Skip to content
#explainable ai Open access

Intercognix Cross-Domain Discovery Report — The nanofluid thermal-conductivity reproducibility problem

Sep 2026 · Zenodo (CERN European Organization for Nuclear Research)

Abstract

Intercognix Research Report 002 (run ID 20260913220518-4881c7, 13 September 2026, 108 pages) documents an AI-assisted cross-domain exploration of a long-standing problem in thermal engineering: why nominally identical suspensions of nanoscale particles in a base fluid yield effective thermal-conductivity enhancements that differ across laboratories by far more than measurement uncertainty can explain, and why classical effective-medium theory cannot reconcile the spread. The exploration began with four unresolved phenomena: the large laboratory-to-laboratory scatter in reported enhancement for nominally identical samples; the presence in some studies but not others of anomalous, non-monotonic, or unexpectedly large enhancement with particle loading, size, or temperature; the unresolved question of whether the discrepancy stems from an unmodelled physical mechanism, an uncontrolled material or process variable, or a systematic measurement artifact; and the absence of a unifying model that predicts enhancement consistently across preparation methods and measurement techniques. Intercognix examined these questions through four independent cross-domain discovery paths, generating 18 candidate hypotheses (a further 7 candidates were set aside during formulation). A separate adversarial review process then challenged each hypothesis through citation checking, prior-work searches, and assessment of assumptions, conflicting evidence, novelty, feasibility, and falsifiability. Following this internal review: 7 hypotheses remained candidates for further testing 10 require reformulation 1 was rejected The seven surviving directions, with the distant field each was drawn from, are: RA-02 — Loading-direction hysteresis (shallow-lake regime-shift ecology): whether apparent enhancement depends on whether a sample was assembled by adding particles to reach a loading or by diluting a more concentrated stock. RC-03 — Preparation dose–state recovery map (single-grain luminescence dating): calorimetrically measured absorbed specific energy (J/g) as an instrument-transferable dose metric for sonication and dispersion. RB-04 — Measurement-window duration as artifact inflator (clinical-trial meta-analysis): a corpus-level meta-regression of reported enhancement on measurement window duration. RD-01 — Wait-to-measure interval (mycorrhizal network transport): elapsed time between the end of dispersion and the start of measurement, logged as a controlled covariate. RA-01 — Probe-induced micro-stirring artifact (measurement reactivity in experimental psychology): whether the heated probe itself induces local flow that inflates apparent conductivity. RB-02 — Calibrator commutability gap (clinical immunoassay harmonization): whether instruments that agree on a calibration fluid still disagree on a particle-laden fluid. RD-05 — Spatial concentration variability (mycorrhizal network transport): spatial coefficient of variation of local particle concentration in the probed volume at the moment of measurement. Several of these converge on the same underlying suspicion: that a substantial part of the irreproducibility is procedural and measurement-related (preparation history, timing, instrument geometry, data-reduction choices) rather than evidence of unmodelled physics. That convergence is itself a testable claim, not a conclusion. These are not established findings or validated mechanisms. They are testable hypotheses proposed for further examination. Each is stated with a proposed measurable variable, a decisive experiment, and an explicit falsifier. The report includes supporting and conflicting evidence, prior-work searches (including engagement with the International Nanofluid Property Benchmark Exercise), citation corrections, assumptions, proposed falsifiers, feasibility concerns, the 10 hypotheses requiring reformulation, the rejected hypothesis, and unresolved questions. Disclosure on review. The hypotheses were generated and challenged by separate AI processes within the Intercognix workflow. The "independent reviewer" referred to in the report is a separate AI process. It is not independent human peer review and should not be interpreted as such. No human domain expert has reviewed this report at the time of deposit. Known limitations identified after the run. The verdict labels ("novelty established," "ready to test now") are generated by the pipeline's tiering rules. The dossiers beneath them are more cautious and should be read as the primary record. Several surviving hypotheses, including RB-04, RD-01, RB-02 and RD-05, have no fully verified citation; RB-04 also has two citations the review could not locate. Two hypotheses (RB-05 and the rejected RB-01) contain citations flagged by the review as misattributed or non-scientific. The survival rate in this run (7 of 18) is much higher than in the Alzheimer's run (2 of 25). The most likely reason is that a reproducibility problem admits many measurement-side hypotheses that are cheap to test; it should not be read as evidence that the hypotheses here are stronger. This was a limited, independent, self-funded run (4 discovery paths, 18 hypotheses). It is not an exhaustive survey of candidate mechanisms. Invitation. This report is published openly to invite scrutiny from thermal and fluids engineers, materials scientists, metrologists, and researchers working on measurement reproducibility. Researchers are specifically invited to identify prior work that may have been missed, weaknesses in the cross-domain analogies, unsupported physical assumptions, alternative explanations, experimental limitations, or evidence that would falsify the proposed hypotheses. Critiques may be sent to connect@intercognix.com, citing the hypothesis ID and section. The objective is not to make AI-generated hypotheses sound convincing. The objective is to make unexpected connections specific enough to examine, challenge, reject, refine, or test. This is version 1.0, deposited unedited as produced by the pipeline apart from a prepended publication note.

