russell@unturf.com
02f8dfec07
candidate_clauses() — NLI now runs only on the top-N source clauses by content-token overlap with the answer claim (max_candidate_clauses=6), not the whole context; records n_candidate_clauses / best_clause_overlap / recombination_risk. Synthetic sweep unchanged (28/28 recombination, 0/26 legit FP, mean 1.45 candidate clauses/record). Real-traffic smoke re-run: STRICT would-demote 30% → 20%, overall 47% → 33% — better, not fixed; recombination-risk split doesn't separate either. Residual STRICT false-contras at ~0.83-0.92 → θc would need ≈ 0.90 (vs the clean-set 0.5); at θc=0.90 the data in hand gives 27/28 synthetic recall, 0/26 legit FP, 0/10 smoke STRICT FP — but n=10 is too small to set on. Next: a fuller ARBORIST_NLI_SHADOW=1 bench-qa run → sweep θc on hundreds of STRICT cells → confirm → set it. θc stays 0.5; runtime NLI demotion stays off. Production verifier unchanged; falsification-hard stays 10/12.
2271 lines
No EOL
68 KiB
JSON
2271 lines
No EOL
68 KiB
JSON
{
|
|
"generated_at": "2026-05-12T18:32:06Z",
|
|
"available": true,
|
|
"reason": "ok",
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"n_records": 116,
|
|
"n_available": 116,
|
|
"elapsed_seconds": 9.3,
|
|
"would_demote_total": 47,
|
|
"would_demote_rate": 0.4052,
|
|
"mean_candidate_clauses": 1.45,
|
|
"max_candidate_clauses_cap": 6,
|
|
"by_bucket": {
|
|
"HYBRID_ENTITY": {
|
|
"n": 2,
|
|
"would_demote": 1,
|
|
"rate": 0.5
|
|
},
|
|
"STRICT_PARAPHRASE": {
|
|
"n": 1,
|
|
"would_demote": 0,
|
|
"rate": 0.0
|
|
},
|
|
"STRICT_SPAN": {
|
|
"n": 9,
|
|
"would_demote": 0,
|
|
"rate": 0.0
|
|
},
|
|
"UNGROUNDED": {
|
|
"n": 50,
|
|
"would_demote": 18,
|
|
"rate": 0.36
|
|
},
|
|
"contradiction": {
|
|
"n": 28,
|
|
"would_demote": 28,
|
|
"rate": 1.0
|
|
},
|
|
"not_contradiction": {
|
|
"n": 26,
|
|
"would_demote": 0,
|
|
"rate": 0.0
|
|
}
|
|
},
|
|
"by_recombination_risk": {
|
|
"risk": {
|
|
"n": 52,
|
|
"would_demote": 31,
|
|
"rate": 0.5962
|
|
},
|
|
"no_risk": {
|
|
"n": 64,
|
|
"would_demote": 16,
|
|
"rate": 0.25
|
|
}
|
|
},
|
|
"false_positive_probe": {
|
|
"n": 26,
|
|
"would_demote": 0,
|
|
"rate": 0.0,
|
|
"note": "would_demote on records labeled want=not_contradiction \u2014 these are shadow FALSE POSITIVES; this is \u00a77 #12 gate item 4 when the input is a real legit-answer sample"
|
|
},
|
|
"rows": [
|
|
{
|
|
"id": "5f-fal-hard-001",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0066,
|
|
"max_entailment": 0.9861,
|
|
"best_clause": "Penicillin was discovered by Alexander Fleming in 1928.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-002",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9977,
|
|
"max_entailment": 0.0004,
|
|
"best_clause": "The Eiffel Tower is in Paris, designed by Gustave Eiffel.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-003",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9926,
|
|
"max_entailment": 0.0069,
|
|
"best_clause": "Mercury is the smallest.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-004",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9932,
|
|
"max_entailment": 0.0011,
|
|
"best_clause": "Water boils at 100 degrees Celsius at sea level.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-005",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9956,
|
|
"max_entailment": 0.0017,
|
|
"best_clause": "Charles Darwin developed evolution.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-006",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9793,
|
|
"max_entailment": 0.0091,
|
|
"best_clause": "The French Revolution began in 1789, with the storming of the Bastille.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-007",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0506,
|
|
"max_entailment": 0.9136,
|
|
"best_clause": "The speed of light is approximately 300,000 kilometers per second in a vacuum.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-008",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9954,
|
|
"max_entailment": 0.0007,
|
|
"best_clause": "The Titanic sank in the North Atlantic Ocean in 1912.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-009",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9897,
|
|
"max_entailment": 0.0118,
|
|
"best_clause": "Shakespeare wrote Hamlet.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-010",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.4814,
|
|
"max_entailment": 0.0844,
|
|
"best_clause": "The mitochondria is the powerhouse of the cell.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-011",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9966,
|
|
"max_entailment": 0.4478,
|
|
"best_clause": "The capital of Australia is Canberra;",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-hard-012",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-hard-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9954,
|
|
"max_entailment": 0.0038,
|
|
"best_clause": "Gold has the chemical symbol Au;",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-001",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-002",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.986,
|
|
"max_entailment": 0.0022,
|
|
"best_clause": "Mount Everest is the tallest mountain at 8849 meters.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-003",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0066,
|
|
"max_entailment": 0.9861,
|
|
"best_clause": "Penicillin was discovered by Alexander Fleming in 1928.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-004",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0014,
|
|
"max_entailment": 0.9847,
|
|
"best_clause": "Water boils at 100 degrees Celsius at sea level.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-005",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0008,
