Module benchctx 

Context benchmark (docs/TEST_PLAN.md §84–87, M0/M4): score task packs against the ground-truth corpus in benchmarks/tasks.json.

Metrics:

• recall: ground-truth entities surfaced in the pack / total ground truth
• precision: ground-truth entities / pack entities of ground-truth kinds
• localization: ground-truth files mentioned in the pack content
• hallucination: nonexistent entities that must NOT surface
• budget: pack under the configured token budget

Structs

BenchSummary

BenchTask

BenchmarkCorpus

GroundTruth

Hallucination

TaskResult

Functions

locate_fixtures_dir

Locate the fixtures directory: walk up from cwd; fall back to the build-time manifest path (dev tooling).

print_summary

run_context_benchmark

Run the full context benchmark against benchmarks/tasks.json.

score_task_public