Truace
Subscribe
The Index recomputed live from the record

What the evidence says.What the public feels.

The record holds 933 sourced gains and 770 sourced problems, averaging 68 and 66 on the index score. Readers have logged 22 public signals on the Pulse, which is kept apart and never counted as evidence.

933 gains
770 problems
Every sourced claim in the Index, one square each, shaded by the strength of its evidence. High Moderate Emerging

Ranks distinct AI gain and problem claims from the published record. Scores reward impact, independent source strength, scale, confidence, and recency.

1,703 results
Show filters and sorting

Download every matching row, not just this page:Export CSVExport JSON

AI gains · 933

209
HealthStableModerate evidence · 1 source

A six-protein plasma signature tested with SVM distinguished fibrotic hypersensitivity pneumonitis from idiopathic pulmonary fibrosis on an independent test set with 71.4% accuracy, offering a non-invasive diagnostic aid.

Between July 2018 and June 2022, investigators enrolled 119 subjects across healthy controls, non-fibrotic HP, fibrotic HP, and IPF cohorts and performed plasma proteomic profiling with WGCNA. They identified 813 proteins, noted enrichment of glycolysis/gluconeogenesis and pyruvate metabolism in FHP, and distilled nine differential proteins to a six-protein signature that was used to train seven machine learning classifiers.

Impact 30%
63
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
84

Updated Jul 19, 2026 · TRV-2026-0264

71Index score
210
CrimeStableModerate evidence · 1 source

Listeners detected ElevenLabs-cloned speech more accurately when audio was transmitted through narrowband GSM (AMR-NB) at 63.7%, compared to other telecom conditions.

By June 2026, researchers tested how telecom transmission affects human detection of cloned speech. They created natural and ElevenLabs-synthesized utterances from nine speakers, processed them through simulated GSM, VoLTE, and VoIP codecs, and asked 95 participants to classify them as human or synthetic.

Impact 30%
63
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
83

Updated Jul 13, 2026 · TRV-2026-0152

71Index score
211
HealthNewly addedModerate evidence · 1 source

Parent-report ML model combining expressive vocabulary and joint attention items achieved over 0.7 sensitivity and specificity for both autism and developmental delay at 18-month visits, approximately doubling sensitivity versus existing screens.

Researchers developed and tested a parent-report screening algorithm called TADAS for the 18-month pediatric visit. Using data from 11,878 toddlers who completed M-CHAT-R/F, Q-CHAT-10-O and ASQ-3, plus extended assessments in a subsample, they trained gradient-boosting models on synthetic data and evaluated them against ADOS-2 and Mullen diagnoses in an authentic holdback sample.

Impact 30%
55
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
98

Updated Sep 30, 2026 · TRV-2026-1218

70Index score
212
HealthStableModerate evidence · 1 source

Researchers were able to extract asthma control and exacerbation data from routine EMRs using SQL and a large language model, suggesting it may be feasible to assess clinical remission in Japanese patients initiating triple therapy.

Researchers tested whether clinical remission in asthma could be measured from routine electronic medical records in Japan by analyzing 32,258 patients starting fluticasone furoate/umeclidinium/vilanterol triple therapy in the JAMDAS database from 2020 to 2024, extracting ACT scores with SQL and exacerbation-related hospitalizations and emergency transports with a large language model.

Impact 30%
55
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
95

Updated Sep 16, 2026 · TRV-2026-1108

70Index score

AI problems · 770

209
HealthStableModerate evidence · 1 source

LLMs answering periodontal questions perform significantly worse in Persian than English and often produce explanations too technical for average patients, lacking clinical nuance for personalized risk assessment.

On 2026-09-16, a peer-reviewed comparative study reported testing 10 large language models on 15 common periodontal questions in English and Persian, with blinded ratings by two board-certified periodontists across six criteria. Performance varied significantly by model architecture, language, and question type, with the most advanced configuration providing highly accurate and well-referenced answers.

Impact 30%
49
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
95

Updated Sep 18, 2026 · TRV-2026-1131

68Index score
210
HealthStableModerate evidence · 1 source

Despite high accuracy, AI fundus diagnostics saw limited clinical adoption due to poor generalizability and disjointed workflow integration.

A systematic review of 34 studies from 2018-2025 found AI systems for fundus image diagnosis of diabetic retinopathy, glaucoma, ROP and AMD frequently exceeded 90% accuracy and matched or outperformed experts, yet adoption in eye clinics remained limited.

Impact 30%
49
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
95

Updated Sep 18, 2026 · TRV-2026-1127

68Index score
211
HealthStableModerate evidence · 1 source

Medical AI faces ambiguity in liability determination due to fragmented multi-source legal texts and persistent difficulties in tracing algorithmic biases through decision flows.

Researchers built a legal knowledge graph for medical AI compliance using four node types and four directed legal relations, then applied a relation-aware graph convolutional network with edge-type weights and attention to align cross-modal legal semantics. They used gradient-based backpropagation to extract bias-propagation subgraphs and community discovery to locate root-node clusters, mapping real-time decision flows to temporary nodes for compliance checking.

Impact 30%
49
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
95

Updated Sep 17, 2026 · TRV-2026-1123

68Index score
212
HealthStableModerate evidence · 1 source

The predictive models remain exploratory with small event counts and no external validation, so they cannot yet be used clinically to mitigate adverse events from enfortumab vedotin therapy.

In a proof-of-concept study published September 16, 2026, investigators used real-world data from 542 patients with advanced urothelial carcinoma to train machine learning models to predict toxicities from enfortumab vedotin, including severe adverse events and specific events like diarrhea and cutaneous toxicity.

Impact 30%
49
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
95

Updated Sep 17, 2026 · TRV-2026-1121

68Index score

Recomputed live from the record · Oct 12, 2026, 2:46 AM