Published evidence dashboard

Findings

Every card shows observed outcomes, the published estimate the engine contradicted, the divergence, and the cross-validation chain. Confirmed findings replicate across two independent datasets and are pre-registered on OSF (osf.io/3ws8g). Divergence percentages reflect combined-evidence rates from cross-validated cohorts where applicable. Per-dataset rates are shown separately for transparency.

Pre-registration discipline: RocSite Discovery does not publish specific quantitative findings on this site without an active OSF pre-registration. The engine identifies many candidate contradictions; public claims are reserved for findings that have completed pre-registration, replication, and methodology review. Findings under validation appear in the Pipeline section without specific cohort numbers.

Confirmed · replicated on two datasets Exploratory · single dataset, OSF-timestamped Pipeline · identified, under validation

Click any evidence hash below to independently verify the engine, protocol version, and timestamp that produced the finding.

Confirmed

Cross-validated across two independent datasets

· 6 findings
Myocardial Infarction × Age > 70
Discovery ID: f8c6412e-17c2-4234-b93e-da7b622d09ff
Confirmed
Observed
21.5%
Published
8.0%
Divergence
+81.9%
MIMIC-IVn = 2,161 · 464 deaths
eICU-CRD confirmationn = 4,069 · mortality 14.55%
Evidence hash6bf704b3907a061b
OSF timestamp2026-04-22T19:34:10Z
Evidence chain
  • Pre-registrationosf.io/3ws8g — Cross-Validated Priority Findings
  • Discovery timestamp2026-04-22 19:34:10 UTC
  • Analysis runoutlier_engine · RocSite Discovery Engine
  • ConsensusAdvocate / Adversary / Arbiter unanimous
  • Replication dataseteICU-CRD 2.0 · 209 US hospitals
Diabetes × Age > 70
Discovery ID: 0c4edd84-c249-465f-86c8-d4f09a720a9d
Confirmed
Observed
13.9%
Published
6.0%
Divergence
+125.7%
MIMIC-IVn = 4,711 · 655 deaths
eICU-CRD confirmationn = 997 · mortality 13.54%
Evidence hash7d2a16e7e901e602
OSF timestamp2026-04-22T19:34:10Z
Evidence chain
Atrial Fibrillation × Age > 70
Discovery ID: a6cb0b26-babe-4540-93b8-73c6c0f27882
Confirmed
Observed
16.6%
Published
10.0%
Divergence
+66.3%
MIMIC-IVn = 11,432 · 1,901 deaths
eICU-CRD confirmationn = 10,369 · mortality 16.26%
Evidence hash7f21bfbc3b83f069
OSF timestamp2026-04-22T19:34:10Z
Evidence chain
Seizure × Age > 70
Discovery ID: 9b051d86-5853-48b1-afee-1477e5349daa
Confirmed
Observed
45.8%
Published
7.5%
Divergence
+510.2% MIMIC-IV
+101.6% eICU
MIMIC-IVn = 59 · 27 deaths
eICU-CRD confirmationn = 1,422 · mortality 15.12%
Evidence hash9b051d86
OSF timestamp2026-04-23T01:04:31Z
Replication noteMIMIC-IV (small cohort, n=59) shows 45.8% mortality with a wide confidence interval. eICU on a larger cohort (n=1,422) shows 15.12% — a more conservative estimate that still substantially exceeds the published ~7.5% benchmark. The pre-registered MIMIC finding stands; the eICU replication tempers the magnitude.
Evidence chain
  • Pre-registrationosf.io/3ws8g — Cross-Validated Priority Findings
  • Analysis runoutlier_engine · RocSite Discovery Engine
  • NoteLargest published-literature gap in the series. Small MIMIC-IV cell size; eICU replication provides broader base-rate anchor.
Pulmonary Embolism × Age > 70
Discovery ID: 660b12cc-0214-4e25-9545-a0ef10413464
Confirmed
Observed
23.0%
Published
11.5%
Divergence
+99.6% MIMIC-IV
+36.0% eICU
MIMIC-IVn = 305 · 70 deaths
eICU-CRD confirmationn = 933 · mortality 15.65%
Evidence hash660b12cc
OSF timestamp2026-04-23T01:04:31Z
Replication noteMIMIC-IV and eICU show partial replication; rates differ across datasets (23.0% vs 15.65%). Both exceed the published ~11.5% benchmark.
Evidence chain
COPD × Age < 50
Discovery ID: c8179898-47ec-4746-b60e-228292b2bb66
Confirmed
Observed
4.0%
Published
14.0%
Divergence
−71.4%
MIMIC-IVn = 50 · 2 deaths
eICU-CRD confirmationn = 648 · mortality 4.01%
Evidence hashc8179898
OSF timestamp2026-04-23T01:04:31Z
Replication noteSmall MIMIC-IV cohort (n=50, 2 deaths); eICU strongly confirms the negative-direction divergence at near-identical 4.01% (n=648). Replication is the load-bearing evidence here, not the MIMIC point estimate.
Evidence chain
Exploratory

Single dataset · OSF-timestamped · replication pending

· 0 findings

No findings currently meet the Exploratory bar. Single-dataset signals must have an active OSF timestamp linking to a real project before appearing here. See Pipeline below for findings under validation.

Pipeline

Identified by the engine, awaiting pre-registration and replication

· ~80 findings

The engine surfaces approximately 80 candidate contradictions currently moving through the RocStars consensus validation queue. These are not listed publicly until they pass internal adversarial review, receive a pre-registration OSF stamp, and clear a replication check.

Currently under validation (selected):

  • Type 2 diabetes, young-adult ICU mortality (temporal trend)
  • Pulmonary embolism, elderly mortality (temporal trend)
  • ARDS, elderly survival (subgroup direction flip)
  • Vasopressor liberal-use signal in septic shock cohorts

Partners and licensees see the full pipeline inside the client portal.

Have a paper, model, or claim you'd like independently tested?

Run it through the RocSite contradiction engine and get a signed Evidence Integrity Score, defensible for FDA, journal review, or investor due diligence.

Request an Evidence Integrity Score →