{"steps":[{"step":"library","label":"Claims Library","cases":[{"name":"4 sources → 10 atomic claims, each span-bound","input":{"sources":["s_casestudy","s_datasheet","s_pricing","s_security"],"library_version":"lib_f4ca2d098c44"},"graph_log":[{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_tco ← s_casestudy","input":{"claim_id":"c_tco"},"output":{"span":[27,78],"evidence":"Northwind Health cut total cost of ownership by 47%"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_los ← s_casestudy","input":{"claim_id":"c_los"},"output":{"span":[91,133],"evidence":"reduced average length of stay by 1.2 days"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_deployed ← s_casestudy","input":{"claim_id":"c_deployed"},"output":{"span":[158,193],"evidence":"deployed across 12 hospital systems"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_speed ← s_datasheet","input":{"claim_id":"c_speed"},"output":{"span":[13,61],"evidence":"processes a full claims dataset in under 4 hours"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_ehr ← s_datasheet","input":{"claim_id":"c_ehr"},"output":{"span":[66,96],"evidence":"integrates with 30 EHR systems"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_soc2 ← s_datasheet","input":{"claim_id":"c_soc2"},"output":{"span":[132,155],"evidence":"SOC 2 Type II certified"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_price ← s_pricing","input":{"claim_id":"c_price"},"output":{"span":[19,53],"evidence":"priced at $0.12 per patient record"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_nofee ← s_pricing","input":{"claim_id":"c_nofee"},"output":{"span":[64,112],"evidence":"no implementation fee for systems under 200 beds"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_hipaa ← s_security","input":{"claim_id":"c_hipaa"},"output":{"span":[19,34],"evidence":"HIPAA compliant"}},{"node":"bind","tool":"claim→evidence graph","event":"OUTPUT","detail":"c_encrypt ← s_security","input":{"claim_id":"c_encrypt"},"output":{"span":[48,80],"evidence":"encrypted at rest and in transit"}}],"output":{"claims":10,"all_span_bound":true},"expected":{"claims":10,"all_span_bound":true},"pass":true}]},{"step":"gate","label":"The Gate — claim verification","cases":[{"name":"Supported security claim → GREEN","input":{"claim_id":"c_soc2","text":"is SOC 2 Type II certified","legal_hold":false},"graph_log":[{"node":"gate","tool":"","event":"INPUT","detail":"verify c_soc2","input":{"claim_id":"c_soc2","text":"is SOC 2 Type II certified","source_id":"s_datasheet","source_version":"v_aa9da0278314","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_datasheet (score 9.80)","input":{"claim":"is SOC 2 Type II certified"},"decision":null,"output":{"evidence":"Helix Analytics is SOC 2 Type II certified","source_id":"s_datasheet","score":9.798}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"no rule constrains","input":{"rule_tags":["security"]},"decision":{"verdict":null,"flags":[]},"output":{"reasons":[]}},{"node":"nli","tool":"LexicalNLI","event":"DECISION","detail":"coverage=1.00","input":{"claim":"is SOC 2 Type II certified","evidence":"Helix Analytics is SOC 2 Type II certified"},"decision":{"label":"entailment","entail_prob":0.98},"output":null},{"node":"ensemble","tool":"diverse judge ensemble","event":"DECISION","detail":"mean agreement 0.95 (numeric & superlative are veto lenses)","input":{"escalated":false},"decision":{"votes":{"coverage":1.0,"numeric":0.95,"superlative":0.9},"mean_p":0.95},"output":null},{"node":"calibrate","tool":"isotonic PAV calibrator","event":"DECISION","detail":"entail_raw 0.900 → confidence 0.900 (green)","input":{"entail_raw":0.9},"decision":{"entail_verdict":"green"},"output":{"confidence":0.9}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"c_soc2 → GREEN (most-severe of entail+rule)","input":null,"decision":{"verdict":"green","entail_verdict":"green","rule_verdict":null},"output":{"confidence":0.9,"flags":[],"reasons":["coverage=1.00"]}}],"output":{"verdict":"green","confidence":0.9,"flags":[]},"expected":{"verdict":"green"},"pass":true},{"name":"Clinical-outcome claim needs a disclaimer → AMBER","input":{"claim_id":"c_los","text":"reduces average length of stay by 1.2 days","legal_hold":false},"graph_log":[{"node":"gate","tool":"","event":"INPUT","detail":"verify c_los","input":{"claim_id":"c_los","text":"reduces average length of stay by 1.2 days","source_id":"s_casestudy","source_version":"v_598c5748478c","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_casestudy (score 11.73)","input":{"claim":"reduces average length of stay by 1.2 days"},"decision":null,"output":{"evidence":"The system reduced average length of stay by 1.2 days","source_id":"s_casestudy","score":11.731}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"amber","input":{"rule_tags":["clinical_outcome"]},"decision":{"verdict":"amber","flags":["clinical_outcome_disclaimer"]},"output":{"reasons":["Clinical-outcome claims require a