Evidence (8974 claims)
Search and filter individual claims pulled from the papers. Looking for a specific finding ("what's the effect on wages?"), you're in the right place. Want to compare whole outcome categories against each other instead? Use the Evidence Explorer.
The board below groups claims two ways: by broad theme (nine paper-level topics) and by outcome category (the 34 claim-level outcomes that the Explorer and Syntheses also use).
Browse by theme
Nine broad, paper-level topics. Click one to filter the claims below.
Adoption
10085 claims
Filter claims →
Productivity
8974 claims
Filtered →
Governance
8062 claims
Filter claims →
Human-AI Collaboration
7749 claims
Filter claims →
Org Design
5057 claims
Filter claims →
Innovation
4896 claims
Filter claims →
Labor Markets
4088 claims
Filter claims →
Skills & Training
3372 claims
Filter claims →
Inequality
2377 claims
Filter claims →
Claims by outcome category
Counts by direction of finding. These are the same 34 outcome categories the Explorer compares and the Syntheses are written for. A linked row has a published synthesis.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 882 | 244 | 117 | 1097 | 2424 |
| Governance & Regulation | 1010 | 469 | 229 | 135 | 1875 |
| Organizational Efficiency | 977 | 235 | 149 | 90 | 1462 |
| Technology Adoption Rate | 781 | 299 | 143 | 128 | 1362 |
| Research Productivity | 506 | 155 | 74 | 363 | 1110 |
| Output Quality | 555 | 219 | 71 | 70 | 915 |
| Decision Quality | 395 | 200 | 95 | 54 | 751 |
| Firm Productivity | 523 | 67 | 101 | 27 | 724 |
| AI Safety & Ethics | 262 | 309 | 75 | 36 | 688 |
| Market Structure | 195 | 201 | 135 | 30 | 566 |
| Task Allocation | 248 | 77 | 96 | 38 | 464 |
| Innovation Output | 300 | 34 | 55 | 20 | 411 |
| Skill Acquisition | 207 | 75 | 65 | 21 | 368 |
| Employment Level | 138 | 67 | 119 | 24 | 350 |
| Fiscal & Macroeconomic | 156 | 80 | 53 | 33 | 329 |
| Task Completion Time | 211 | 38 | 13 | 16 | 280 |
| Firm Revenue | 183 | 52 | 29 | 5 | 270 |
| Consumer Welfare | 131 | 77 | 48 | 13 | 269 |
| Inequality Measures | 50 | 141 | 54 | 9 | 254 |
| Worker Satisfaction | 104 | 85 | 25 | 13 | 227 |
| Error Rate | 87 | 112 | 11 | 5 | 215 |
| Automation Exposure | 69 | 69 | 37 | 20 | 198 |
| Wages & Compensation | 102 | 49 | 31 | 11 | 193 |
| Team Performance | 115 | 30 | 30 | 11 | 187 |
| Regulatory Compliance | 88 | 74 | 17 | 7 | 186 |
| Training Effectiveness | 109 | 22 | 14 | 21 | 168 |
| Developer Productivity | 116 | 21 | 15 | 8 | 161 |
| Job Displacement | 12 | 92 | 26 | 1 | 131 |
| Hiring & Recruitment | 57 | 12 | 9 | 5 | 83 |
| Skill Obsolescence | 6 | 59 | 10 | 2 | 77 |
| Social Protection | 43 | 17 | 8 | 2 | 70 |
| Creative Output | 35 | 21 | 9 | 4 | 70 |
| Labor Share of Income | 18 | 23 | 17 | 1 | 59 |
| Worker Turnover | 15 | 16 | — | 4 | 35 |
| Industry | — | — | — | 1 | 1 |
Productivity
Remove filter
The positive impact of AI application on enterprise innovation efficiency is stronger in state-owned enterprises.
Heterogeneity/subsample analysis using firm ownership status in the 2012–2023 A-share panel showing larger effects for state-owned enterprises.
The positive impact of AI application on enterprise innovation efficiency is stronger in firms located in central and western regions of China.
Heterogeneity/subsample analysis on the 2012–2023 panel of Shanghai and Shenzhen A-share listed firms showing larger estimated effects for firms in central and western regions.
The effect of AI application on enterprise innovation efficiency is mediated by enterprise ESG performance.
Mediation analysis on Shanghai and Shenzhen A-share listed firms (2012–2023) demonstrating a significant mediating role for ESG performance.
The effect of AI application on enterprise innovation efficiency is mediated by the enterprise's data factor utilization level.
Mediation analysis (empirical) using the 2012–2023 A-share firm panel showing significant mediating effects of data factor utilization.
The effect of AI application on enterprise innovation efficiency is mediated by improvements in enterprise "new-quality productivity".
Mediation analysis (empirical) on the 2012–2023 panel of Shanghai and Shenzhen A-share listed firms showing a significant mediating role for new-quality productivity.
AI application can significantly improve enterprise innovation efficiency.
Empirical analysis of Shanghai and Shenzhen A-share listed enterprises using panel data from 2012–2023; baseline regressions showing a significant positive relationship between AI application measures and enterprise innovation efficiency.
