Our Work

The Correction Freeze Problem (CFP) describes four independent ways in which a corrective mechanism can fail to fire. The Error Correction Model (ECM), together with its threshold and band varian…

The Correction Freeze Problem (CFP) describes four independent ways in which a corrective mechanism can fail to fire. The Error Correction Model (ECM), together with its threshold and band variants (TVECM and BVECM), is the standard econometric tool for measuring correction. This paper proposes a formal link between the two. We argue that the ECM gives CFP’s abstract update rule an estimable form: a negative adjustment coefficient 𝛼 marks successful correction, and CFP supplies conditions under which 𝛼 could fall to zero within a regime. The TVECM and BVECM then correspond to CFP’s threshold and cost mechanisms. We take Mekong River sediment starvation as a promising first case in which the threshold parameter 𝛾 might be estimated from obtainable data, and we lay out an estimation protocol built on Mekong River Commission monitoring records, dam construction dates and published sediment load series. The paper reports no estimates. It closes with a general estimation strategy and a list of open issues, chiefly how CFP’s threshold on increments relates to the TVECM threshold on levels, and which series should play the part of the perceived state

Evidica is a local literature-review workspace: search across PubMed, OpenAlex and Europe PMC, human triage, citation management, and a manuscript editor in which every claim is tied to the exact sour…

Evidica is a local literature-review workspace: search across PubMed, OpenAlex and Europe PMC, human triage, citation management, and a manuscript editor in which every claim is tied to the exact source passage a person chose, with that decision kept in a hash-linked event log and exported as an audit package.

This release provides a frozen and auditable World Bank panel for monitoring water and sanitation services in 26 Latin-American and Caribbean economies from 2000 through 2023. The primary dataset cont…

This release provides a frozen and auditable World Bank panel for monitoring water and sanitation services in 26 Latin-American and Caribbean economies from 2000 through 2023. The primary dataset contains 624 country-year rows assembled from ten World Development Indicators series: total, urban and rural basic drinking-water coverage; total, urban and rural basic sanitation coverage; under-five mortality; population; GDP per capita; and urban population share.The release includes the exact ten World Bank API JSON responses used to construct the panel, a provenance registry with SHA-256 digests, the derived panel, validation reports, eight analytical validation tables, seven figures, a PowerShell 7 offline replay verifier and a live-reacquisition utility for future versioned vintages.The frozen replay validates all ten source hashes, reconstructs all 624 country-year rows, compares 6,864 numeric cells with the released panel, and reports zero mismatches and a maximum absolute numeric difference of 0. World Bank aggregate codes are excluded by an explicit LAC-26 ISO3 country lock, and missing observations remain missing.The analytical tables and figures are supplied as technical validation and examples of reuse. They include descriptive service levels, urban-rural gaps, cross-country dispersion, descriptive beta-convergence, an associational fixed-effects model relating the WASH service index to under-five mortality, and a descriptive typology. These outputs are not causal estimates.

Continuous in-hive temperature monitoring can replace invasive manual inspections, but machine learning (ML) studies in precision beekeeping rarely control temporal data leakage or check whether model…

Continuous in-hive temperature monitoring can replace invasive manual inspections, but machine learning (ML) studies in precision beekeeping rarely control temporal data leakage or check whether models fit on the target microcontroller. This study evaluates both aspects for the automation of a physiologically grounded thermal alert rule on an ESP32-based node. Three DHT22 sensors (brood chamber, honey super and exterior) were deployed on a Langstroth hive in Arequipa, Peru, from 12 February to 15 June 2026 (420,614 records). Hourly aggregation yielded 2232 windows with 12 temperature features, labelled Normal, Cold Stress, Heat Stress or Critical by a threshold heuristic anchored on the 34–35 °C brood-nest optimum. Decision Tree (DT), Random Forest (RF) and Logistic Regression (LR) classifiers were compared with a majority-class baseline and with the rule itself on a chronological 80/20 hold-out, and their serialized sizes were screened against the ESP32 memory. DT reached 96.4% accuracy (95% CI 94.3–97.8%) and 98.8% balanced accuracy with a 3.1 KB footprint (0.6% of SRAM); all its errors were conservative Critical over-alerts. RF scored 100% on this autumn-dominated period, but its 549.4 KB serialized model exceeds the 520 KB SRAM; LR reached 95.7% (balanced accuracy 89.1%). Because labels derive from the inputs, these metrics measure rule-reproduction fidelity; the directly coded rule remains the simplest implementation, and the ML pipeline serves as a controlled benchmark for retraining once expert labels exist. No model was executed on the ESP32, so the results establish memory feasibility rather than validated deployment.

