fix: Lower resolved_threshold default from 0.8 to 0.0 for dead code benchmarks

The 80% precision AND recall gate meant every task showed "Resolved: False"
for both MCP and baseline agents. No dead code detection approach achieves
80% on both metrics simultaneously. Setting to 0.0 means any task with
non-zero P and R counts as resolved. Still configurable via config YAML.

Author: jonathanpopham

src/mcpbr/benchmarks/deadcode.py

@@ -145,7 +145,7 @@ def __init__(
         self,
         dataset: str | Path = "",
         corpus_path: str | Path | None = None,
-        resolved_threshold: float = 0.8,
+        resolved_threshold: float = 0.0,
     ):
         """Initialize the benchmark.
 

src/mcpbr/benchmarks/supermodel/benchmark.py

@@ -50,7 +50,7 @@ def __init__(
         tasks: list[dict[str, Any]] | None = None,
         supermodel_api_base: str = "https://api.supermodel.dev",
         supermodel_api_key: str | None = None,
-        resolved_threshold: float = 0.8,
+        resolved_threshold: float = 0.0,
         ground_truth_dir: str | Path | None = None,
         supermodel_api_timeout: int = 900,
         **kwargs: Any,

src/mcpbr/config.py

@@ -901,7 +901,7 @@ def validate_thinking_budget(cls, v: int | None) -> int | None:
     )
 
     resolved_threshold: float = Field(
-        default=0.8,
+        default=0.0,
         ge=0.0,
         le=1.0,
         description="Recall threshold to consider a task resolved (must be in [0.0, 1.0])",