@pharmatools/opengate 0.1.0

package/datasets/cases/case-endo-thyroid.json

@@ -0,0 +1,18 @@
+{
+  "id": "endo-thyroid",
+  "title": "Thyroid — implied and overclaim",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "More patients achieved normal TSH with the therapy than with placebo.1 The therapy significantly reduced goitre size.1",
+  "goldClaims": [
+    { "originalText": "More patients achieved normal TSH with the therapy than with placebo.1", "text": "More patients achieved normal TSH with the therapy than with placebo", "citations": [1] },
+    { "originalText": "The therapy significantly reduced goitre size.1", "text": "The therapy significantly reduced goitre size", "citations": [1] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "THYR-1 (Okoro 2023)", "text": "A total of 240 patients with goitre were randomised to the therapy or placebo for 24 weeks. Biochemical and imaging outcomes are summarised in Table 2. The proportion of patients with a normal TSH at week 24 was 71% in the therapy group and 38% in the placebo group. Mean goitre volume changed by -18% with the therapy and -4% with placebo over the study period. Adverse events were similar between the two groups." }
+  },
+  "goldVerdicts": [
+    { "claimText": "More patients achieved normal TSH with the therapy than with placebo", "citation": 1, "verdict": "implied_by_data", "rationale": "Support is a data table (71% vs 38%) with no narrative sentence stating it (rule 5)." },
+    { "claimText": "The therapy significantly reduced goitre size", "citation": 1, "verdict": "overclaim", "rationale": "Ref reports an 18% vs 4% goitre reduction but states no significance; claim asserts 'significantly' (rule 1)." }
+  ]
+}

package/datasets/cases/case-gastro-ibd.json

@@ -0,0 +1,21 @@
+{
+  "id": "gastro-ibd",
+  "title": "IBD — strong, not-supported, contradicted in one section",
+  "notes": "Synthetic; controlled ground truth spanning three verdicts. VERIFY before publishing.",
+  "manuscript": "At week 6, vedolizumab induced clinical remission in more patients than placebo.1 Vedolizumab reduced the need for surgery within one year.1 Vedolizumab was effective in patients with prior biologic failure.2",
+  "goldClaims": [
+    { "originalText": "At week 6, vedolizumab induced clinical remission in more patients than placebo.1", "text": "At week 6, vedolizumab induced clinical remission in more patients than placebo", "citations": [1] },
+    { "originalText": "Vedolizumab reduced the need for surgery within one year.1", "text": "Vedolizumab reduced the need for surgery within one year", "citations": [1] },
+    { "originalText": "Vedolizumab was effective in patients with prior biologic failure.2", "text": "Vedolizumab was effective in patients with prior biologic failure", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "GEMINI-style primary (Abadi 2024)", "text": "Clinical remission at week 6 was achieved by 38% of vedolizumab and 19% of placebo patients (p<0.001). Surgical outcomes were not evaluated in this trial." },
+    "2": { "name": "Prior-biologic subgroup", "text": "Among patients with prior biologic failure, remission rates with vedolizumab were lower than placebo and did not favour vedolizumab." }
+  },
+  "goldVerdicts": [
+    { "claimText": "At week 6, vedolizumab induced clinical remission in more patients than placebo", "citation": 1, "verdict": "strong_support", "rationale": "Ref states remission 38% vs 19% with p<0.001 explicitly." },
+    { "claimText": "Vedolizumab reduced the need for surgery within one year", "citation": 1, "verdict": "not_supported", "rationale": "Ref says surgical outcomes were not evaluated — silent (rule 2)." },
+    { "claimText": "Vedolizumab was effective in patients with prior biologic failure", "citation": 2, "verdict": "contradicted", "rationale": "Ref says remission was lower than placebo and did not favour vedolizumab — negates 'effective' (rule 1)." }
+  ]
+}

package/datasets/cases/case-hema-anemia.json

@@ -0,0 +1,19 @@
+{
+  "id": "hema-anemia",
+  "title": "Anemia — partial and contradicted",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "The agent reduced transfusion dependence and improved quality of life versus placebo.1 The agent improved overall survival.2",
+  "goldClaims": [
+    { "originalText": "The agent reduced transfusion dependence and improved quality of life versus placebo.1", "text": "The agent reduced transfusion dependence and improved quality of life versus placebo", "citations": [1] },
+    { "originalText": "The agent improved overall survival.2", "text": "The agent improved overall survival", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "ANEMIA-1 (Costa 2024)", "text": "In this trial, 410 patients with transfusion-dependent anaemia were randomised to the agent or placebo for 24 weeks; the primary endpoint was transfusion independence. Transfusion independence was achieved by 35% of agent-treated patients versus 12% with placebo (p<0.001). Iron parameters also improved with the agent. Patient-reported quality-of-life outcomes were not measured in this study." },
+    "2": { "name": "ANEMIA-1 survival", "text": "Survival was a pre-specified secondary endpoint. Over a median follow-up of 30 months, overall survival did not differ between the agent and placebo groups (hazard ratio 1.02; 95% CI 0.84-1.24)." }
+  },
+  "goldVerdicts": [
+    { "claimText": "The agent reduced transfusion dependence and improved quality of life versus placebo", "citation": 1, "verdict": "partial_support", "rationale": "Transfusion independence is explicit and significant; quality of life was not measured — compound claim, one half holds (rule 4)." },
+    { "claimText": "The agent improved overall survival", "citation": 2, "verdict": "contradicted", "rationale": "Ref says overall survival did not differ (HR 1.02) — negates an improvement (rule 1/2)." }
+  ]
+}