disclaimer"]}},{"node":"nli","tool":"LexicalNLI","event":"DECISION","detail":"coverage=0.80","input":{"claim":"reduces average length of stay by 1.2 days","evidence":"The system reduced average length of stay by 1.2 days"},"decision":{"label":"entailment","entail_prob":0.8},"output":null},{"node":"ensemble","tool":"","event":"ESCALATE","detail":"NLI uncertain (0.80) → spend LLM judges","input":null,"decision":null,"output":null},{"node":"ensemble","tool":"diverse judge ensemble","event":"DECISION","detail":"mean agreement 0.8833 (numeric & superlative are veto lenses)","input":{"escalated":true},"decision":{"votes":{"coverage":0.8,"numeric":0.95,"superlative":0.9},"mean_p":0.8833},"output":null},{"node":"calibrate","tool":"isotonic PAV calibrator","event":"DECISION","detail":"entail_raw 0.800 → confidence 0.800 (green)","input":{"entail_raw":0.8},"decision":{"entail_verdict":"green"},"output":{"confidence":0.8}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"c_los → AMBER (most-severe of entail+rule)","input":null,"decision":{"verdict":"amber","entail_verdict":"green","rule_verdict":"amber"},"output":{"confidence":0.8,"flags":["clinical_outcome_disclaimer"],"reasons":["coverage=0.80","Clinical-outcome claims require a disclaimer"]}}],"output":{"verdict":"amber","confidence":0.8,"flags":["clinical_outcome_disclaimer"]},"expected":{"verdict":"amber"},"pass":true},{"name":"ROI claim needs a disclaimer → AMBER","input":{"claim_id":"c_tco","text":"cuts total cost of ownership by 47%","legal_hold":false},"graph_log":[{"node":"gate","tool":"","event":"INPUT","detail":"verify c_tco","input":{"claim_id":"c_tco","text":"cuts total cost of ownership by 47%","source_id":"s_casestudy","source_version":"v_598c5748478c","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_casestudy (score 7.81)","input":{"claim":"cuts total cost of ownership by 47%"},"decision":null,"output":{"evidence":"In an 18-month deployment, Northwind Health cut total cost of ownership by 47%","source_id":"s_casestudy","score":7.811}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"amber","input":{"rule_tags":["roi_outcome"]},"decision":{"verdict":"amber","flags":["roi_outcome_disclaimer"]},"output":{"reasons":["ROI/outcome claims require an 'individual results may vary' disclaimer"]}},{"node":"nli","tool":"LexicalNLI","event":"DECISION","detail":"coverage=0.75","input":{"claim":"cuts total cost of ownership by 47%","evidence":"In an 18-month deployment, Northwind Health cut total cost of ownership by 47%"},"decision":{"label":"entailment","entail_prob":0.8},"output":null},{"node":"ensemble","tool":"","event":"ESCALATE","detail":"NLI uncertain (0.80) → spend LLM judges","input":null,"decision":null,"output":null},{"node":"ensemble","tool":"diverse judge ensemble","event":"DECISION","detail":"mean agreement 0.8667 (numeric & superlative are veto lenses)","input":{"escalated":true},"decision":{"votes":{"coverage":0.75,"numeric":0.95,"superlative":0.9},"mean_p":0.8667},"output":null},{"node":"calibrate","tool":"isotonic PAV calibrator","event":"DECISION","detail":"entail_raw 0.775 → confidence 0.775 (green)","input":{"entail_raw":0.775},"decision":{"entail_verdict":"green"},"output":{"confidence":0.775}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"c_tco → AMBER (most-severe of entail+rule)","input":null,"decision":{"verdict":"amber","entail_verdict":"green","rule_verdict":"amber"},"output":{"confidence":0.775,"flags":["roi_outcome_disclaimer"],"reasons":["coverage=0.75","ROI/outcome claims require an 'individual results may vary' disclaimer"]}}],"output":{"verdict":"amber","confidence":0.775,"flags":["roi_outcome_disclaimer"]},"expected":{"verdict":"amber"},"pass":true},{"name":"ROI claim under the MLR legal hold → RED","input":{"claim_id":"c_tco","text":"cuts total cost of ownership by 47%","legal_hold":true},"graph_log":[{"node":"gate","tool":"","event":"INPUT","detail":"verify c_tco","input":{"claim_id":"c_tco","text":"cuts total cost of ownership by 47%","source_id":"s_casestudy","source_version":"v_598c5748478c","rules_version":"r_f7bf1787cb7d"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_casestudy (score 7.81)","input":{"claim":"cuts total cost of ownership by 47%"},"decision":null,"output":{"evidence":"In an 18-month deployment, Northwind Health cut total cost of ownership by 47%","source_id":"s_casestudy","score":7.811}},{"node":"rules","tool":"rules engine r_f7bf1787cb7d","event":"DECISION","detail":"red","input":{"rule_tags":["roi_outcome"]},"decision":{"verdict":"red","flags":["mlr_hold_tco","roi_outcome_disclaimer"]},"output":{"reasons":["MLR legal hold pending substantiation review of the 47% TCO claim","ROI/outcome claims require an 'individual results may vary' disclaimer"]}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"c_tco → RED (compliance veto, cascade short-circuit)","input":null,"decision":{"verdict":"red","via":"rule"},"output":{"confidence":0.02,"flags":["mlr_hold_tco","roi_outcome_disclaimer"]}}],"output":{"verdict":"red","confidence":0.02,"flags":["mlr_hold_tco","roi_outcome_disclaimer"]},"expected":{"verdict":"red"},"pass":true},{"name":"Number