fix: address phase two review and theme benchmark page

backend/app/benchmarks/agent.py

@@ -11,15 +11,27 @@ INVALID = {'TOOL_NOT_FOUND', 'TOOL_NOT_ALLOWED', 'TOOL_ARGUMENT_INVALID', 'TOOL_
 
 def score(case, run, events, latency, repeat):
     calls = [e.data for e in events if e.event.value == 'ToolCall']
-    unmatched = list(calls)
-    selected = accurate = 0
-    for expected in case.expected_tools:
-        candidates = [c for c in unmatched if c.get('name') == expected.name]
-        if not candidates:
-            continue
-        exact = next((c for c in candidates if all(k in c.get('arguments', {}) and c['arguments'][k] == v for k,v in expected.arguments.items())), None)
-        chosen = exact or candidates[0]
-        unmatched.remove(chosen); selected += 1; accurate += int(exact is not None)
+    # Maximum bipartite matching: broad parameter subsets must not consume the
+    # only call satisfying a more specific expectation. Each call is used once.
+    matched = {}
+    def assign(expected_index, visited):
+        expected = case.expected_tools[expected_index]
+        for call_index, call in enumerate(calls):
+            if call_index in visited or call.get('name') != expected.name:
+                continue
+            arguments = call.get('arguments', {})
+            if not all(key in arguments and arguments[key] == value for key, value in expected.arguments.items()):
+                continue
+            visited.add(call_index)
+            if call_index not in matched or assign(matched[call_index], visited):
+                matched[call_index] = expected_index
+                return True
+        return False
+    accurate = sum(assign(index, set()) for index in range(len(case.expected_tools)))
+    from collections import Counter
+    actual_names = Counter(call.get('name') for call in calls)
+    expected_names = Counter(tool.name for tool in case.expected_tools)
+    selected = sum(min(count, actual_names[name]) for name, count in expected_names.items())
     results = run.tool_results
     checks = {
         'completed': run.status.value == 'completed',

backend/app/export/exporters/docx.py

@@ -115,7 +115,13 @@ class DocxExporter:
             from PIL import Image
             png = node.attributes['static_png']
             with Image.open(BytesIO(png)) as image:
-                width = min(5.8, image.width / (180 if node.type == 'math_block' else 96))
+                section = self._doc.sections[-1]
+                available_width = (section.page_width - section.left_margin - section.right_margin) / 914400
+                # Leave room for Word's containing paragraph line/spacing.
+                available_height = (section.page_height - section.top_margin - section.bottom_margin) / 914400 - 0.25
+                width = min(5.8, available_width,
+                    image.width / (180 if node.type == 'math_block' else 96),
+                    available_height * image.width / image.height)
             self._doc.add_picture(BytesIO(png), width=Inches(width))
             return
         handler = getattr(self, f"_block_{node.type}", None)

backend/tests/test_phase2_completion.py

@@ -164,3 +164,36 @@ def test_repeated_static_assets_share_document_resource_budget():
     warnings = enrich_document(document)
     assert sum(bool(node.attributes.get('static_png')) for node in document.children) == 64
     assert any('预算' in warning for warning in warnings)
+
+
+@pytest.mark.parametrize('order', [(1, 2), (2, 1)])
+def test_agent_parameter_matching_is_independent_of_call_order(order):
+    from types import SimpleNamespace as NS
+    from app.benchmarks.agent import score
+    from app.contracts import AgentDatasetCase
+    case = AgentDatasetCase(case_id='overlap', prompt='test', allowed_tools=['math.add'],
+        expected_tools=[{'name':'math.add','arguments':{}}, {'name':'math.add','arguments':{'left':1}}])
+    events = [NS(event=NS(value='ToolCall'), data={'name':'math.add','arguments':{'left':value}}) for value in order]
+    run = NS(status=NS(value='completed'),tool_results=[],output='',citations=[],run_id='test',current_step=1,token_usage=0,error_code=None)
+    result = score(case, run, events, 1, 0)
+    assert result.success and result.accurate_calls == result.selected_calls == 2
+    # Two expectations cannot reuse one matching call.
+    result = score(case, run, events[:1], 1, 0)
+    assert not result.success and result.accurate_calls == 1
+
+
+@pytest.mark.parametrize('page_size', ['A4', 'Letter'])
+@pytest.mark.parametrize('dimensions', [(200, 2000), (2000, 200)])
+def test_docx_static_images_fit_both_page_dimensions(page_size, dimensions):
+    from app.export.markdown import parse_document
+    from app.export.exporters.docx import DocxExporter
+    from app.contracts import ExportOptions
+    from docx import Document
+    png = BytesIO(); Image.new('RGB', dimensions, 'white').save(png, 'PNG')
+    document = parse_document('```mermaid\nflowchart TD\n A-->B\n```')
+    document.children[0].attributes['static_png'] = png.getvalue()
+    result = DocxExporter().render(document, ExportOptions(page_size=page_size))
+    word = Document(BytesIO(result.content)); section = word.sections[0]; shape = word.inline_shapes[0]
+    assert shape.width <= section.page_width - section.left_margin - section.right_margin
+    assert shape.height < section.page_height - section.top_margin - section.bottom_margin
+    assert shape.width / shape.height == pytest.approx(dimensions[0] / dimensions[1], rel=1e-5)