Capacity Lives Between the Parts

An ACS paper reports a supramolecular gel made by evaporation-induced co-assembly of poly(vinyl alcohol), phytic acid and lignosulfonate into a ternary hydrogen-bond network. At the optimized lignosulfonate content, the material reached 267 kPa tensile strength and 0.2676 S/m ionic conductivity, remained flexible and conductive after seven days at −20 °C, responded after 3,000 seconds of cyclic loading and heated above 65 °C under 0.1 W/cm² solar irradiation; paired with a thermoelectric generator, it drove a miniature fan. The record is a publisher abstract for a laboratory material, not evidence of long-term wear, repeated thermal cycling or manufacturing yield.

A Frontiers paper combines a 1 mm TPU plate carrying three resonator geometries with a 15 mm polyurethane sound-absorbing layer, then tunes the structure with a Kriging surrogate model and genetic algorithm. In a semi-anechoic test at 2,000–5,000 rpm, four microphones one metre from the motor measured RMS reductions below 2 kHz of 11.20, 9.11, 7.54 and 6.06 dB(A), respectively. The authors say the setup could not directly quantify the frequency–transmission-loss relationship; standing-wave-tube validation, manufacturing tolerance, durability and service life remain open.

A conformable triboelectric e-skin patch captures labial motion without an external sensor battery and feeds the signal to a multilayer convolutional neural network. The authors report 98.75% across 15 lexical categories with 200 samples each, but this is a non-independent laboratory estimate: the main dataset came from three healthy adults, participant-specific samples appeared in both training and test splits, and test-set loss was monitored for early stopping. A separate four-person leave-one-subject-out evaluation is described, but its exact results are in supplementary material; comfort, speech-impairment diversity and long-duration wear were not clinically evaluated.

A paper formalizes precision weeding as a coupled active-perception and manipulation problem for a quadruped mobile manipulator, enumerating feasible task-multiplexing configurations before applying a two-stage optimizer. Across simulation and a physical robot experiment, the authors report fewer stops, replans and path length, with 36.1% average energy reduction against their baseline. The accessible primary abstract does not expose field conditions, trial scale, baseline details or statistical robustness, so the result remains a bounded robotics experiment rather than farm deployment evidence.

ComfyUI 0.42.0 widened its open visual-node system with MOSS-image support, a Krea 3 partner node, Qwen 3.x performance work and embedding support, ARM Linux tests, 3D remeshing changes and additional portable hardware builds. Version 0.42.1 followed three hours and 23 minutes later to fix crashes some users experienced in the MOSS image-template workflow. Both records are first-party release notes, with no independent performance verification; together they show compatibility and repair arriving on the same graph boundary.