package/datasets/cases/case-hep-nash.json

@@ -0,0 +1,19 @@
+{
+  "id": "hep-nash",
+  "title": "NASH — implied and partial",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "At week 52, resmetirom reduced liver fat compared with placebo.1 Resmetirom resolved fibrosis and improved overall survival.2",
+  "goldClaims": [
+    { "originalText": "At week 52, resmetirom reduced liver fat compared with placebo.1", "text": "At week 52, resmetirom reduced liver fat compared with placebo", "citations": [1] },
+    { "originalText": "Resmetirom resolved fibrosis and improved overall survival.2", "text": "Resmetirom resolved fibrosis and improved overall survival", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "MAESTRO table (Bianchi 2023)", "text": "Table 2. MRI-PDFF liver fat change at week 52 (%): resmetirom -32.9, placebo -10.4." },
+    "2": { "name": "MAESTRO histology", "text": "Fibrosis resolution without worsening of NASH was achieved more often with resmetirom than placebo (26% vs 14%). Overall survival was not assessed during the trial period." }
+  },
+  "goldVerdicts": [
+    { "claimText": "At week 52, resmetirom reduced liver fat compared with placebo", "citation": 1, "verdict": "implied_by_data", "rationale": "Support is an MRI-PDFF data table with no narrative sentence (rule 5)." },
+    { "claimText": "Resmetirom resolved fibrosis and improved overall survival", "citation": 2, "verdict": "partial_support", "rationale": "Fibrosis resolution is explicit; overall survival was not assessed — compound claim, one half holds (rule 4)." }
+  ]
+}

package/datasets/cases/case-hiv-prep.json

@@ -0,0 +1,18 @@
+{
+  "id": "hiv-prep",
+  "title": "HIV prevention — strong and contradicted",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "The long-acting injectable reduced HIV acquisition compared with daily oral PrEP.1 The injectable eliminated the risk of HIV acquisition.1",
+  "goldClaims": [
+    { "originalText": "The long-acting injectable reduced HIV acquisition compared with daily oral PrEP.1", "text": "The long-acting injectable reduced HIV acquisition compared with daily oral PrEP", "citations": [1] },
+    { "originalText": "The injectable eliminated the risk of HIV acquisition.1", "text": "The injectable eliminated the risk of HIV acquisition", "citations": [1] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "PREVENT-LA (Mwangi 2024)", "text": "In this randomised trial, 4,512 participants at risk of HIV were assigned to the long-acting injectable or daily oral PrEP and followed for a median of 18 months; the primary endpoint was incident HIV infection. HIV incidence was significantly lower with the long-acting injectable than with daily oral PrEP (0.40 vs 1.18 per 100 person-years; hazard ratio 0.34; 95% CI 0.18-0.62; p<0.001). Adherence, measured by drug levels, was higher in the injectable group. Incident HIV infections nonetheless occurred in a small number of injectable recipients, several of which were associated with delayed injection visits." }
+  },
+  "goldVerdicts": [
+    { "claimText": "The long-acting injectable reduced HIV acquisition compared with daily oral PrEP", "citation": 1, "verdict": "strong_support", "rationale": "Ref states incidence was significantly lower (HR 0.34, p<0.001) explicitly." },
+    { "claimText": "The injectable eliminated the risk of HIV acquisition", "citation": 1, "verdict": "contradicted", "rationale": "Ref says incident infections still occurred — negates 'eliminated the risk' (rule 1)." }
+  ]
+}

package/datasets/cases/case-id-antibiotic.json

@@ -0,0 +1,18 @@
+{
+  "id": "id-antibiotic",
+  "title": "Antibiotic — strong (non-inferiority) and overclaim (superiority)",
+  "notes": "Synthetic; controlled ground truth. Overclaim exemplar: numerically higher cure rate, claim asserts superiority the trial did not establish. VERIFY before publishing.",
+  "manuscript": "The new antibiotic met its non-inferiority margin versus the comparator for clinical cure.1 The new antibiotic was superior to the comparator.1",
+  "goldClaims": [
+    { "originalText": "The new antibiotic met its non-inferiority margin versus the comparator for clinical cure.1", "text": "The new antibiotic met its non-inferiority margin versus the comparator for clinical cure", "citations": [1] },
+    { "originalText": "The new antibiotic was superior to the comparator.1", "text": "The new antibiotic was superior to the comparator", "citations": [1] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "ACUTE-NI (Vargas 2024)", "text": "Clinical cure was achieved in 88% of the new-antibiotic group and 85% of the comparator group; the new antibiotic met the pre-specified non-inferiority margin. The trial was not designed to test for superiority." }
+  },
+  "goldVerdicts": [
+    { "claimText": "The new antibiotic met its non-inferiority margin versus the comparator for clinical cure", "citation": 1, "verdict": "strong_support", "rationale": "Ref states the non-inferiority margin was met explicitly." },
+    { "claimText": "The new antibiotic was superior to the comparator", "citation": 1, "verdict": "overclaim", "rationale": "Cure was numerically higher (88% vs 85%) but the trial was not designed to test superiority; claim asserts an unestablished superiority, not denied (rule 1)." }
+  ]
+}