|
|
"max_entailment": 0.9919,
|
|
"best_clause": "Hydrogen has atomic number 1 and is the lightest chemical element.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-006",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9561,
|
|
"max_entailment": 0.0096,
|
|
"best_clause": "There are eight planets in the solar system since Pluto's reclassification.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-007",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9893,
|
|
"max_entailment": 0.0031,
|
|
"best_clause": "DNA has a double helix structure first described by Watson and Crick.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-008",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0011,
|
|
"max_entailment": 0.9908,
|
|
"best_clause": "Gold has the chemical symbol Au and atomic number 79.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-009",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9945,
|
|
"max_entailment": 0.0005,
|
|
"best_clause": "Intel released a new CPU yesterday.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-010",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.001,
|
|
"max_entailment": 0.9845,
|
|
"best_clause": "The Pacific Ocean is the largest ocean on Earth.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-011",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0009,
|
|
"max_entailment": 0.9931,
|
|
"best_clause": "Sputnik 1 was the first artificial satellite launched in 1957 by the Soviet Union.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-012",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.996,
|
|
"max_entailment": 0.0003,
|
|
"best_clause": "The capital of France is Paris.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.3333,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-013",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0014,
|
|
"max_entailment": 0.9779,
|
|
"best_clause": "Plain text matching context exactly.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-014",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0035,
|
|
"max_entailment": 0.9891,
|
|
"best_clause": "Penicillin was discovered by Alexander Fleming in 1928.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-015",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0038,
|
|
"max_entailment": 0.9671,
|
|
"best_clause": "Cuneiform is one of the earliest known writing systems originating in ancient Sumer.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-016",
|
|
"bucket": "HYBRID_ENTITY",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0076,
|
|
"max_entailment": 0.9787,
|
|
"best_clause": "Mount Everest is the tallest mountain on Earth at about 8849 meters.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-017",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0039,
|
|
"max_entailment": 0.977,
|
|
"best_clause": "DNA has a double helix structure first described by Watson and Crick.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-018",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0018,
|
|
"max_entailment": 0.9887,
|
|
"best_clause": "Hydrogen has atomic number 1 and is the lightest chemical element.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-019",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0008,
|
|
"max_entailment": 0.9915,
|
|
"best_clause": "There are eight planets in the solar system.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-020",
|
|
"bucket": "STRICT_PARAPHRASE",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0009,
|
|
"max_entailment": 0.9843,
|
|
"best_clause": "Mandarin Chinese is the most widely spoken native language in the world.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-021",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0015,
|
|
"max_entailment": 0.9893,
|
|
"best_clause": "The Python ast module parses source code into an abstract syntax tree.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-022",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0063,
|
|
"max_entailment": 0.9785,
|
|
"best_clause": "Golden retrievers are dogs which are mammals.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-023",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-024",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9936,
|
|
"max_entailment": 0.0023,
|
|
"best_clause": "AMD released a new GPU yesterday.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-025",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0123,
|
|
"max_entailment": 0.9577,
|
|
"best_clause": "Pluto was reclassified as a dwarf planet in 2006.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-026",
|
|
"bucket": "HYBRID_ENTITY",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9958,
|
|
"max_entailment": 0.0011,
|
|
"best_clause": "The Nile is one of the longest rivers in the world at about 6650 kilometers.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.5,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-027",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.99,
|
|
"max_entailment": 0.0006,
|
|
"best_clause": "Sputnik 1 was launched in 1957.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-028",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.998,
|
|
"max_entailment": 0.0005,
|
|
"best_clause": "The Eiffel Tower is in Paris France.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-029",
|
|
"bucket": "STRICT_SPAN",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0011,
|
|
"max_entailment": 0.9933,
|
|
"best_clause": "Pride and Prejudice was published in 1813.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-030",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9972,
|
|