The review synthesizes fragmented evidence and links AI use to SME performance improvements, while outlining directions for future research on sustainable AI adoption.
Self-description of the article's contribution based on the authors' focused literature review (2016-2024).
Cloud-based AI solutions, targeted employee training, and explainable AI are identified strategies to overcome AI adoption challenges in SMEs.
Recommendations synthesized from the reviewed literature (2016-2024); presented as enabling strategies rather than results from a single empirical intervention).
AI supports more data-driven financial planning for SMEs.
Identified across the reviewed empirical and conceptual studies in the 2016-2024 literature (synthesis rather than new empirical estimate).
AI enables real-time fraud detection for SMEs.
Synthesis of empirical and conceptual literature reporting AI applications in fraud detection (review-level claim; no aggregated quantitative effect provided).
AI enables more accurate credit risk assessment for SMEs.
Review synthesizing studies on credit scoring and risk assessment within the 2016-2024 corpus (no single pooled sample size or unified effect estimate provided).
AI improves cash flow and financial forecasting for SMEs.
Synthesis of empirical studies and conceptual papers in the 2016-2024 literature reviewed (review article does not report primary sample sizes/effect estimates).
AI offers strong potential to enhance the financial stability and growth of SMEs when supported by suitable organizational capacities and governance.
Focused review of high-quality research (2016-2024) synthesizing empirical and conceptual studies on AI applications in SME finance (no single-sample primary data reported).
There is a need for privacy-preserving deployments and richer, structure-aware representations of human knowledge for practical use.
Authors' recommendation/conclusion drawn from observed accuracy/limitations and privacy considerations in using long-term Slack logs.
Gemini 2.5 Flash achieved the lowest error (MAE 21.13%).
Reported model evaluation results comparing MAE across models; Gemini 2.5 Flash reported as lowest with MAE 21.13%.
We analyze 27,188 messages from 43 users to investigate whether LLMs can infer individual domain knowledge from long-term Slack logs.
Dataset description reported in the paper: 27,188 Slack messages from 43 users.
Our project website, including the leaderboard, dataset, and code, is available at https://dong7313.github.io/muse-benchmark/.
Statement in abstract and provided URL pointing to project artifacts.
Together, MUSE provides a realistic benchmark and evaluation framework for advancing Text-to-CAD from geometric generation toward true engineering design.
Paper's stated contribution and intended purpose (abstract) and provision of dataset/benchmark artifacts via project website.
To enable scalable evaluation, we use a rubric-based visual language model (VLM) judge and validate its reliability through human annotation.
Method and validation claim in abstract stating use of rubric-based VLM and validation against human annotations.
The final stage uses design-specific rubrics to assess functionality, manufacturability, and assemblability, moving beyond shape matching toward practical design quality.
Paper's description of the benchmark's evaluation rubric and intended assessment criteria (abstract).
MUSE pairs practical design instances with structured Design Specifications and evaluates generated models through a three-stage protocol: code check, geometric check, and design-intent alignment.
Methodological description in abstract indicating dataset pairing and three-stage evaluation protocol.
We introduce MUSE, a Text-to-CAD benchmark focused on complex, editable boundary representation (B-Rep) assemblies.
Paper contribution / dataset creation described in abstract; supported by project website and accompanying dataset/code.
Two non-negotiable design requirements guide the architecture: cognitive-load redistribution (DR1) and bounded autonomy with alignment (DR2).
Design requirements explicitly stated in the paper guiding the HARMONY architecture.
The model introduces 'Orchestration Leverage' as a candidate productivity metric suited to human–agent hybrid systems.
Conceptual proposal within the paper (new metric introduced as part of HARMONY).
We propose HARMONY (Hybrid Agentic Research Model for Organisational New Yield), a four-pillar socio-technical architecture comprising ResOps (Industrialized Execution), the Control Tower (Strategic Visibility and Drift Detection), the Ethics Fabric (Bounded Autonomy by Design), and the Talent Studio (Sciencepreneur Capability).
Design Science Research artifact (proposed operating model described in the paper).
Augment Engineering completes a three-discipline progression: Prompt Engineering (one tool), Context Engineering (reproducible pipelines), Augment Engineering (a portfolio of tools across domains).
Conceptual framing presented in the paper describing a proposed progression of disciplines.
A Wright's Law fit (n = 82 artifacts, p < 0.01) shows production acceleration across the artifact portfolio.
Quantitative model reported in the paper: Wright's Law fit on 82 artifacts with reported p-value < 0.01.
A Cochran-Armitage trend test (n = 200 interactions across two chat LLMs, p < 0.01) shows first-pass acceptance rising with prompt-sophistication level.
Quantitative test reported in the paper: Cochran-Armitage trend test on 200 interactions across two chat LLMs, reported p-value < 0.01.
A 5-month formative case study (Nov 2025 to Mar 2026) documents a single practitioner applying Augment Engineering skills across a ten-component orchestration stack spanning seven professional domains, producing work products that would traditionally involve separate domain specialists.