drifted 47%→57% → RED (numeric veto)","input":{"claim_id":"(free)","text":"Helix Analytics cuts total cost of ownership by 57%","legal_hold":false},"graph_log":[{"node":"gate","tool":"","event":"INPUT","detail":"verify free_21e16b9c","input":{"claim_id":"free_21e16b9c","text":"Helix Analytics cuts total cost of ownership by 57%","source_id":"none","source_version":"none","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_casestudy (score 6.27)","input":{"claim":"Helix Analytics cuts total cost of ownership by 57%"},"decision":null,"output":{"evidence":"In an 18-month deployment, Northwind Health cut total cost of ownership by 47%","source_id":"s_casestudy","score":6.272}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"no rule constrains","input":{"rule_tags":[]},"decision":{"verdict":null,"flags":[]},"output":{"reasons":[]}},{"node":"nli","tool":"LexicalNLI","event":"DECISION","detail":"claim asserts a number not entailed by the source","input":{"claim":"Helix Analytics cuts total cost of ownership by 57%","evidence":"In an 18-month deployment, Northwind Health cut total cost of ownership by 47%"},"decision":{"label":"contradiction","entail_prob":0.12},"output":null},{"node":"ensemble","tool":"diverse judge ensemble","event":"DECISION","detail":"mean agreement 0.4833 (numeric & superlative are veto lenses)","input":{"escalated":false},"decision":{"votes":{"coverage":0.5,"numeric":0.05,"superlative":0.9},"mean_p":0.4833},"output":null},{"node":"calibrate","tool":"isotonic PAV calibrator","event":"DECISION","detail":"entail_raw 0.050 → confidence 0.050 (red)","input":{"entail_raw":0.05},"decision":{"entail_verdict":"red"},"output":{"confidence":0.05}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"free_21e16b9c → RED (most-severe of entail+rule)","input":null,"decision":{"verdict":"red","entail_verdict":"red","rule_verdict":null},"output":{"confidence":0.05,"flags":[],"reasons":["not entailed by any approved source","claim asserts a number not entailed by the source"]}}],"output":{"verdict":"red","confidence":0.05,"flags":[]},"expected":{"verdict":"red"},"pass":true},{"name":"Unsupported superlative (#1) → RED","input":{"claim_id":"(free)","text":"Helix Analytics is the #1 best-in-class platform","legal_hold":false},"graph_log":[{"node":"gate","tool":"","event":"INPUT","detail":"verify free_9ce16780","input":{"claim_id":"free_9ce16780","text":"Helix Analytics is the #1 best-in-class platform","source_id":"none","source_version":"none","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_datasheet (score 3.18)","input":{"claim":"Helix Analytics is the #1 best-in-class platform"},"decision":null,"output":{"evidence":"The platform processes a full claims dataset in under 4 hours","source_id":"s_datasheet","score":3.178}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"red","input":{"rule_tags":[]},"decision":{"verdict":"red","flags":["unsubstantiated_superlative"]},"output":{"reasons":["Unsubstantiated superlative without a cited source"]}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"free_9ce16780 → RED (compliance veto, cascade short-circuit)","input":null,"decision":{"verdict":"red","via":"rule"},"output":{"confidence":0.02,"flags":["unsubstantiated_superlative"]}}],"output":{"verdict":"red","confidence":0.02,"flags":["unsubstantiated_superlative"]},"expected":{"verdict":"red"},"pass":true},{"name":"Guaranteed outcome (the planted lie) → RED","input":{"claim_id":"(free)","text":"Helix Analytics guarantees a 60% reduction in hospital readmissions","legal_hold":false},"graph_log":[{"node":"gate","tool":"","event":"INPUT","detail":"verify free_e4254dac","input":{"claim_id":"free_e4254dac","text":"Helix Analytics guarantees a 60% reduction in hospital readmissions","source_id":"none","source_version":"none","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_casestudy (score 2.67)","input":{"claim":"Helix Analytics guarantees a 60% reduction in hospital readmissions"},"decision":null,"output":{"evidence":"Helix Analytics is now deployed across 12 hospital systems","source_id":"s_casestudy","score":2.668}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"red","input":{"rule_tags":[]},"decision":{"verdict":"red","flags":["no_guaranteed_outcomes"]},"output":{"reasons":["FTC: no guaranteed financial or clinical outcomes"]}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"free_e4254dac → RED (compliance veto, cascade short-circuit)","input":null,"decision":{"verdict":"red","via":"rule"},"output":{"confidence":0.02,"flags":["no_guaranteed_outcomes"]}}],"output":{"verdict":"red","confidence":0.02,"flags":["no_guaranteed_outcomes"]},"expected":{"verdict":"red"},"pass":true}]},{"step":"enrich_gate","label":"The Enrichment Gate — fact verification","cases":[{"name":"Clean consented fact → USABLE","input":{"key":"recent_news","value":"opened a regional