Numerical source code and data accompanying the article “Exact replica-sector hierarchy of multi-resolvent correlations in random free fermions”. The deposit contains the complete set of s…

Numerical source code and data accompanying the article “Exact replica-sector hierarchy of multi-resolvent correlations in random free fermions”. The deposit contains the complete set of scripts used for the numerical verification, together with their machine-readable output files and the figures generated from them.Contents. 232 jl files, 8 py files; 48,649 lines in total.Article abstract. The spectral weight with which a many-body eigenstate contributes to a given channel is not fixed by the smooth one-point ETH envelope. Its higher-order correlations involve joint fluctuations of spectral weights across many eigenstates, and their energy-resolved structure is naturally encoded by connected products of channel resolvents---an organization that has been lacking. We construct one. In random free fermions, where every channel overlap is a squared Slater minor of a Haar-orthogonal one-body eigenvector matrix, the r-resolvent moment admits an exact replica-sector decomposition: a sum over the double-coset sectors of the orthogonal Weingarten algebra, in which the permutation secto …What the code computes. Numerical protocol app:num The Monte Carlo draws real GOE one-body Hamiltonians h in the Mehta normalization ( h_aa^2=2, h_ab^2=1), diagonalizes them to obtain the Haar one-body eigenvector matrix U and levels E_a, and builds the many-body levels _A= _a AE_a and the exact overlaps p_A^C=| U[A,C]|^2 for all D= Nk label and mode sets (Julia, MKL backend). All quantities in Sec. sec:bulk are bin-averaged over the sixteen channel-triple patterns: the direct cumulant C^(3) uses the exact flat-envelope centering _U[p]=1/D, i.e. R_a= R_a- 1D (z-H)^-1 per realization; the sector reconstruction uses Eq. eq:bulk3 with _3 computed per realization from the many-body spectrum and, at (14,2), the closed-form kernel table; at (8,4) the sector values _3 are measured in the same ensemble, so the check tests the sector dependence of together with the U factorization. The deep-plane coefficients use the moment fields s_r= (H^r)_aa in the channel basis with the channel Hamiltonian built by second quantization and checked elementwise; with bin-averaging over the channel-triple patterns, K_111 is consistent with its exact zero within the bin noise (machine precision at (8,4)), and K_211 agrees with the …Requirements. Julia packages: Printf, Serialization, LinearAlgebra, Random, Statistics, MKL, FFTW. Python modules: symbolica, fractions, re, json, collections, importlib, math. All numerical results were produced on a single workstation (CPU, MKL backend); no external data are required and every script is self-contained.Funding. National Natural Science Foundation of China, grant 12305035. (Zenodo 的 funders 词表当前为空,资助信息写入 description。)License. Code released under the MIT License; see LICENSE. If you use this code, please cite the article above and this deposit.

This open educational resource (OER) module provides a mathematically rigorous, systems-level foundation in Biomedical Engineering, bridging physical principles, computational tools, and clinical prac…

This open educational resource (OER) module provides a mathematically rigorous, systems-level foundation in Biomedical Engineering, bridging physical principles, computational tools, and clinical practice.  Covering multi-scale physiology from cellular mechanics and biomaterials to organ-level cardiovascular and neuro-technologies, the module integrates governing continuum laws—including Hagen-Poiseuille vascular flow, Fickian membrane mass transport, and lumped-parameter Windkessel hemodynamics. Modern 2026 computational paradigms are integrated, focusing on Physics-Informed Neural Networks (PINNs) for continuous blood velocity estimation, Fourier Neural Operators (FNOs) for rapid non-linear soft-tissue deformation, and invariant filtering for assistive robotics.  Features include:- A step-by-step engineering calculation analyzing post-stenotic vascular resistance and balloon angioplasty recovery.- A multi-variable systems engineering trade-off matrix spanning orthopedics, neural interfaces, prosthetic heart valves, and bioelectronics.- An interactive single-compartment pharmacokinetic elimination model.- Three tiers of comprehensive practice problems: conceptual review, scenario analysis, and quantitative numerical evaluations with fully worked solutions. Permanent web resource: https://prep4uni.online/stem/physical-technologies/biomedical-engineering/

Why does a complex system obey the particular laws it does, rather than others? Wepropose a single answer, made precise: the laws active at any scale are the symmetries ofthe energy minimum the system…

Why does a complex system obey the particular laws it does, rather than others? Wepropose a single answer, made precise: the laws active at any scale are the symmetries ofthe energy minimum the system occupies.Formally, for a state ϕ0 that minimizes an energy functional E(ϕ), the operative symmetry group is the ISOTROPY (stabilizer) subgroup G0 = {g : g.ϕ0 = ϕ0} the largest group of transformations leaving that minimum invariant and the invariants of G0 are the conserved quantities and selection rules of the eective physics there.This is a generalization of Landau's theory of phase transitions and the standard account of spontaneous symmetry breaking: the minimum carries less symmetry than the functional, and the unbroken remainder is exactly G0. The at (zero-Hessian) directions at the minimum are the broken generators, and by Goldstone's theorem they are the massless modes a rigorous bridge between the geometry of the basin and the particle content of the theory.Emergence, in this view, is the selection of WHICH minimum the system falls into; the laws of nature at a given scale are the isotropy group of that basin.We illustrate the principle across the early-universe vacuum, galactic dynamics, and associative-memory networks, and we note how it grounds the symmetry structure of the DSM-861 program. The Standard Model gauge group is used as an illustration of the principle, not derived from it.

Archive of the CodonPausingKit release v1.0.0, that freezes the code in place for our publication in Science Advances.

This dataset contains the pretrained DeepCov-EnKF forecast model and B-matrix estimation model, together with model output arrays used in the study. The forecast and B-matrix model files were split in…

This dataset contains the pretrained DeepCov-EnKF forecast model and B-matrix estimation model, together with model output arrays used in the study. The forecast and B-matrix model files were split into numbered parts for reliable upload; the included restoration script and manifest allow them to be reconstructed and verified. The study code is available at 10.5281/zenodo.20993446, and the processed input data at 10.5281/zenodo.22908836.