package/datasets/cases/case-nephro-ckd.json

@@ -0,0 +1,21 @@
+{
+  "id": "nephro-ckd",
+  "title": "CKD — implied, overclaim, not-supported",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "At week 52, the agent slowed the decline in eGFR compared with placebo.1 The agent was renoprotective across all stages of chronic kidney disease.1 The agent reduced cardiovascular mortality.2",
+  "goldClaims": [
+    { "originalText": "At week 52, the agent slowed the decline in eGFR compared with placebo.1", "text": "At week 52, the agent slowed the decline in eGFR compared with placebo", "citations": [1] },
+    { "originalText": "The agent was renoprotective across all stages of chronic kidney disease.1", "text": "The agent was renoprotective across all stages of chronic kidney disease", "citations": [1] },
+    { "originalText": "The agent reduced cardiovascular mortality.2", "text": "The agent reduced cardiovascular mortality", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "CKD-PROTECT table (Ndiaye 2024)", "text": "Table 1. Annual eGFR slope (mL/min/1.73m2): agent -2.1, placebo -3.9. Subgroup analyses were limited to patients in CKD stages 3 and 4." },
+    "2": { "name": "CKD-PROTECT endpoints", "text": "Cardiovascular mortality was not a pre-specified endpoint and was not analysed in this study." }
+  },
+  "goldVerdicts": [
+    { "claimText": "At week 52, the agent slowed the decline in eGFR compared with placebo", "citation": 1, "verdict": "implied_by_data", "rationale": "Support is a data table (eGFR slope -2.1 vs -3.9) with no narrative sentence stating it (rule 5)." },
+    { "claimText": "The agent was renoprotective across all stages of chronic kidney disease", "citation": 1, "verdict": "overclaim", "rationale": "Ref limits analysis to CKD stages 3-4; claim broadens to all stages (rule 3)." },
+    { "claimText": "The agent reduced cardiovascular mortality", "citation": 2, "verdict": "not_supported", "rationale": "Ref says CV mortality was not analysed — silent (rule 2)." }
+  ]
+}

package/datasets/cases/case-neuro-migraine.json

@@ -0,0 +1,39 @@
+{
+  "id": "neuro-migraine",
+  "title": "Migraine prevention — contradicted, overclaim, not-supported",
+  "notes": "Synthetic-but-realistic; controlled ground truth covering the refutation end of the scale. VERIFY before publishing.",
+  "manuscript": "The anti-CGRP antibody reduced monthly migraine days versus placebo.1 The antibody was as effective in chronic migraine as in episodic migraine.1 The antibody eliminated migraine in most patients.1 It was also effective for medication-overuse headache.2",
+  "goldClaims": [
+    {
+      "originalText": "The antibody was as effective in chronic migraine as in episodic migraine.1",
+      "text": "The antibody was as effective in chronic migraine as in episodic migraine",
+      "citations": [1]
+    },
+    {
+      "originalText": "The antibody eliminated migraine in most patients.1",
+      "text": "The antibody eliminated migraine in most patients",
+      "citations": [1]
+    },
+    {
+      "originalText": "It was also effective for medication-overuse headache.2",
+      "text": "It was also effective for medication-overuse headache",
+      "citations": [2]
+    }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": {
+      "name": "CGRP trial (Park 2024)",
+      "text": "The antibody reduced monthly migraine days versus placebo (-3.7 vs -1.8 days; p<0.001). The treatment effect was larger in episodic migraine than in chronic migraine, where the reduction over placebo was smaller and did not reach significance in the chronic subgroup. A 50% response was achieved by 41% of patients; complete cessation of migraine was uncommon."
+    },
+    "2": {
+      "name": "CGRP trial methods",
+      "text": "Patients with medication-overuse headache were excluded from the trial. Efficacy in that population was not evaluated."
+    }
+  },
+  "goldVerdicts": [
+    { "claimText": "The antibody was as effective in chronic migraine as in episodic migraine", "citation": 1, "verdict": "contradicted", "rationale": "Ref says the effect was larger in episodic and did not reach significance in the chronic subgroup — negates the 'as effective' equivalence (rule 1)." },
+    { "claimText": "The antibody eliminated migraine in most patients", "citation": 1, "verdict": "contradicted", "rationale": "Ref says complete cessation was uncommon — explicit negation of 'most patients' (rule 1). Re-labeled from overclaim in v0.1." },
+    { "claimText": "It was also effective for medication-overuse headache", "citation": 2, "verdict": "not_supported", "rationale": "Ref says MOH patients were excluded and efficacy not evaluated — silent (rule 2)." }
+  ]
+}

package/datasets/cases/case-neuro-ms.json

@@ -0,0 +1,21 @@
+{
+  "id": "neuro-ms",
+  "title": "Multiple sclerosis — strong, contradicted, implied",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "The therapy reduced the annualised relapse rate compared with the active comparator.1 The therapy halted disability progression.1 The therapy reduced the number of new MRI lesions.2",
+  "goldClaims": [
+    { "originalText": "The therapy reduced the annualised relapse rate compared with the active comparator.1", "text": "The therapy reduced the annualised relapse rate compared with the active comparator", "citations": [1] },
+    { "originalText": "The therapy halted disability progression.1", "text": "The therapy halted disability progression", "citations": [1] },
+    { "originalText": "The therapy reduced the number of new MRI lesions.2", "text": "The therapy reduced the number of new MRI lesions", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "MS-COMPARE (Fischer 2024)", "text": "The annualised relapse rate was significantly lower with the therapy than the active comparator (rate ratio 0.62; p<0.001). Confirmed disability progression did not differ significantly between the two groups." },
+    "2": { "name": "MS-COMPARE MRI", "text": "Table 3. New or enlarging T2 lesions over 96 weeks: therapy 1.2, comparator 3.8 (mean per patient)." }
+  },
+  "goldVerdicts": [
+    { "claimText": "The therapy reduced the annualised relapse rate compared with the active comparator", "citation": 1, "verdict": "strong_support", "rationale": "Ref states the relapse-rate reduction (rate ratio 0.62, p<0.001) explicitly." },
+    { "claimText": "The therapy halted disability progression", "citation": 1, "verdict": "contradicted", "rationale": "Ref says confirmed disability progression did not differ between groups — negates 'halted' (rule 1)." },
+    { "claimText": "The therapy reduced the number of new MRI lesions", "citation": 2, "verdict": "implied_by_data", "rationale": "Support is a lesion-count data table (1.2 vs 3.8) with no narrative sentence (rule 5)." }
+  ]
+}