"max_entailment": 0.0003,
|
|
"best_clause": "The capital of France is Paris.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-031",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-032",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0191,
|
|
"max_entailment": 0.3648,
|
|
"best_clause": "Paris is the capital of France.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-033",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0023,
|
|
"max_entailment": 0.0033,
|
|
"best_clause": "Tigers are large mammals.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.4,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-034",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-035",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0193,
|
|
"max_entailment": 0.4569,
|
|
"best_clause": "Paris is the capital of France.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-036",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0022,
|
|
"max_entailment": 0.0039,
|
|
"best_clause": "Tigers are large mammals.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.4,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-037",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-038",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0194,
|
|
"max_entailment": 0.3792,
|
|
"best_clause": "Paris is the capital of France.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-039",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0023,
|
|
"max_entailment": 0.0039,
|
|
"best_clause": "Tigers are large mammals.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.4,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-040",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-041",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0208,
|
|
"max_entailment": 0.3534,
|
|
"best_clause": "Paris is the capital of France.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-042",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0022,
|
|
"max_entailment": 0.0034,
|
|
"best_clause": "Tigers are large mammals.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.4,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-043",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-044",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0225,
|
|
"max_entailment": 0.4006,
|
|
"best_clause": "Paris is the capital of France.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-045",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0023,
|
|
"max_entailment": 0.0036,
|
|
"best_clause": "Tigers are large mammals.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.4,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-046",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-047",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0225,
|
|
"max_entailment": 0.4071,
|
|
"best_clause": "Paris is the capital of France.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-048",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0023,
|
|
"max_entailment": 0.0035,
|
|
"best_clause": "Tigers are large mammals.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.4,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-049",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0,
|
|
"max_entailment": 0.0,
|
|
"best_clause": null,
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 0,
|
|
"best_clause_overlap": 0.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "no_candidate_clauses"
|
|
},
|
|
{
|
|
"id": "5f-fal-live-050",
|
|
"bucket": "UNGROUNDED",
|
|
"is_fp_probe": false,
|
|
"src_file": "falsification-live-v1.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0208,
|
|
"max_entailment": 0.3305,
|
|
"best_clause": "Paris is the capital of France.",
|
|
"n_clauses": 1,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-001",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9975,
|
|
"max_entailment": 0.0066,
|
|
"best_clause": "Mercury is the smallest planet in the Solar System.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-002",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.997,
|
|
"max_entailment": 0.0017,
|
|
"best_clause": "Charles Darwin developed the theory of evolution.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-003",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9806,
|
|
"max_entailment": 0.1759,
|
|
"best_clause": "The Nile is the longest river in Africa.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-004",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9922,
|
|
"max_entailment": 0.713,
|
|
"best_clause": "Mount Everest is the highest mountain above sea level.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-005",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9967,
|
|
"max_entailment": 0.0013,
|
|
"best_clause": "Christopher Marlowe wrote Doctor Faustus.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-006",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9956,
|
|
"max_entailment": 0.0018,
|
|
"best_clause": "Python was created by Guido van Rossum.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-007",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9972,
|
|
"max_entailment": 0.0007,
|
|
"best_clause": "The Arctic Ocean is the smallest ocean.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-008",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9968,
|
|
"max_entailment": 0.0011,
|
|
"best_clause": "Beijing is the capital of China.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-009",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9944,
|
|
"max_entailment": 0.003,
|
|
"best_clause": "Usain Bolt holds the world record in the 100 metres.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8333,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-010",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9853,