View source

Similar papers

#artificial intelligence Conference Open access Apr 2020

ECCOLA - a Method for Implementing Ethically Aligned AI Systems

The method, ECCOLA, is presented, which aims at making the high-level AI ethics principles more practical, making it possible for developers to more easily implement them in practice.

Ville Vakkuri, Kai-Kristian Kemell, P. Abrahamsson · 64 citations · ⚡6
#computer vision Review Apr 2024

AI-powered Code Review with LLMs: Early Results

The goal is to not only refine the accuracy of the LLM-based tool but also to underscore its potential in streamlining the software development lifecycle through proactive code improvement and education.

Z. Rasheed, Malik Abdul Sami, Muhammad Waseem et al. · 62 citations · ⚡3
#computer vision Open access Mar 2024

LLM-based agents for automating the enhancement of user story quality: An early report

The use of large language models to automatically improve the user story quality in Austrian Post Group IT agile teams is explored, with a reference model for an Autonomous LLM-based Agent System developed and implemented at the company.

Zheying Zhang, M. Rayhan, Tomas Herda et al. · 48 citations · ⚡4
#computer vision Review Mar 2024

System for systematic literature review using multiple AI agents: Concept and an empirical evaluation

This paper introduces a novel multi-AI-agent system designed to fully automate SLRs, and demonstrates how it substantially reduces the time and effort traditionally required for SLRs while maintaining comprehensiveness and precision.

Abdul Malik Sami, Z. Rasheed, Kai-Kristian Kemell et al. · 44 citations · ⚡2
#computer vision Feb 2024

Can Large Language Models Serve as Data Analysts? A Multi-Agent Assisted Approach for Qualitative Data Analysis

The proposed LLM-based multi-agent system automates qualitative data analysis process, creating opportunities for researchers and practitioners, and future improvements focus on enhancing multilingual performance and integrating continuous expert feedback.

Z. Rasheed, Muhammad Waseem, Aakash Ahmad et al. · 41 citations
#artificial intelligence Conference Open access Jun 2018

The Key Concepts of Ethics of Artificial Intelligence

It is suggested that the focus on finding keywords is the first step in guiding and providing direction for future research in the AI ethics field.

Ville Vakkuri, P. Abrahamsson · 39 citations · ⚡2

Related blog posts

Microsoft Research Blog Oct 7, 2026

Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses

Training AI agents with reinforcement learning can be challenging because their tools, context, and decision-making are managed by complex frameworks. Agent Lightning connects existing agents to RL training, making it easier to improve them without rebuilding them. The post Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses appeared first on Microsoft Research.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.