Case study reported in the paper describing one practitioner's activities over five months across a 10-component stack in seven domains; sample size = 1 practitioner.
The paper presents a six-phase orchestration methodology and four portability metrics for Augment Engineering.
Stated methodological contribution within the paper (description of methodology and metrics).
Augment Engineering is a discipline of orchestrating multiple purpose-built AI tools across distinct professional domains, applying prompt and context engineering as portable competencies that transfer across tool boundaries.
Definition and conceptual development presented in the paper (methodological contribution).
Prompt engineering (interaction-level optimization) and context engineering (structured input pipeline design) are domain-portable meta-skills: a practitioner who masters them can apply them to any purpose-built AI tool in any domain.
Conceptual claim supported by the paper's argumentation and exemplified by a single-practitioner case study.
The framework has implications for digital health, education, AI personalisation, and personal agency.
Authors' discussion in paper of potential implications across these application domains; presented qualitatively.
The authors list six operational requirements for state-aware systems.
Explicit statement in paper that six operational requirements are listed; descriptive rather than empirically tested in abstract.
The authors derive seven testable predictions from the state-aware framework.
Explicit statement in paper that seven testable predictions are derived from the framework; no individual prediction effects quantified in abstract.
The paper is supported by a 24-month observational base from a deployed behavioural platform spanning more than 200,000 consented users across four occupational personas (research period 2023 to 2026).
Empirical dataset described in the paper: observational deployment over 24 months, >200,000 consented users, four occupational personas, timeframe given (2023–2026).
The framework is motivated by six strands of established evidence: causal inference, predictive processing, allostasis, attentional bottleneck, chronobiology, and computational psychiatry.
Explicit statement in paper describing the literature strands used to motivate the framework.
Taken together, these claims imply that the outcome of a given event is controllable, conditionally, on the state-trajectory at the time of intervention.
Synthesis/implication drawn by authors from the conceptual framework and the six literature strands; argued but not quantified in abstract.
The conscious channel through which outcomes are reportable is a narrow attentional bottleneck whose contents are themselves state-dependent.
Theoretical claim supported by attentional bottleneck literature cited in the paper; presented as part of the conceptual framework.
The weighting vector (state) is dynamic at sub-daily timescales.
Claim motivated by chronobiology and related literature cited in the paper; authors state the sub-daily dynamism as part of their framework.
The relationship between state, decision, and outcome is causal rather than correlational.
Argument grounded in causal inference literature cited by the authors; presented as a core theoretical claim in the paper rather than demonstrated by a specific randomized experiment in the abstract.
A state can be defined as the time-indexed weighting vector over the dimensions that govern how an individual's biology, physiology, and neuropsychology process the next event into a decision and an outcome.
Explicit definitional claim / framework component introduced by the authors; justified conceptually via multidisciplinary literature cited in the paper.
Human outcomes are controllable in a precise and operational sense through interventions that target the state and its weighting at the moment a decision is being formed.
Theoretical argument in the paper, motivated by the six literature strands; supported in part by the authors' deployed behavioural platform (see separate claim about dataset) but no randomized effect sizes reported in abstract.
This persistent variability belongs in a dynamic latent state of the person (i.e., is best modelled as a time-varying latent state).
Conceptual claim supported by integration of six strands of established evidence (causal inference, predictive processing, allostasis, attentional bottleneck, chronobiology, computational psychiatry) cited in the paper.
Within-person variability persists: the same individual, presented with the same observable input, produces different outcomes on different occasions, and different individuals produce divergent outcomes that no observable covariate fully predicts.
Statement motivated by literature review across behavioural sciences; argued in paper as empirical puzzle rather than proven with new statistics in this manuscript.
Agents share successes and failures to reduce redundant exploration during long-running experiments.
Design of AutoScientists includes mechanisms for recording and sharing experimental outcomes; asserted benefit in paper that this reduces redundant exploration (qualitative and supported by experimental comparisons).
Applied without modification across all 217 ProteinGym assays, the same method improves over the prior state of the art by +6.5% (Spearman correlation).
Empirical evaluation across all 217 assays in the ProteinGym benchmark; reported aggregate improvement in Spearman correlation versus prior state-of-the-art.
On ProteinGym fitness prediction, AutoScientists discovers a method for ACE2-Spike binding that improves over the current state-of-the-art model by +12.5% in Spearman correlation.
Empirical evaluation on the ACE2-Spike assay within the ProteinGym benchmark; reported relative improvement in Spearman correlation versus prior state-of-the-art.
On GPT training optimization, AutoScientists continues discovering improvements from a starting champion where the single-agent approach finds none (7 vs. 0 accepted improvements).
Empirical comparison of discovered/accepted improvements during GPT training optimization; counts of accepted improvements for AutoScientists (7) versus single-agent approach (0).
On GPT training optimization, AutoScientists reaches a target validation bits-per-byte 1.9x faster than Autoresearch.
Empirical training-time comparison between AutoScientists and Autoresearch on GPT training optimization tasks; reported speedup multiplier to reach a validation bits-per-byte target.