Add benchmark contamination graph guard

[pine-harbor]

/claim #17

Summary

Adds benchmark-contamination-graph-guard, a focused Scientific Knowledge Graph Integration guard for model-to-benchmark graph edges. It detects when a model evaluation claim may be contaminated because the benchmark, held-out split, label source, or derived data appeared in training/pretraining/fine-tuning data before entity pages or AI recommendations treat the result as clean evidence.

This is intentionally narrow: it does not rebuild graph ingestion, entity extraction, ontology aliasing, geospatial provenance, organism/strain boundaries, chemical identity, software dependency provenance, calibration evidence, negative evidence, funding provenance, or generic recommendation-path explanation.

What It Covers

• Model graph edges such as trained_on, pretrained_on, fine_tuned_on, derived_from, and label_source.
• Benchmark claim edges such as evaluated_on, claims_sota_on, recommended_for, and compared_against.
• Direct training/pretraining overlap with the claimed benchmark.
• Training/evaluation split hash collisions.
• Shared training and benchmark label sources.
• Benchmark release date versus model training cutoff.
• Independent holdout evidence, split checksums, and locked evaluation-set evidence.
• Unsafe entity-page and recommendation publication while contamination risk is unresolved.

Included Artifacts

• Dependency-free Node module: benchmark-contamination-graph-guard/src/index.js
• Synthetic sample packets: benchmark-contamination-graph-guard/data/sample_benchmark_graph_packets.json
• Tests: benchmark-contamination-graph-guard/test/benchmark-contamination-graph-guard.test.js
• Demo/report generator: benchmark-contamination-graph-guard/scripts/demo.js
• Generated reviewer packet, JSON summary, SVG summary, and MP4 slate in benchmark-contamination-graph-guard/reports/

Validation

• npm.cmd test: 8 tests passed.
• npm.cmd run check: syntax checks passed.
• npm.cmd run demo: generated 4-packet report with PUBLISH: 1, REVIEW: 1, HOLD: 2.
• npm.cmd run video: generated reports/demo.mp4.
• git diff --check: passed.
• ffprobe: H.264, 1280x720, 8.000 seconds, 30 fps, 240 frames.

The MP4 is a generated FFmpeg color-source/drawtext slate, not a desktop or browser recording. It contains no private data.