🔍

Verascan Report

Data Contamination & Leakage Audit

Threshold: {{ report.threshold }} Methods: {{ report.methods_used | join(', ') }} {{ timestamp }}

{% if report.contamination_rate == 0 %}

Clean — No Contamination Detected

0 of {{ report.eval_size }} evaluation examples leaked into the training set.

{% elif report.contamination_rate < 0.03 %}

Low Risk Contamination ({{ contamination_pct }})

{{ report.total_matches }} matching {% if report.total_matches == 1 %}pair{% else %}pairs{% endif %} detected across datasets.

{% else %}

Significant Contamination Detected ({{ contamination_pct }})

{{ report.total_matches }} matching {% if report.total_matches == 1 %}pair{% else %}pairs{% endif %} detected — evaluation results may be inflated.

{% endif %}

Evaluation Set Size

Total eval samples audited

Training Set Size

Reference training corpus

Total Matches Flagged

Pairs meeting threshold

Detection Breakdown

Exact: {{ report.exact_matches }} {% if report.ngram_matches > 0 or 'ngram' in report.methods_used %} N-gram: {{ report.ngram_matches }} {% endif %} Fuzzy: {{ report.fuzzy_matches }} {% if report.semantic_matches > 0 or 'semantic' in report.methods_used %} Semantic: {{ report.semantic_matches }} {% endif %}

Across {{ report.methods_used | length }} engine(s)

{% if report.benchmarks %}

Audited Benchmarks

{% for bm in report.benchmarks %} {{ bm }}: {{ report.benchmark_counts.get(bm, 0) if report.benchmark_counts else 0 }} {% endfor %}

Contamination count per benchmark

{% endif %}

{% if matches %}

{% if report.exact_matches > 0 %} {% endif %} {% if report.ngram_matches > 0 %} {% endif %} {% if report.fuzzy_matches > 0 %} {% endif %} {% if report.semantic_matches > 0 %} {% endif %}

{% if report.benchmarks %}

Benchmark: {% for bm in report.benchmarks %} {% endfor %}

{% endif %}

{% for m in matches %}

#{{ loop.index }} {% if m.benchmark %} {{ m.benchmark }} {% endif %} {{ m.method }} Score: {{ '%.3f' | format(m.score) }}

{% if m.benchmark %}[{{ m.benchmark }}] {% endif %}Eval [#{{ m.eval_index }}] ↔ Train [#{{ m.train_index }}]

Word-Level Diff Comparison Train → Eval Diff

Evaluation Text Index #{{ m.eval_index }}

Training Text Index #{{ m.train_index }}

{% endfor %}

{% else %}

✅

No contamination detected

None of the evaluation samples exceeded the similarity threshold against the training set.

{% endif %}