facility","source":"company_news_rss","age_seconds":0,"consent":true},"graph_log":[{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"recent_news (company_news_rss) → usable","input":{"key":"recent_news","source":"company_news_rss","age_seconds":0},"decision":{"verdict":"usable","basis":"public_record","reasons":[]},"output":null}],"output":{"verdict":"usable","basis":"public_record","reasons":[]},"expected":{"verdict":"usable"},"pass":true},{"name":"Disallowed source (scraped) → BLOCKED","input":{"key":"seniority","value":"VP","source":"scraped_linkedin","age_seconds":0,"consent":true},"graph_log":[{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"seniority (scraped_linkedin) → blocked: source 'scraped_linkedin' is not in the enrichment allow-list","input":{"key":"seniority","source":"scraped_linkedin","age_seconds":0},"decision":{"verdict":"blocked","basis":"","reasons":["source 'scraped_linkedin' is not in the enrichment allow-list"]},"output":null}],"output":{"verdict":"blocked","basis":"","reasons":["source 'scraped_linkedin' is not in the enrichment allow-list"]},"expected":{"verdict":"blocked"},"pass":true},{"name":"Stale fact past its TTL → BLOCKED","input":{"key":"recent_news","value":"old","source":"company_news_rss","age_seconds":10000000,"consent":true},"graph_log":[{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"recent_news (company_news_rss) → blocked: stale: age 10000000s > ttl 86400s","input":{"key":"recent_news","source":"company_news_rss","age_seconds":10000000},"decision":{"verdict":"blocked","basis":"public_record","reasons":["stale: age 10000000s > ttl 86400s"]},"output":null}],"output":{"verdict":"blocked","basis":"public_record","reasons":["stale: age 10000000s > ttl 86400s"]},"expected":{"verdict":"blocked"},"pass":true},{"name":"PHI key → BLOCKED","input":{"key":"diagnosis","value":"redacted","source":"firmographic_sim","age_seconds":0,"consent":true},"graph_log":[{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"diagnosis (firmographic_sim) → blocked: key 'diagnosis' is PHI/PII-blocked","input":{"key":"diagnosis","source":"firmographic_sim","age_seconds":0},"decision":{"verdict":"blocked","basis":"contract_vendor_dpa","reasons":["key 'diagnosis' is PHI/PII-blocked"]},"output":null}],"output":{"verdict":"blocked","basis":"contract_vendor_dpa","reasons":["key 'diagnosis' is PHI/PII-blocked"]},"expected":{"verdict":"blocked"},"pass":true},{"name":"Recipient never consented → BLOCKED","input":{"key":"intent_topic","value":"ROI","source":"intent_sim","age_seconds":0,"consent":false},"graph_log":[{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"intent_topic (intent_sim) → blocked: recipient did not consent","input":{"key":"intent_topic","source":"intent_sim","age_seconds":0},"decision":{"verdict":"blocked","basis":"contract_vendor_dpa","reasons":["recipient did not consent"]},"output":null}],"output":{"verdict":"blocked","basis":"contract_vendor_dpa","reasons":["recipient did not consent"]},"expected":{"verdict":"blocked"},"pass":true}]},{"step":"optimizer","label":"Optimizer — bandit over verified arms","cases":[{"name":"A Gate-blocked lie can never be selected (twin converges to it)","input":{"recipients":1000,"constrained":true},"graph_log":[{"node":"bandit","tool":"Thompson bandit + synthetic CTA oracle","event":"INPUT","detail":"campaign c1/email over 1000 recipients (verified arms only)","input":{"campaign":"c1","channel":"email","constrained":true,"n":1000,"warm_started":false,"active_arms":{"cfo__core":["cfo__core__email__A","cfo__core__email__B"],"cfo__ent":["cfo__ent__email__A","cfo__ent__email__B"],"clinops__core":["clinops__core__email__A","clinops__core__email__B"],"clinops__ent":["clinops__ent__email__A","clinops__ent__email__B"],"it_security__core":["it_security__core__email__A","it_security__core__email__B"],"it_security__ent":["it_security__ent__email__A","it_security__ent__email__B"],"quality__core":["quality__core__email__A","quality__core__email__B"],"quality__ent":["quality__ent__email__A","quality__ent__email__B"]}},"decision":null,"output":null},{"node":"bandit","tool":"","event":"DECISION","detail":"quality__ent: learned winner A","input":null,"decision":{"segment":"quality__ent","winner_arm":"A","posterior_mean":0.255,"est_ctr":0.244,"winner_is_lie":false},"output":null},{"node":"bandit","tool":"","event":"DECISION","detail":"clinops__ent: learned winner B","input":null,"decision":{"segment":"clinops__ent","winner_arm":"B","posterior_mean":0.25,"est_ctr":0.242,"winner_is_lie":false},"output":null},{"node":"bandit","tool":"","event":"DECISION","detail":"it_security__core: learned winner B","input":null,"decision":{"segment":"it_security__core","winner_arm":"B","posterior_mean":0.245,"est_ctr":0.242,"winner_is_lie":false},"output":null},{"node":"bandit","tool":"","event":"DECISION","detail":"quality__core: learned winner B","input":null,"decision":{"segment":"quality__core","winner_arm":"B","posterior_mean":0.18,"est_ctr":0.173,"winner_is_lie":false},"output":null},{"node":"bandit","tool":"","event":"DECISION","detail":"cfo__core: learned winner A","input":null,"decision":{"segment":"cfo__core","winner_arm":"A","posterior_mean":0.235,"est_ctr":0.235,"winner_is_lie":false},"output":null},{"node":"bandit","tool":"","event":"DECISION","detail":"cfo__ent: learned winner B","input":null,"decision":{"segment":"cfo__ent","winner_arm":"B","posterior_mean":0.175,"est_ctr":0.164,"winner_is_lie":false},"output":null},{"node":"bandit","tool":"","event":"DECISION","detail":"clinops__core: learned winner A","input":null,"decision":{"segment":"clinops__core","winner_arm":"A","posterior_mean":0.237,"est_ctr":0.235,"winner_is_lie":false},"output":null},{"node":"bandit","tool":"","event":"DECISION","detail":"it_security__ent: learned winner A","input":null,"decision":{"segment":"it_security__ent","winner_arm":"A","posterior_mean":0.232,"est_ctr":0.222,"winner_is_lie":false},"output":null},{"node":"bandit","tool":"","event":"OUTPUT","detail":"c1/email: lie selected 0× · final regret 19.68","input":null,"decision":null,"output":{"selections_of_lie":0,"final_regret":19.68,"winner_is_lie_anywhere":false}},{"node":"twin","tool":"Thompson bandit + synthetic CTA oracle","event":"INPUT","detail":"campaign twin/email over 1000 recipients (lie in pool)","input":{"campaign":"twin","channel":"email","constrained":false,"n":1000,"warm_started":false,"active_arms":{"cfo__core":["cfo__core__email__A","cfo__core__email__B","cfo__core__email__LIE"],"cfo__ent":["cfo__ent__email__A","cfo__ent__email__B","cfo__ent__email__LIE"],"clinops__core":["clinops__core__email__A","clinops__core__email__B","clinops__core__email__LIE"],"clinops__ent":["clinops__ent__email__A","clinops__ent__email__B","clinops__ent__email__LIE"],"it_security__core":["it_security__core__email__A","it_security__core__email__B","it_security__core__email__LIE"],"it_security__ent":["it_security__ent__email__A","it_security__ent__email__B","it_security__ent__email__LIE"],"quality__core":["quality__core__email__A","quality__core__email__B","quality__core__email__LIE"],"quality__ent":["quality__ent__email__A","quality__ent__email__B","quality__ent__email__LIE"]}},"decision":null,"output":null},{"node":"twin","tool":"","event":"DECISION","detail":"quality__ent: learned winner LIE — LIE","input":null,"decision":{"segment":"quality__ent","winner_arm":"LIE","posterior_mean":0.176,"est_ctr":0.133,"winner_is_lie":true},"output":null},{"node":"twin","tool":"","event":"DECISION","detail":"clinops__ent: learned winner LIE — LIE","input":null,"decision":{"segment":"clinops__ent","winner_arm":"LIE","posterior_mean":0.431,"est_ctr":0.429,"winner_is_lie":true},"output":null},{"node":"twin","tool":"","event":"DECISION","detail":"it_security__core: learned winner LIE — LIE","input":null,"decision":{"segment":"it_security__core","winner_arm":"LIE","posterior_mean":0.321,"est_ctr":0.318,"winner_is_lie":true},"output":null}],"output":{"constrained_lie_selections":0,"twin_lie_selections":734},"expected":{"constrained_lie_selections":0},"pass":true}]},{"step":"drift","label":"Drift Monitor — surgical re-verify","cases":[{"name":"Legal-hold flip re-verifies exactly the held claim → RED","input":{"event":"legal_hold","change":"mlr_hold_tco -> ON"},"graph_log":[{"node":"drift","tool":"claim→source dependency graph","event":"INPUT","detail":"legal_hold: mlr_hold_tco -> ON","input":{"event":"legal_hold","change":"mlr_hold_tco -> ON","affected_claims":["c_tco"]},"decision":null,"output":null},{"node":"drift","tool":"surgical re-Gate (cache-miss only)","event":"DRIFT","detail":"re-verified 1/1 affected claim(s) — c_tco amber→red","input":null,"decision":{"recomputed_count":1,"recomputed_ids":["c_tco"],"before":{"c_tco":"amber"},"after":{"c_tco":"red"}},"output":null},{"node":"drift","tool":"","event":"OUTPUT","detail":"paused 2 · unblocked 0","input":null,"decision":null,"output":{"paused_variants":["cfo__core__email__A","cfo__ent__email__A"],"unblocked_variants":[]}}],"output":{"before":{"c_tco":"amber"},"after":{"c_tco":"red"},"recomputed":["c_tco"]},"expected":{"after_red":true,"recomputed_equals_affected":true},"pass":true}]},{"step":"assurance","label":"Assurance Lab — Gate vs single judge","cases":[{"name":"Gate catch-rate > single-judge baseline at fixed false-reject","input":{"n_traps":37,"n_clean":10},"graph_log":[{"node":"assurance","tool":"adversarial trap generator (4 mutations) + the real Gate + single-judge baseline","event":"INPUT","detail":"47 traps through the Gate vs a number-blind single judge","input":{"n_traps":47,"by_mutation":{"clean":10,"number_drift":7,"unsupported_superlative":10,"false_equivalence":10,"true_but_unsayable":10}},"decision":null,"output":null},{"node":"assurance","tool":"","event":"DECISION","detail":"Gate catch 100% vs single-judge 24% at 