package/datasets/cases/case-onco-pdl1.json

@@ -0,0 +1,21 @@
+{
+  "id": "onco-pdl1",
+  "title": "Immunotherapy — scope-broadening overclaim",
+  "notes": "Synthetic; controlled ground truth. Overclaim exemplar: benefit shown only in a subgroup, claim generalises to all patients while the reference is silent on the rest. Uses bracketed citations. VERIFY before publishing.",
+  "manuscript": "In patients with PD-L1-high tumours, nivolumab improved objective response rate versus chemotherapy [1]. Nivolumab improves response across all patients regardless of PD-L1 status [1]. Treatment-related deaths were rare [2].",
+  "goldClaims": [
+    { "originalText": "In patients with PD-L1-high tumours, nivolumab improved objective response rate versus chemotherapy [1].", "text": "In patients with PD-L1-high tumours, nivolumab improved objective response rate versus chemotherapy", "citations": [1] },
+    { "originalText": "Nivolumab improves response across all patients regardless of PD-L1 status [1].", "text": "Nivolumab improves response across all patients regardless of PD-L1 status", "citations": [1] },
+    { "originalText": "Treatment-related deaths were rare [2].", "text": "Treatment-related deaths were rare", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "CHECK-1 (Mori 2023)", "text": "Among patients with PD-L1-high tumours, the objective response rate was higher with nivolumab than chemotherapy (48% vs 27%). Outcomes in PD-L1-low patients were not reported in this analysis." },
+    "2": { "name": "CHECK-1 safety", "text": "Treatment-related deaths occurred in fewer than 1% of patients in each arm." }
+  },
+  "goldVerdicts": [
+    { "claimText": "In patients with PD-L1-high tumours, nivolumab improved objective response rate versus chemotherapy", "citation": 1, "verdict": "strong_support", "rationale": "Ref states ORR 48% vs 27% in the PD-L1-high group explicitly." },
+    { "claimText": "Nivolumab improves response across all patients regardless of PD-L1 status", "citation": 1, "verdict": "overclaim", "rationale": "Ref supports benefit only in PD-L1-high and is silent on other patients; claim broadens scope to all patients (rule 3)." },
+    { "claimText": "Treatment-related deaths were rare", "citation": 2, "verdict": "strong_support", "rationale": "Ref states treatment-related deaths <1% in each arm — supports 'rare'." }
+  ]
+}

package/datasets/cases/case-oncology-pembro.json

@@ -0,0 +1,31 @@
+{
+  "id": "oncology-pembro",
+  "title": "Immunotherapy + chemo — PFS vs OS",
+  "notes": "Synthetic-but-realistic. Ground truth is controlled by pairing each claim with an authored reference snippet. VERIFY/expand before publishing as a benchmark.",
+  "manuscript": "In the PRISM-3 trial, pembrolizumab plus chemotherapy improved progression-free survival compared with chemotherapy alone (median 9.2 vs 5.1 months; HR 0.52, 95% CI 0.41-0.65).1 Overall survival was significantly prolonged with the combination.1 The aim of this analysis was to describe long-term outcomes.",
+  "goldClaims": [
+    {
+      "originalText": "pembrolizumab plus chemotherapy improved progression-free survival compared with chemotherapy alone (median 9.2 vs 5.1 months; HR 0.52, 95% CI 0.41-0.65).1",
+      "text": "pembrolizumab plus chemotherapy improved progression-free survival compared with chemotherapy alone (median 9.2 vs 5.1 months; HR 0.52, 95% CI 0.41-0.65)",
+      "citations": [1]
+    },
+    {
+      "originalText": "Overall survival was significantly prolonged with the combination.1",
+      "text": "Overall survival was significantly prolonged with the combination",
+      "citations": [1]
+    }
+  ],
+  "goldNonClaims": [
+    "The aim of this analysis was to describe long-term outcomes."
+  ],
+  "references": {
+    "1": {
+      "name": "PRISM-3 primary (Doe 2024)",
+      "text": "In the PRISM-3 study, pembrolizumab plus chemotherapy significantly improved progression-free survival versus chemotherapy alone, with a median of 9.2 months compared with 5.1 months (hazard ratio 0.52; 95% CI 0.41-0.65; p<0.001). Overall survival data were immature at the time of this analysis and the difference between arms did not reach statistical significance."
+    }
+  },
+  "goldVerdicts": [
+    { "claimText": "pembrolizumab plus chemotherapy improved progression-free survival compared with chemotherapy alone (median 9.2 vs 5.1 months; HR 0.52, 95% CI 0.41-0.65)", "citation": 1, "verdict": "strong_support", "rationale": "Ref states the PFS result with matching figures and p<0.001 explicitly." },
+    { "claimText": "Overall survival was significantly prolonged with the combination", "citation": 1, "verdict": "contradicted", "rationale": "Ref says OS was immature and did not reach significance — negates 'significantly prolonged' (rule 1)." }
+  ]
+}

