Hippo Evaluation Scorecard

Corpus

Summary

MetricMeanMedian
recall@k0.2100.000
mrr0.2130.000
ndcg@k0.1670.000
source_diversity0.8900.971
coverage_gap0.0000.000
groundedness——
keyword_hit_rate0.0000.000
latency_ms_p504918.6—
latency_ms_p955037.3—

Stratified by enrichment_model

modelnmean recall@kmean gapmean ground
google/gemma-4-26b-a4b80.2740.000—
google/gemma-4-31b390.2100.000—
gpt-oss-120b160.2410.000—
gpt-oss-120b-mlx-crack200.2800.000—
qwen/qwen3.6-27b1—0.000—
qwen3.5-35b-a3b200.1780.000—
qwen3.6-35b-a3b-ud-mlx370.2110.000—

Caveats

Per-question

idintenttopgapdivgroundkwdegraded
q01why-decision1.0000.0001.000—✗
q02how-it-works1.0000.0001.000—✗
q03why-decision1.0000.0001.000—✗
q04how-it-works1.0000.0000.971—✗
q05state-lookup1.0000.0001.000—✗
q06why-decision1.0000.0000.783—✗
q07how-it-works1.0000.0000.971—✗
q08state-lookup1.0000.0001.000—✗
q09how-it-works1.0000.0000.971—✗
q10how-it-works1.0000.0000.971—✗
q11why-decision1.0000.0001.000—✗
q12how-it-works1.0000.0000.000—✗
q13state-lookup1.0000.0000.783—✗
q14state-lookup1.0000.0000.881—✗
q15why-decision1.0000.0000.971—✗
q16state-lookup1.0000.0000.881—✗
q17why-decision1.0000.0000.971—✗
q18how-it-works1.0000.0001.000—✗
q19state-lookup1.0000.0000.469—✗
q20how-it-works1.0000.0000.971—✗
q21why-decision1.0000.0001.000—✗
q22adversarial1.0000.0000.881—✗
q23why-decision1.0000.0000.971—✗
q24how-it-works1.0000.0000.971—✗
q25how-it-works1.0000.0000.971—✗
q26cross-source1.0000.0000.971—✗
q27how-it-works1.0000.0000.722—✗
q28why-decision1.0000.0001.000—✗
q29how-it-works1.0000.0000.722—✗
q30why-decision1.0000.0001.000—✗
q31adversarial1.0000.0000.722—✗
q32how-it-works1.0000.0000.971—✗
q33why-decision1.0000.0000.971—✗
q34state-lookup1.0000.0000.971—✗
q35adversarial1.0000.0000.881—✗
q36state-lookup1.0000.0000.971—✗
q37cross-source1.0000.0000.971—✗
q38adversarial1.0000.0000.881—✗
q39state-lookup1.0000.0000.469—✗
q40state-lookup1.0000.0000.971—✗

Coverage gaps (weakest 10 questions)