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

237
ScienceStableHigh evidence · 5 sources

Deep learning models composed of multiple processing layers improved state-of-the-art performance in speech recognition, visual object recognition, object detection, drug discovery and genomics by 2015

Published in May 2015, this peer-reviewed overview describes deep learning as models with multiple processing layers that learn multi-level representations of data, trained via backpropagation to adjust parameters between layers

Impact 30%
49
Evidence 25%
100
Scale 20%
35
Confidence 15%
100
Recency 10%
83

Updated Jul 13, 2026 · TRV-2026-0209

70Index score
238
HealthStableHigh evidence · 5 sources

Random Forest survival model achieved comparable predictive accuracy to Cox regression on colon cancer mortality data, with low concordance error.

Published in 2017, this methods paper explored Random Forest as an alternative to Cox regression for survival analysis. Using 66,807 colon cancer cases from the SEER database, the authors built both a Cox model and a Random Forest model to derive mortality-associated risk factors and compared their predictive performance.

Impact 30%
49
Evidence 25%
100
Scale 20%
35
Confidence 15%
100
Recency 10%
83

Updated Jul 13, 2026 · TRV-2026-0208

70Index score
239
ScienceStableHigh evidence · 4 sources

A deep neural network trained on DFT data learned a transferable potential that predicts total energies for organic molecules with chemical accuracy versus DFT, generalizing to systems up to 54 atoms despite training on smaller molecules.

On Feb 8 2017, researchers described ANI, a deep neural network architecture that uses atomic environment vectors derived from modified symmetry functions to learn molecular energies from quantum mechanical DFT calculations, and introduced Normal Mode Sampling to efficiently sample potential surfaces.

Impact 30%
49
Evidence 25%
100
Scale 20%
35
Confidence 15%
100
Recency 10%
83

Updated Jul 13, 2026 · TRV-2026-0206

70Index score
240
ScienceStableHigh evidence · 5 sources

Decision-tree ensemble approach called deep forest achieves excellent performance across different domains using the same default setting while requiring fewer hyper-parameters and automatically determining model complexity without gradient-based adjustment or backpropagation.

As of the 2018-10-08 publication, researchers proposed deep forest, a decision-tree ensemble that stacks non-differentiable modules in layers to replicate deep learning characteristics without using backpropagation for training.

Impact 30%
49
Evidence 25%
100
Scale 20%
35
Confidence 15%
100
Recency 10%
83

Updated Jul 13, 2026 · TRV-2026-0205

70Index score

AI problems · 770

237
HealthStableModerate evidence · 1 source

Students with medium and low baseline history-taking proficiency showed relatively limited score improvements from LLM-VSP self-practice, with practice frequency alone not independently predicting final performance.

In a 2026 prospective cohort study, 168 third-year medical students were grouped by voluntary use of an LLM-powered virtual standardized patient system for extracurricular history-taking practice versus routine instruction alone. After propensity score matching to 40 pairs, the LLM-VSP group scored higher on an end-of-term Objective Structured Clinical Examination with real standardized patients.

Impact 30%
49
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
94

Updated Sep 10, 2026 · TRV-2026-1044

68Index score
238
HealthStableModerate evidence · 1 source

Most hand surgery AI tools are deployed in unaudited workflows and rarely remeasured after release after testing only on training-like data, leaving the hand surgeon accountable for patient outcomes shaped by opaque models.

On September 9, 2026, a peer-reviewed article in The Journal of Hand Surgery described how neural networks, machine learning models, and large language models are entering hand surgery for radiograph reading, outcome prediction, and chart drafting, while noting that adoption has outpaced validation and proposing a four-part stewardship framework for surgeons.

Impact 30%
49
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
94

Updated Sep 10, 2026 · TRV-2026-1043

68Index score
239
HealthStableModerate evidence · 1 source

Pediatric medical imaging is substantially underrepresented in AI development and validation, with scarce datasets and off-label use of adult models risking bias and patient safety.

On 2026-09-09, an AJR Expert Panel Narrative Review assessed the state of artificial intelligence in pediatric radiology, finding that while AI has transformed general radiology, pediatric imaging remains substantially underrepresented in development, validation, regulation, and implementation.

Impact 30%
49
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
94

Updated Sep 10, 2026 · TRV-2026-1042

68Index score
240
HealthStableModerate evidence · 1 source

AI tools translating radiology reports into plain language can produce translation errors that compromise comprehension and safety, causing reports to be graded unsafe and warrant withholding from patients.

From February to December 2025, researchers developed a 5-attribute rubric  clarity, content, certainty, tone, verbosity  to grade AI-generated patient-friendly radiology reports, using survey-workshop cycles with 19 participants and testing with ChatGPT-4.1 and Claude-4.0 outputs from public radiology impressions. Evaluation involved six research-team members and 111 additional participants, plus AI evaluation with ChatGPT-5, comparing rubric grades to prespecified reference standards and to subjective decisions about withholding unsafe reports.

Impact 30%
49
Evidence 25%
95
Scale 20%
35
Confidence 15%
87
Recency 10%
94

Updated Sep 10, 2026 · TRV-2026-1041

68Index score

Recomputed live from the record · Oct 12, 2026, 6:35 AM