|
|
"max_entailment": 0.0071,
|
|
"best_clause": "The Congo Basin contains the second largest tropical rainforest.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-011",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9965,
|
|
"max_entailment": 0.0018,
|
|
"best_clause": "Sandro Botticelli painted The Birth of Venus.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-012",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9936,
|
|
"max_entailment": 0.0102,
|
|
"best_clause": "The cheetah is the fastest land animal.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-013",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.998,
|
|
"max_entailment": 0.0616,
|
|
"best_clause": "Africa is the hottest continent.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-014",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9956,
|
|
"max_entailment": 0.0023,
|
|
"best_clause": "Mozart composed The Magic Flute.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-015",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9941,
|
|
"max_entailment": 0.0055,
|
|
"best_clause": "The blue whale is the largest animal ever known to have lived.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.7143,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-016",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9979,
|
|
"max_entailment": 0.0135,
|
|
"best_clause": "Russia is the largest country by area.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-017",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9931,
|
|
"max_entailment": 0.31,
|
|
"best_clause": "Pierre Curie won a Nobel Prize in Physics.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-018",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9948,
|
|
"max_entailment": 0.026,
|
|
"best_clause": "Charles Babbage designed the Analytical Engine.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-019",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9748,
|
|
"max_entailment": 0.0374,
|
|
"best_clause": "Mercury, the closest planet to the Sun, is the smallest.",
|
|
"n_clauses": 3,
|
|
"n_candidate_clauses": 3,
|
|
"best_clause_overlap": 0.3333,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-020",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9802,
|
|
"max_entailment": 0.0096,
|
|
"best_clause": "Marie Curie was awarded the Nobel Prize in Chemistry in 1911 for her discovery of radium and polonium.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-021",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9445,
|
|
"max_entailment": 0.0153,
|
|
"best_clause": "The Great Wall of China is the longest wall ever built, stretching thousands of kilometres.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-022",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9483,
|
|
"max_entailment": 0.2615,
|
|
"best_clause": "In computing history, Linux was originally created by Linus Torvalds in 1991.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-023",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9531,
|
|
"max_entailment": 0.1426,
|
|
"best_clause": "Venus is the hottest planet in the Solar System owing to its dense carbon-dioxide atmosphere.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-024",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9916,
|
|
"max_entailment": 0.0024,
|
|
"best_clause": "The Burj Khalifa in Dubai stands 828 metres tall, making it the tallest building in the world.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.7143,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-025",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9898,
|
|
"max_entailment": 0.0011,
|
|
"best_clause": "His contemporary William Makepeace Thackeray wrote Vanity Fair.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-026",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.7589,
|
|
"max_entailment": 0.4461,
|
|
"best_clause": "Antarctica, though icy, is technically the largest desert of any kind because of its extremely low precipitation.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-027",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9832,
|
|
"max_entailment": 0.0051,
|
|
"best_clause": "Jupiter has the largest number of confirmed moons of any planet.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8333,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "recomb-028",
|
|
"bucket": "contradiction",
|
|
"is_fp_probe": false,
|
|
"src_file": "recombination.jsonl",
|
|
"available": true,
|
|
"would_demote": true,
|
|
"max_contradiction": 0.9849,
|
|
"max_entailment": 0.0071,
|
|
"best_clause": "The peregrine falcon, in a hunting dive, exceeds 320 km/h, making it the fastest animal of any kind.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-001",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.009,
|
|
"max_entailment": 0.0461,
|
|
"best_clause": "Batman lives in Gotham City.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.5714,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-002",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0038,
|
|
"max_entailment": 0.0033,
|
|
"best_clause": "Marie Curie was a physicist and chemist.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-003",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0033,
|
|
"max_entailment": 0.1003,
|
|
"best_clause": "The Eiffel Tower is located in Paris.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-004",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.003,
|
|
"max_entailment": 0.0031,
|
|
"best_clause": "Albert Einstein was born in Germany.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.7143,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-005",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0021,