0% false-reject","input":null,"decision":{"gate_catch":1.0,"baseline_catch":0.2432,"gate_false_reject":0.0,"gate_by_type":{"number_drift":1.0,"unsupported_superlative":1.0,"false_equivalence":1.0,"true_but_unsayable":1.0},"baseline_by_type":{"number_drift":0.0,"unsupported_superlative":0.4,"false_equivalence":0.4,"true_but_unsayable":0.1}},"output":{"ece":0.0,"n_traps":37,"n_clean":10}},{"node":"fact_audit","tool":"enrichment fact-trap harness","event":"DECISION","detail":"fact-gate caught 100% of un-shippable facts at 0% false-block","input":null,"decision":{"catch_rate":1.0,"false_block":0.0},"output":{"by_trap":{"clean":"usable","disallowed":"blocked","stale":"blocked","phi":"blocked","nonconsent":"blocked"}}}],"output":{"gate_catch":1.0,"baseline_catch":0.2432,"false_reject":0.0,"ece":0.0},"expected":{"gate_catch_gt_baseline":true},"pass":true}]},{"step":"website","label":"Website channel — same Gate, both channels","cases":[{"name":"Same claim_id → same verdict on email & website; 0 red rendered","input":{"segments":8},"graph_log":[{"node":"website","tool":"Thompson bandit + synthetic CTA oracle","event":"INPUT","detail":"campaign web/website over 1000 recipients (verified arms only)","input":{"campaign":"web","channel":"website","constrained":true,"n":1000,"warm_started":true,"active_arms":{"cfo__core":["cfo__core__website__B"],"cfo__ent":["cfo__ent__website__B"],"clinops__core":["clinops__core__website__A","clinops__core__website__B"],"clinops__ent":["clinops__ent__website__A","clinops__ent__website__B"],"it_security__core":["it_security__core__website__A","it_security__core__website__B"],"it_security__ent":["it_security__ent__website__A","it_security__ent__website__B"],"quality__core":["quality__core__website__A","quality__core__website__B"],"quality__ent":["quality__ent__website__A","quality__ent__website__B"]}},"decision":null,"output":null},{"node":"website","tool":"","event":"DECISION","detail":"quality__ent: learned winner A","input":null,"decision":{"segment":"quality__ent","winner_arm":"A","posterior_mean":0.38,"est_ctr":0.375,"winner_is_lie":false},"output":null},{"node":"website","tool":"","event":"DECISION","detail":"clinops__ent: learned winner B","input":null,"decision":{"segment":"clinops__ent","winner_arm":"B","posterior_mean":0.229,"est_ctr":0.227,"winner_is_lie":false},"output":null},{"node":"website","tool":"","event":"DECISION","detail":"it_security__core: learned winner B","input":null,"decision":{"segment":"it_security__core","winner_arm":"B","posterior_mean":0.211,"est_ctr":0.21,"winner_is_lie":false},"output":null},{"node":"website","tool":"","event":"DECISION","detail":"quality__core: learned winner B","input":null,"decision":{"segment":"quality__core","winner_arm":"B","posterior_mean":0.158,"est_ctr":0.151,"winner_is_lie":false},"output":null},{"node":"website","tool":"","event":"DECISION","detail":"cfo__core: learned winner B","input":null,"decision":{"segment":"cfo__core","winner_arm":"B","posterior_mean":0.048,"est_ctr":0.044,"winner_is_lie":false},"output":null},{"node":"website","tool":"","event":"DECISION","detail":"cfo__ent: learned winner B","input":null,"decision":{"segment":"cfo__ent","winner_arm":"B","posterior_mean":0.256,"est_ctr":0.25,"winner_is_lie":false},"output":null},{"node":"website","tool":"","event":"DECISION","detail":"clinops__core: learned winner A","input":null,"decision":{"segment":"clinops__core","winner_arm":"A","posterior_mean":0.251,"est_ctr":0.249,"winner_is_lie":false},"output":null},{"node":"website","tool":"","event":"DECISION","detail":"it_security__ent: learned winner A","input":null,"decision":{"segment":"it_security__ent","winner_arm":"A","posterior_mean":0.18,"est_ctr":0.174,"winner_is_lie":false},"output":null},{"node":"website","tool":"","event":"OUTPUT","detail":"web/website: lie selected 0× · final regret 17.12","input":null,"decision":null,"output":{"selections_of_lie":0,"final_regret":17.12,"winner_is_lie_anywhere":false}}],"output":{"verdict_mismatches":[],"red_claims_rendered":0},"expected":{"verdict_mismatches":[],"red_claims_rendered":0},"pass":true}]}],"summary":{"passed":17,"total":17},"lifecycle":{"name":"Maria Chen — form → enrich → Gate → personalize (end to end)","input":{"recipient":"Northwind Health","segment":"clinops__ent","consent":true},"graph_log":[{"node":"enrich_connectors","tool":"","event":"INPUT","detail":"enrich r_eval_maria (Northwind Health, consent=True)","input":{"recipient_id":"r_eval_maria","company":"Northwind Health","consent":true,"mode":"synthetic"},"decision":null,"output":null},{"node":"enrich_connectors","tool":"email domain parse (real, no network)","event":"TOOL","detail":"email_domain: 1 fact(s)","input":null,"decision":null,"output":[{"key":"company_domain","value":"northwindhealth.org"}]},{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"company_domain (email_domain) → usable","input":{"key":"company_domain","source":"email_domain","age_seconds":0},"decision":{"verdict":"usable","basis":"legitimate_interest_b2b","reasons":[]},"output":null},{"node":"enrich_connectors","tool":"public news (simulated, deterministic)","event":"TOOL","detail":"company_news_rss: 1 fact(s)","input":null,"decision":null,"output":[{"key":"recent_news","value":"Northwind Health reported a push to cut administrative cost"}]},{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"recent_news (company_news_rss) → usable","input":{"key":"recent_news","source":"company_news_rss","age_seconds":0},"decision":{"verdict":"usable","basis":"public_record","reasons":[]},"output":null},{"node":"enrich_connectors","tool":"firmographics (simulated, Clearbit/ZoomInfo class)","event":"TOOL","detail":"firmographic_sim: 3 fact(s)","input":null,"decision":null,"output":[{"key":"num_facilities","value":"19"},{"key":"ehr_vendor","value":"Allscripts/Veradigm"},{"key":"size_band","value":"9+ hospital IDN"}]},{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"num_facilities (firmographic_sim) → usable","input":{"key":"num_facilities","source":"firmographic_sim","age_seconds":0},"decision":{"verdict":"usable","basis":"contract_vendor_dpa","reasons":[]},"output":null},{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"ehr_vendor (firmographic_sim) → usable","input":{"key":"ehr_vendor","source":"firmographic_sim","age_seconds":0},"decision":{"verdict":"usable","basis":"contract_vendor_dpa","reasons":[]},"output":null},{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"size_band (firmographic_sim) → usable","input":{"key":"size_band","source":"firmographic_sim","age_seconds":0},"decision":{"verdict":"usable","basis":"contract_vendor_dpa","reasons":[]},"output":null},{"node":"enrich_connectors","tool":"account intent (simulated, 6sense/Bombora class)","event":"TOOL","detail":"intent_sim: 2 fact(s)","input":null,"decision":null,"output":[{"key":"intent_topic","value":"reducing length of stay"},{"key":"in_market","value":"true"}]},{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"intent_topic (intent_sim) → disclaimer: account-level inferred signal — flag as inferred","input":{"key":"intent_topic","source":"intent_sim","age_seconds":0},"decision":{"verdict":"disclaimer","basis":"contract_vendor_dpa","reasons":["account-level inferred signal — flag as inferred"]},"output":null},{"node":"enrich_gate","tool":"enrichment-source policy ep_fd66f64c63","event":"DECISION","detail":"in_market (intent_sim) → disclaimer: account-level inferred signal — flag as inferred","input":{"key":"in_market","source":"intent_sim","age_seconds":0},"decision":{"verdict":"disclaimer","basis":"contract_vendor_dpa","reasons":["account-level inferred signal — flag as inferred"]},"output":null},{"node":"enrich_synth","tool":"synthesizer","event":"OUTPUT","detail":"7 usable · 0 blocked · signals: intent_topic, lead_with, in_market, account_tier, ehr_vendor, recent_news","input":null,"decision":null,"output":{"usable":7,"blocked":0,"signals":{"intent_topic":"reducing length of stay","lead_with":"reducing length of stay","in_market":true,"account_tier":"9+ hospital IDN","ehr_vendor":"Allscripts/Veradigm","recent_news":"Northwind Health reported a push to cut administrative cost"}}},{"node":"gate","tool":"","event":"INPUT","detail":"verify c_los","input":{"claim_id":"c_los","text":"reduces average length of stay by 1.2 days","source_id":"s_casestudy","source_version":"v_598c5748478c","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_casestudy (score 11.73)","input":{"claim":"reduces average length of stay by 1.2 days"},"decision":null,"output":{"evidence":"The system reduced average length of stay by 1.2 days","source_id":"s_casestudy","score":11.731}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"amber","input":{"rule_tags":["clinical_outcome"]},"decision":{"verdict":"amber","flags":["clinical_outcome_disclaimer"]},"output":{"reasons":["Clinical-outcome claims require a disclaimer"]}},{"node":"nli","tool":"LexicalNLI","event":"DECISION","detail":"coverage=0.80","input":{"claim":"reduces average length of stay by 1.2 days","evidence":"The system reduced average length of stay by 1.2 days"},"decision":{"label":"entailment","entail_prob":0.8},"output":null},{"node":"ensemble","tool":"","event":"ESCALATE","detail":"NLI uncertain (0.80) → spend LLM judges","input":null,"decision":null,"output":null},{"node":"ensemble","tool":"diverse judge ensemble","event":"DECISION","detail":"mean agreement 0.8833 (numeric & superlative are veto lenses)","input":{"escalated":true},"decision":{"votes":{"coverage":0.8,"numeric":0.95,"superlative":0.9},"mean_p":0.8833},"output":null},{"node":"calibrate","tool":"isotonic PAV calibrator","event":"DECISION","detail":"entail_raw 0.800 → confidence 0.800 (green)","input":{"entail_raw":0.8},"decision":{"entail_verdict":"green"},"output":{"confidence":0.8}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"c_los → AMBER (most-severe of entail+rule)","input":null,"decision":{"verdict":"amber","entail_verdict":"green","rule_verdict":"amber"},"output":{"confidence":0.8,"flags":["clinical_outcome_disclaimer"],"reasons":["coverage=0.80","Clinical-outcome claims require a disclaimer"]}},{"node":"gate","tool":"","event":"INPUT","detail":"verify c_speed","input":{"claim_id":"c_speed","text":"processes a full claims dataset in under 4 hours","source_id":"s_datasheet","source_version":"v_aa9da0278314","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_datasheet (score 14.06)","input":{"claim":"processes a full claims dataset in under 4 hours"},"decision":null,"output":{"evidence":"The platform processes a full claims dataset in under 4 hours","source_id":"s_datasheet","score":14.059}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"no rule constrains","input":{"rule_tags":["performance"]},"decision":{"verdict":null,"flags":[]},"output":{"reasons":[]}},{"node":"nli","tool":"LexicalNLI","event":"DECISION","detail":"coverage=1.00","input":{"claim":"processes a full claims dataset in under 4 hours","evidence":"The platform processes a full claims dataset in under 4 hours"},"decision":{"label":"entailment","entail_prob":0.98},"output":null},{"node":"ensemble","tool":"diverse judge ensemble","event":"DECISION","detail":"mean agreement 0.95 (numeric & superlative are veto lenses)","input":{"escalated":false},"decision":{"votes":{"coverage":1.0,"numeric":0.95,"superlative":0.9},"mean_p":0.95},"output":null},{"node":"calibrate","tool":"isotonic PAV calibrator","event":"DECISION","detail":"entail_raw 0.900 → confidence 0.900 (green)","input":{"entail_raw":0.9},"decision":{"entail_verdict":"green"},"output":{"confidence":0.9}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"c_speed → GREEN (most-severe of entail+rule)","input":null,"decision":{"verdict":"green","entail_verdict":"green","rule_verdict":null},"output":{"confidence":0.9,"flags":[],"reasons":["coverage=1.00"]}},{"node":"gate","tool":"","event":"INPUT","detail":"verify c_deployed","input":{"claim_id":"c_deployed","text":"is deployed across 12 hospital systems","source_id":"s_casestudy","source_version":"v_598c5748478c","rules_version":"r_b5f41bb751a8"},"decision":null,"output":null},{"node":"retrieve","tool":"BM25Okapi (rank_bm25)","event":"TOOL","detail":"top evidence from s_casestudy (score 8.21)","input":{"claim":"is deployed across 12 hospital systems"},"decision":null,"output":{"evidence":"Helix Analytics is now deployed across 12 hospital systems","source_id":"s_casestudy","score":8.208}},{"node":"rules","tool":"rules engine r_b5f41bb751a8","event":"DECISION","detail":"no rule constrains","input":{"rule_tags":["proof"]},"decision":{"verdict":null,"flags":[]},"output":{"reasons":[]}},{"node":"nli","tool":"LexicalNLI","event":"DECISION","detail":"coverage=1.00","input":{"claim":"is deployed across 12 hospital systems","evidence":"Helix Analytics is now deployed across 12 hospital systems"},"decision":{"label":"entailment","entail_prob":0.98},"output":null},{"node":"ensemble","tool":"diverse judge ensemble","event":"DECISION","detail":"mean agreement 0.95 (numeric & superlative are veto lenses)","input":{"escalated":false},"decision":{"votes":{"coverage":1.0,"numeric":0.95,"superlative":0.9},"mean_p":0.95},"output":null},{"node":"calibrate","tool":"isotonic PAV calibrator","event":"DECISION","detail":"entail_raw 0.900 → confidence 0.900 (green)","input":{"entail_raw":0.9},"decision":{"entail_verdict":"green"},"output":{"confidence":0.9}},{"node":"ledger","tool":"","event":"OUTPUT","detail":"c_deployed → GREEN (most-severe of entail+rule)","input":null,"decision":{"verdict":"green","entail_verdict":"green","rule_verdict":null},"output":{"confidence":0.9,"flags":[],"reasons":["coverage=1.00"]}},{"node":"personalize","tool":"fact inliner (gated)","event":"OUTPUT","detail":"inlined 4 gated fact(s); 0 blocked fact(s) withheld","input":null,"decision":{"inlined":["recent_news","ehr_vendor","size_band","intent_topic"],"withheld":[]},"output":{"receipts":[{"key":"recent_news","text":"We saw Northwind Health reported a push to cut administrative cost.","source":"company_news_rss","source_kind":"free","basis":"public_record","verdict":"usable"},{"key":"ehr_vendor","text":"Note: we integrate with Allscripts/Veradigm.","source":"firmographic_sim","source_kind":"paid","basis":"contract_vendor_dpa","verdict":"usable"},{"key":"size_band","text":"Tailored for a 9+ hospital IDN.","source":"firmographic_sim","source_kind":"paid","basis":"contract_vendor_dpa","verdict":"usable"},{"key":"intent_topic","text":"Since Northwind Health is exploring reducing length of stay (inferred), here's what's most relevant:","source":"intent_sim","source_kind":"paid","basis":"contract_vendor_dpa","verdict":"disclaimer"}]}}],"output":{"verified_claims":[{"claim_id":"c_los","verdict":"amber"},{"claim_id":"c_speed","verdict":"green"},{"claim_id":"c_deployed","verdict":"green"}],"usable_facts":7,"blocked_facts":0,"inlined":["recent_news","ehr_vendor","size_band","intent_topic"]},"checks":[{"name":"every shown product claim cleared the Gate (no RED)","pass":true},{"name":"every inlined personal fact carries a lawful basis","pass":true},{"name":"no blocked fact reached the copy","pass":true}],"pass":true}}