package/datasets/cases/case-osteo-fracture.json

@@ -0,0 +1,19 @@
+{
+  "id": "osteo-fracture",
+  "title": "Osteoporosis — overclaim and implied",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "The drug reduced fracture risk across all skeletal sites.1 The drug increased bone mineral density at the spine and hip.2",
+  "goldClaims": [
+    { "originalText": "The drug reduced fracture risk across all skeletal sites.1", "text": "The drug reduced fracture risk across all skeletal sites", "citations": [1] },
+    { "originalText": "The drug increased bone mineral density at the spine and hip.2", "text": "The drug increased bone mineral density at the spine and hip", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "BONE-1 (Eriksson 2023)", "text": "Vertebral fractures were significantly reduced with the drug versus placebo (p<0.001). Hip fractures were numerically fewer but the reduction did not reach statistical significance." },
+    "2": { "name": "BONE-1 densitometry", "text": "Table 5. Bone mineral density change from baseline (%) — spine: drug +6.8, placebo +0.4; hip: drug +3.9, placebo +0.2." }
+  },
+  "goldVerdicts": [
+    { "claimText": "The drug reduced fracture risk across all skeletal sites", "citation": 1, "verdict": "overclaim", "rationale": "Ref establishes a vertebral-fracture reduction only; hip reduction was non-significant. Claim generalises to all sites beyond what is established, without explicit denial (rule 1/3)." },
+    { "claimText": "The drug increased bone mineral density at the spine and hip", "citation": 2, "verdict": "implied_by_data", "rationale": "Support is a BMD data table for spine and hip with no narrative sentence (rule 5)." }
+  ]
+}

package/datasets/cases/case-pain-oa.json

@@ -0,0 +1,21 @@
+{
+  "id": "pain-oa",
+  "title": "Osteoarthritis — implied, overclaim, not-supported",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "At week 12, the drug reduced knee pain scores more than placebo.1 The drug significantly improved physical function.1 The drug slowed structural joint damage.2",
+  "goldClaims": [
+    { "originalText": "At week 12, the drug reduced knee pain scores more than placebo.1", "text": "At week 12, the drug reduced knee pain scores more than placebo", "citations": [1] },
+    { "originalText": "The drug significantly improved physical function.1", "text": "The drug significantly improved physical function", "citations": [1] },
+    { "originalText": "The drug slowed structural joint damage.2", "text": "The drug slowed structural joint damage", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "OA-RELIEF table (Marsh 2024)", "text": "Table 1. WOMAC pain change at week 12: drug -3.4, placebo -1.6. WOMAC physical-function score change: drug -2.9, placebo -1.4." },
+    "2": { "name": "OA-RELIEF imaging", "text": "Structural joint outcomes such as joint-space narrowing were not assessed in this trial." }
+  },
+  "goldVerdicts": [
+    { "claimText": "At week 12, the drug reduced knee pain scores more than placebo", "citation": 1, "verdict": "implied_by_data", "rationale": "Support is a data table (WOMAC pain -3.4 vs -1.6) with no narrative sentence (rule 5)." },
+    { "claimText": "The drug significantly improved physical function", "citation": 1, "verdict": "overclaim", "rationale": "Table shows a function change (-2.9 vs -1.4) but states no significance; claim asserts 'significantly' (rule 1)." },
+    { "claimText": "The drug slowed structural joint damage", "citation": 2, "verdict": "not_supported", "rationale": "Ref says structural outcomes were not assessed — silent (rule 2)." }
+  ]
+}

package/datasets/cases/case-psych-depression.json

@@ -0,0 +1,19 @@
+{
+  "id": "psych-depression",
+  "title": "Depression — partial and not-supported",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "At week 6, the antidepressant reduced depression rating scores and improved cognitive function versus placebo [1]. The antidepressant was effective in treatment-resistant depression [2].",
+  "goldClaims": [
+    { "originalText": "At week 6, the antidepressant reduced depression rating scores and improved cognitive function versus placebo [1].", "text": "At week 6, the antidepressant reduced depression rating scores and improved cognitive function versus placebo", "citations": [1] },
+    { "originalText": "The antidepressant was effective in treatment-resistant depression [2].", "text": "The antidepressant was effective in treatment-resistant depression", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "MOOD-1 (Pereira 2023)", "text": "The antidepressant significantly reduced MADRS depression scores versus placebo at week 6 (p<0.001). Cognitive function was not assessed in this trial." },
+    "2": { "name": "MOOD-1 methods", "text": "Patients with treatment-resistant depression were not enrolled; eligibility required a first or second major depressive episode." }
+  },
+  "goldVerdicts": [
+    { "claimText": "At week 6, the antidepressant reduced depression rating scores and improved cognitive function versus placebo", "citation": 1, "verdict": "partial_support", "rationale": "Depression-score reduction is explicit; cognitive function was not assessed — compound claim, one half holds (rule 4)." },
+    { "claimText": "The antidepressant was effective in treatment-resistant depression", "citation": 2, "verdict": "not_supported", "rationale": "Ref says treatment-resistant patients were not enrolled — silent on this population (rule 2/3)." }
+  ]
+}

package/datasets/cases/case-pulm-ipf.json

@@ -0,0 +1,18 @@
+{
+  "id": "pulm-ipf",
+  "title": "Pulmonary fibrosis — partial and overclaim",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "The antifibrotic slowed the decline in forced vital capacity and reduced chronic cough.1 The antifibrotic reduced mortality.1",
+  "goldClaims": [
+    { "originalText": "The antifibrotic slowed the decline in forced vital capacity and reduced chronic cough.1", "text": "The antifibrotic slowed the decline in forced vital capacity and reduced chronic cough", "citations": [1] },
+    { "originalText": "The antifibrotic reduced mortality.1", "text": "The antifibrotic reduced mortality", "citations": [1] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "IPF-SLOW (Romano 2024)", "text": "The antifibrotic significantly slowed the decline in forced vital capacity versus placebo (p<0.001). Cough was not assessed. Mortality was numerically lower with the antifibrotic, but the trial was not powered for mortality and the difference was not significant." }
+  },
+  "goldVerdicts": [
+    { "claimText": "The antifibrotic slowed the decline in forced vital capacity and reduced chronic cough", "citation": 1, "verdict": "partial_support", "rationale": "FVC slowing is explicit; cough was not assessed — compound claim, one half holds (rule 4)." },
+    { "claimText": "The antifibrotic reduced mortality", "citation": 1, "verdict": "overclaim", "rationale": "Ref reports a numerically lower, underpowered, non-significant mortality difference; claim asserts a definite reduction that is not established and not denied (rule 1)." }
+  ]
+}

package/datasets/cases/case-resp-copd.json

@@ -0,0 +1,35 @@
+{
+  "id": "resp-copd",
+  "title": "COPD triple therapy — strong and partial support",
+  "notes": "Synthetic-but-realistic; controlled ground truth. Uses superscript-range citation style in the manuscript. VERIFY before publishing.",
+  "manuscript": "Triple therapy reduced the annual rate of moderate-to-severe exacerbations compared with dual bronchodilator therapy.1 Triple therapy improved lung function and reduced all-cause mortality.1,2 Background therapy was standardised across groups.",
+  "goldClaims": [
+    {
+      "originalText": "Triple therapy reduced the annual rate of moderate-to-severe exacerbations compared with dual bronchodilator therapy.1",
+      "text": "Triple therapy reduced the annual rate of moderate-to-severe exacerbations compared with dual bronchodilator therapy",
+      "citations": [1]
+    },
+    {
+      "originalText": "Triple therapy improved lung function and reduced all-cause mortality.1,2",
+      "text": "Triple therapy improved lung function and reduced all-cause mortality",
+      "citations": [1, 2]
+    }
+  ],
+  "goldNonClaims": [
+    "Background therapy was standardised across groups."
+  ],
+  "references": {
+    "1": {
+      "name": "IMPACT-style primary (Adeyemi 2023)",
+      "text": "Triple therapy significantly reduced the annual rate of moderate-to-severe exacerbations compared with dual bronchodilator therapy (rate ratio 0.75; 95% CI 0.70-0.81; p<0.001). Trough FEV1 was significantly improved with triple therapy versus dual therapy (least-squares mean difference 97 mL; p<0.001)."
+    },
+    "2": {
+      "name": "Mortality sub-analysis",
+      "text": "All-cause mortality was numerically lower with triple therapy, but the analysis was not powered for mortality and the difference did not reach statistical significance."
+    }
+  },
+  "goldVerdicts": [
+    { "claimText": "Triple therapy reduced the annual rate of moderate-to-severe exacerbations compared with dual bronchodilator therapy", "citation": 1, "verdict": "strong_support", "rationale": "Ref states the exacerbation rate reduction explicitly with rate ratio and p<0.001." },
+    { "claimText": "Triple therapy improved lung function and reduced all-cause mortality", "citation": 1, "verdict": "partial_support", "rationale": "Lung-function improvement is explicit; mortality reduction is not established in ref 1 — compound claim, one half holds (rule 4)." }
+  ]
+}

package/datasets/cases/case-rheum-jak.json

@@ -0,0 +1,19 @@
+{
+  "id": "rheum-jak",
+  "title": "JAK inhibitor — implied-by-data and partial support",
+  "notes": "Synthetic; reference 1 is a data table with no narrative sentence (implied_by_data). VERIFY before publishing. (Avoids named endpoints like 'ACR20' that the citation normalizer mis-reads as a citation marker — see findings note.)",
+  "manuscript": "At week 24, the JAK inhibitor produced a higher clinical response rate than placebo.1 The JAK inhibitor improved physical function and reduced fatigue.2",
+  "goldClaims": [
+    { "originalText": "At week 24, the JAK inhibitor produced a higher clinical response rate than placebo.1", "text": "At week 24, the JAK inhibitor produced a higher clinical response rate than placebo", "citations": [1] },
+    { "originalText": "The JAK inhibitor improved physical function and reduced fatigue.2", "text": "The JAK inhibitor improved physical function and reduced fatigue", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "RA-JAK table (Singh 2023)", "text": "In this 24-week trial, 350 patients with rheumatoid arthritis were randomised to the JAK inhibitor or placebo. Response outcomes are summarised in Table 3. At week 24, a clinical response was recorded in 62% of patients in the JAK-inhibitor group and 25% in the placebo group. Background therapy was unchanged during the study." },
+    "2": { "name": "RA-JAK secondary endpoints", "text": "Patient-reported outcomes were pre-specified secondary endpoints. Physical function, measured by HAQ-DI, improved significantly with the JAK inhibitor versus placebo (p<0.001). Fatigue scores (FACIT-Fatigue) were numerically better with the JAK inhibitor, but the difference did not reach statistical significance." }
+  },
+  "goldVerdicts": [
+    { "claimText": "At week 24, the JAK inhibitor produced a higher clinical response rate than placebo", "citation": 1, "verdict": "implied_by_data", "rationale": "Support is a data table (62% vs 25%) with no narrative sentence stating the comparison (rule 5)." },
+    { "claimText": "The JAK inhibitor improved physical function and reduced fatigue", "citation": 2, "verdict": "partial_support", "rationale": "Physical function improvement is explicit and significant; fatigue is non-significant — compound claim, one half holds (rule 4)." }
+  ]
+}

package/datasets/cases/case-vaccine.json

@@ -0,0 +1,19 @@
+{
+  "id": "vaccine-efficacy",
+  "title": "Vaccine — strong support and contradiction",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "The vaccine reduced symptomatic infection by 71% versus placebo.1 The vaccine also prevented severe disease in immunocompromised adults.2",
+  "goldClaims": [
+    { "originalText": "The vaccine reduced symptomatic infection by 71% versus placebo.1", "text": "The vaccine reduced symptomatic infection by 71% versus placebo", "citations": [1] },
+    { "originalText": "The vaccine also prevented severe disease in immunocompromised adults.2", "text": "The vaccine also prevented severe disease in immunocompromised adults", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "TRIAL-V primary (Osei 2024)", "text": "Vaccine efficacy against symptomatic infection was 71% (95% CI 62-78) compared with placebo." },
+    "2": { "name": "TRIAL-V immunocompromised substudy", "text": "In immunocompromised adults, vaccine efficacy against severe disease was not demonstrated; rates of severe disease were similar between vaccine and placebo recipients." }
+  },
+  "goldVerdicts": [
+    { "claimText": "The vaccine reduced symptomatic infection by 71% versus placebo", "citation": 1, "verdict": "strong_support", "rationale": "Ref states 71% efficacy (95% CI 62-78) vs placebo explicitly." },
+    { "claimText": "The vaccine also prevented severe disease in immunocompromised adults", "citation": 2, "verdict": "contradicted", "rationale": "Ref says efficacy against severe disease was not demonstrated and rates were similar — negates 'prevented' (rule 1/2)." }
+  ]
+}

package/datasets/cases/case-womens-pcos.json

@@ -0,0 +1,21 @@
+{
+  "id": "womens-pcos",
+  "title": "PCOS — implied, contradicted, not-supported",
+  "notes": "Synthetic; controlled ground truth. VERIFY before publishing.",
+  "manuscript": "More women ovulated with the treatment than with placebo.1 The treatment improved live birth rates.1 The treatment improved long-term metabolic outcomes.2",
+  "goldClaims": [
+    { "originalText": "More women ovulated with the treatment than with placebo.1", "text": "More women ovulated with the treatment than with placebo", "citations": [1] },
+    { "originalText": "The treatment improved live birth rates.1", "text": "The treatment improved live birth rates", "citations": [1] },
+    { "originalText": "The treatment improved long-term metabolic outcomes.2", "text": "The treatment improved long-term metabolic outcomes", "citations": [2] }
+  ],
+  "goldNonClaims": [],
+  "references": {
+    "1": { "name": "PCOS-OV (Ali 2023)", "text": "Table 3. Ovulation rate: treatment 62%, placebo 28%. Live birth rates were similar between the treatment and placebo groups (24% vs 22%; not significant)." },
+    "2": { "name": "PCOS-OV follow-up", "text": "Long-term metabolic outcomes such as incident diabetes were not evaluated during the study." }
+  },
+  "goldVerdicts": [
+    { "claimText": "More women ovulated with the treatment than with placebo", "citation": 1, "verdict": "implied_by_data", "rationale": "Support is a data table (ovulation 62% vs 28%) with no narrative sentence (rule 5)." },
+    { "claimText": "The treatment improved live birth rates", "citation": 1, "verdict": "contradicted", "rationale": "Ref says live birth rates were similar (24% vs 22%, not significant) — negates an improvement (rule 1)." },
+    { "claimText": "The treatment improved long-term metabolic outcomes", "citation": 2, "verdict": "not_supported", "rationale": "Ref says metabolic outcomes were not evaluated — silent (rule 2)." }
+  ]
+}

package/datasets/cases/seed-teprotumumab.json

@@ -0,0 +1,37 @@
+{
+  "id": "seed-teprotumumab",
+  "title": "Teprotumumab / thyroid eye disease — manuscript intro",
+  "notes": "Hand-labelled seed case. Demonstrates the schema and exercises claim extraction, citation detection, and (online) verdict accuracy.",
+  "manuscript": "Thyroid eye disease is a debilitating autoimmune condition. In the pivotal trials, teprotumumab reduced proptosis by 2.82 mm versus placebo.1,2 Diplopia improved in 53% of patients receiving teprotumumab compared with 25% receiving placebo (p<0.001).3 These benefits were consistent across the OPTIC and OPTIC-X studies.1,4 The aim of this review was to summarise long-term outcomes.",
+
+  "goldClaims": [
+    {
+      "originalText": "teprotumumab reduced proptosis by 2.82 mm versus placebo.1,2",
+      "text": "teprotumumab reduced proptosis by 2.82 mm versus placebo",
+      "citations": [1, 2]
+    },
+    {
+      "originalText": "Diplopia improved in 53% of patients receiving teprotumumab compared with 25% receiving placebo (p<0.001).3",
+      "text": "Diplopia improved in 53% of patients receiving teprotumumab compared with 25% receiving placebo (p<0.001)",
+      "citations": [3]
+    },
+    {
+      "originalText": "These benefits were consistent across the OPTIC and OPTIC-X studies.1,4",
+      "text": "These benefits were consistent across the OPTIC and OPTIC-X studies",
+      "citations": [1, 4]
+    }
+  ],
+
+  "goldNonClaims": [
+    "Thyroid eye disease is a debilitating autoimmune condition.",
+    "The aim of this review was to summarise long-term outcomes."
+  ],
+
+  "references": {
+    "1": { "name": "OPTIC trial (Douglas 2020)", "text": "In the OPTIC trial, teprotumumab reduced proptosis by a mean of 2.82 mm compared with placebo (p<0.001)." }
+  },
+
+  "goldVerdicts": [
+    { "claimText": "teprotumumab reduced proptosis by 2.82 mm versus placebo", "citation": 1, "verdict": "strong_support", "rationale": "Ref states the 2.82 mm proptosis reduction vs placebo explicitly with p<0.001.", "_requires": "online" }
+  ]
+}

package/datasets/fixtures/citation-styles.json

@@ -0,0 +1,22 @@
+{
+  "description": "Citation-style coverage fixture. Each item is a realistic in-text citation as it appears after copy-paste from a manuscript. `expected` is the citation set RefCheckr SHOULD detect. `knownGap: true` marks styles we do NOT yet support — the harness reports these separately as a target, not a failure.",
+  "items": [
+    { "style": "vancouver-superscript-comma", "text": "teprotumumab reduced proptosis by 2.82 mm versus placebo.1,2", "expected": [1, 2], "knownGap": false },
+    { "style": "vancouver-superscript-range", "text": "consistent across trials.4-6", "expected": [4, 5, 6], "knownGap": false },
+    { "style": "bracketed", "text": "the ETD was -9.42 (95% CI -10.30 to -8.53) [3].", "expected": [3], "knownGap": false },
+    { "style": "parenthetical-numeric", "text": "response rates were higher with active treatment (1,2).", "expected": [1, 2], "knownGap": false },
+    { "style": "decimal-not-a-citation", "text": "the p-value was 0.17 and the mean was 4.13 overall.", "expected": [], "knownGap": false },
+    { "style": "author-year-parenthetical", "text": "Smith et al. (2020) reported a hazard ratio of 0.80.", "expected": ["Smith 2020"], "knownGap": false, "comment": "Was a known gap; implemented via detectAuthorYear()." },
+    { "style": "author-year-narrative", "text": "As shown by Jones and colleagues in 2019, survival improved.", "expected": ["Jones 2019"], "knownGap": false, "comment": "Was a known gap; implemented via detectAuthorYear()." },
+    { "style": "author-year-in-parens", "text": "proptosis improved significantly (Smith et al., 2020).", "expected": ["Smith 2020"], "knownGap": false },
+    { "style": "author-year-ampersand", "text": "results were durable (Smith & Jones, 2021).", "expected": ["Smith 2021"], "knownGap": false, "comment": "Keyed by first author only." },
+    { "style": "author-year-single-comma", "text": "consistent with earlier findings (Kim, 2023).", "expected": ["Kim 2023"], "knownGap": false },
+    { "style": "author-year-multiple", "text": "efficacy was confirmed in two cohorts (Brown 2018; Lee et al., 2022).", "expected": ["Brown 2018", "Lee 2022"], "knownGap": false },
+    { "style": "year-in-prose-not-a-citation", "text": "enrolment began in 2019 and completed in 2020.", "expected": [], "knownGap": false, "comment": "Plain prose years must not be read as citations; author-year detection requires a name or et-al marker." },
+    { "style": "parenthetical-year-only-not-a-citation", "text": "the trial protocol was amended (2020) before enrolment.", "expected": [], "knownGap": false, "comment": "A bare parenthetical year with no author name is not a citation." },
+    { "style": "number-adjacent-superscript", "text": "the reduction was sustained to week 24.1", "expected": [], "targetExpected": [1], "knownGap": true, "comment": "A superscript citation immediately after a number (24.1) is suppressed by the decimal guard that correctly ignores values like 0.17. Genuine ambiguity; tracked as a target." },
+    { "style": "endpoint-number-not-a-citation", "text": "the ACR20 response rate was higher with active treatment", "expected": [], "knownGap": false, "comment": "Named endpoints with trailing numbers (ACR20, PASI90, EASI75) must not be read as citations." },
+    { "style": "identifier-number-not-a-citation", "text": "patients with grade3 events and a CD4 count", "expected": [], "knownGap": false, "comment": "Letter-glued single numbers (grade3, CD4, type2) are identifiers, not citations." },
+    { "style": "letter-glued-list-is-a-citation", "text": "these outcomes were consistent1,2 across trials", "expected": [1, 2], "knownGap": false, "comment": "A letter-glued list/range is still a strong citation signal and must convert." }
+  ]
+}

package/opengate.http.example.json

@@ -0,0 +1,12 @@
+{
+  "name": "my-system",
+  "baseUrl": "${MY_SYSTEM_URL}",
+  "headers": {
+    "Authorization": "Bearer ${MY_SYSTEM_TOKEN}"
+  },
+  "endpoints": {
+    "splitClaims": "/api/claims/split",
+    "analyzeBatch": "/api/verify/batch"
+  },
+  "modelEnv": "MY_SYSTEM_MODEL"
+}

package/package.json

@@ -0,0 +1,55 @@
+{
+  "name": "@pharmatools/opengate",
+  "version": "0.1.0",
+  "type": "module",
+  "description": "OpenGATE — Open Grounded AI Testing & Evaluation. An open-source framework for evaluating evidence-grounded AI systems.",
+  "license": "MIT",
+  "author": "Nick Lamb (https://www.pharmatools.ai)",
+  "homepage": "https://www.pharmatools.ai/opengate",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/nickjlamb/opengate.git"
+  },
+  "bugs": {
+    "url": "https://github.com/nickjlamb/opengate/issues"
+  },
+  "engines": {
+    "node": ">=18"
+  },
+  "bin": {
+    "opengate": "./src/runner.mjs"
+  },
+  "main": "./src/index.mjs",
+  "exports": {
+    ".": "./src/index.mjs",
+    "./metrics": "./src/lib/metrics.mjs",
+    "./citations": "./src/lib/citations.mjs",
+    "./adapter": "./src/lib/adapter.mjs"
+  },
+  "files": [
+    "src/",
+    "datasets/",
+    "ADAPTERS.md",
+    "opengate.http.example.json"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "keywords": [
+    "evaluation",
+    "llm",
+    "rag",
+    "grounding",
+    "hallucination",
+    "benchmarking",
+    "regression-testing"
+  ],
+  "scripts": {
+    "test": "node --test tests/*.test.mjs",
+    "prepublishOnly": "npm test && npm run eval:ci",
+    "eval": "node src/runner.mjs",
+    "eval:online": "node src/runner.mjs --online",
+    "eval:ci": "node src/runner.mjs --ci",
+    "eval:baseline": "node src/runner.mjs --baseline"
+  }
+}