|
|
"max_entailment": 0.0123,
|
|
"best_clause": "Mount Everest is the highest mountain above sea level.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8571,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-006",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0136,
|
|
"max_entailment": 0.2455,
|
|
"best_clause": "Python is a programming language.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-007",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0025,
|
|
"max_entailment": 0.0034,
|
|
"best_clause": "The Nile flows through Egypt.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.5714,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-008",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0041,
|
|
"max_entailment": 0.1253,
|
|
"best_clause": "Shakespeare was an English playwright.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-009",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0025,
|
|
"max_entailment": 0.0183,
|
|
"best_clause": "Tokyo is the capital of Japan.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.7143,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-010",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0056,
|
|
"max_entailment": 0.1995,
|
|
"best_clause": "The blue whale is a marine mammal.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-011",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0234,
|
|
"max_entailment": 0.0181,
|
|
"best_clause": "Leonardo da Vinci was an Italian polymath.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.7143,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-012",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0024,
|
|
"max_entailment": 0.0145,
|
|
"best_clause": "The Amazon River is in South America.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-013",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.002,
|
|
"max_entailment": 0.0046,
|
|
"best_clause": "Jupiter is a gas giant.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8333,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-014",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0019,
|
|
"max_entailment": 0.018,
|
|
"best_clause": "Usain Bolt is a Jamaican sprinter.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-015",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0463,
|
|
"max_entailment": 0.22,
|
|
"best_clause": "Antarctica is a continent.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-016",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0058,
|
|
"max_entailment": 0.0055,
|
|
"best_clause": "Mozart was an Austrian composer.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-017",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0216,
|
|
"max_entailment": 0.0232,
|
|
"best_clause": "Vatican City is an independent city-state.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.7143,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-018",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0053,
|
|
"max_entailment": 0.014,
|
|
"best_clause": "The cheetah is a big cat.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-019",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0068,
|
|
"max_entailment": 0.0045,
|
|
"best_clause": "It was completed in 1889 for the World's Fair.",
|
|
"n_clauses": 3,
|
|
"n_candidate_clauses": 3,
|
|
"best_clause_overlap": 0.6667,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-020",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.006,
|
|
"max_entailment": 0.0031,
|
|
"best_clause": "Marie Curie was a physicist and chemist born in Warsaw.",
|
|
"n_clauses": 3,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.5,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-021",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0335,
|
|
"max_entailment": 0.002,
|
|
"best_clause": "It is set in Denmark.",
|
|
"n_clauses": 3,
|
|
"n_candidate_clauses": 3,
|
|
"best_clause_overlap": 0.4,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-022",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0009,
|
|
"max_entailment": 0.9769,
|
|
"best_clause": "Jupiter is the largest planet in the Solar System by mass and volume.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 1,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-023",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0421,
|
|
"max_entailment": 0.0084,
|
|
"best_clause": "It is located in Dubai.",
|
|
"n_clauses": 3,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-024",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.0636,
|
|
"max_entailment": 0.0232,
|
|
"best_clause": "He also created the Git version-control system in 2005.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.8,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-025",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.1172,
|
|
"max_entailment": 0.9792,
|
|
"best_clause": "Antarctica is the largest desert overall.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 1.0,
|
|
"recombination_risk": false,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
},
|
|
{
|
|
"id": "legit-026",
|
|
"bucket": "not_contradiction",
|
|
"is_fp_probe": true,
|
|
"src_file": "legit_summary.jsonl",
|
|
"available": true,
|
|
"would_demote": false,
|
|
"max_contradiction": 0.9458,
|
|
"max_entailment": 0.9732,
|
|
"best_clause": "The peregrine falcon is the fastest animal in the world when diving.",
|
|
"n_clauses": 2,
|
|
"n_candidate_clauses": 2,
|
|
"best_clause_overlap": 0.75,
|
|
"recombination_risk": true,
|
|
"model_version": "nli-shadow-v1-minilm2-l6-h768",
|
|
"theta_contra": 0.5,
|
|
"theta_entail": 0.9,
|
|
"reason": "ok"
|
|
}
|
|
]
|
|
} |