_integration.py

@@ -0,0 +1,52 @@
+"""Gated live integration tests against a real Excel + ModelChoice add-in.
+
+These are SKIPPED unless MODELCHOICE_LIVE=1 and a workbook with at least one
+ModelChoice tree is open. They exercise the real COM bridge (no fakes) — the
+attach, read, and rollback path — and serve as a regression harness for the
+GetActiveObject attach fix (0.0.13). Run with:
+
+    MODELCHOICE_LIVE=1 uv run pytest tests/test_integration.py -v
+
+Read-only by default; they never write to the workbook.
+"""
+
+from __future__ import annotations
+
+import os
+
+import pytest
+
+pytestmark = pytest.mark.skipif(
+    os.environ.get("MODELCHOICE_LIVE") != "1",
+    reason="live Excel test; set MODELCHOICE_LIVE=1 with a ModelChoice workbook open",
+)
+
+
+@pytest.fixture
+def live_bridge():
+    from modelchoice_mcp.bridge import ModelChoiceBridge
+
+    return ModelChoiceBridge()
+
+
+def test_attach_and_list(live_bridge) -> None:
+    """The bridge attaches to the running Excel and finds at least one tree."""
+    trees = live_bridge.list_trees()
+    assert isinstance(trees, dict)
+    assert trees, "no ModelChoice trees in the active workbook"
+
+
+def test_roll_up_matches_rendered_cells(live_bridge) -> None:
+    """Our Python rollback agrees with the add-in's own MC_V_ cells (when the
+    tree has been rendered)."""
+    trees = live_bridge.list_trees()
+    name = next(iter(trees))
+    result = live_bridge.roll_up(name)
+    assert result.expected_value == pytest.approx(result.expected_value)
+
+    cells = live_bridge.read_node_values()
+    if not cells:
+        pytest.skip("tree not rendered (no MC_V_ cells) — open it in ModelChoice")
+    for node_id, node in result.node_results.items():
+        if node_id in cells:
+            assert node.expected_value == pytest.approx(cells[node_id], abs=0.01)

tests/test_server.py

@@ -7,7 +7,7 @@
 
 
 def test_version_set() -> None:
-    assert __version__ == "0.0.18"
+    assert __version__ == "0.0.19"
 
 
 def test_server_name() -> None: