Add benchmark default matrix reporting

[ictechgy]

Summary

• add context-guard-bench report-level default_matrix with six token-reduction lanes: trimming, artifact escrow, tool pruning, cache advice, adaptive-k, and optional compression
• classify lanes as default-on, advisory, experimental, or reject/rework from existing matched-pair evidence, with lane match method, policy ceiling/clamp, reason codes, and report-only claim boundaries
• render the matrix in benchmark dashboards and update benchmark docs/sample report/tests

Ralplan evidence

• Context snapshot: .omx/context/g005-paired-replay-default-matrix-20260615T042329Z.md
• Plan: .omx/plans/ralplan-g005-paired-replay-default-matrix.md
• Architect: APPROVE, blockers none — .omx/artifacts/ralplan-g005-architect-20260615T042403Z.md
• Critic: APPROVE, blockers none — .omx/artifacts/ralplan-g005-critic-20260615T042527Z.md

Validation

• python3 scripts/sync_plugin_copies.py --check ✅
• python3 -m py_compile context-guard-kit/benchmark_runner.py plugins/context-guard/bin/context-guard-bench tests/test_context_guard_kit.py scripts/release_smoke.py ✅
• PYTHONDONTWRITEBYTECODE=1 python3 -m unittest -k benchmark tests.test_context_guard_kit.BenchmarkRunnerTests (35 tests) ✅
• evidence replay smoke for docs/benchmark-fixtures/token-savings-12task.evidence.example.jsonl emitted default_matrix JSON and ## Default matrix dashboard ✅
• python3 scripts/release_smoke.py --timeout 20 ✅
• PYTHONDONTWRITEBYTECODE=1 python3 scripts/prepublish_check.py (697 tests) ✅
• git diff --check ✅

[ictechgy]

G005 quad-review + validation evidence before merge:

• Local validation passed:
  • python3 scripts/sync_plugin_copies.py --check
  • python3 -m py_compile context-guard-kit/benchmark_runner.py plugins/context-guard/bin/context-guard-bench tests/test_context_guard_kit.py scripts/release_smoke.py
  • PYTHONDONTWRITEBYTECODE=1 python3 -m unittest -k benchmark tests.test_context_guard_kit.BenchmarkRunnerTests (35 tests)
  • 12-task replay smoke for docs/benchmark-fixtures/token-savings-12task.* verified default_matrix schema/public-claim false/6 lanes/classifications/dashboard header
  • python3 scripts/release_smoke.py --timeout 20
  • PYTHONDONTWRITEBYTECODE=1 python3 scripts/prepublish_check.py (697 tests)
  • git diff --check
• PR CI passed: test-and-prepublish on Python 3.11, Python 3.12, and macOS 3.12 (run 27524671842).
• Quad review loop: APPROVE / no blockers from Codex, Claude, Agy, and Forge.
  • Codex: .omx/artifacts/quad-review-pr206-codex-20260615T050757Z.md
  • Claude: .omx/artifacts/quad-review-pr206-claude-20260615T050757Z.md
  • Agy: .omx/artifacts/quad-review-pr206-agy-20260615T050757Z.md
  • Forge fallback: .omx/artifacts/quad-review-pr206-forge-fallback-20260615T064824Z.md

Nonblocking follow-ups captured from review: lane attribution remains key/name heuristic based for future fixtures; future non-current policy ceilings may need explicit clamp handling. Neither blocks current G005 scope.