Rediscovery benchmark

Integrating thermal imaging-derived breathing pattern features with social media interaction network features in a multimodal deep learning architecture will significantly improve the prediction accuracy of clinical depression severity (PHQ-8 scores) in adolescents compared to unimodal baselines.

IV: Modality of feature integration (multimodal fusion of thermal breathing and social interaction features vs. isolated unimodal features) · DV: Prediction accuracy of depression severity (measured by Mean Absolute Error on PHQ-8 scores)

Mechanism: Thermal breathing patterns serve as a proxy for autonomic nervous system arousal and physiological stress responses, while social media interaction patterns (e.g., posting frequency, sentiment, friend network density) reflect behavioral withdrawal and cognitive rumination associated with depression. A deep multimodal fusion layer aligns these complementary physiological and behavioral embeddings, allowing the network to capture latent depression severity markers that are obscured when either signal is analyzed in isolation.

grounded on: R138927, R139014, R138931, R138934

The substitution chemistry of β-cyclodextrin derivatives dictates a route-specific threshold for mucosal drug permeation enhancement, such that hydroxypropyl substitution optimizes ocular and nasal bioavailability while dimethyl substitution optimizes pulmonary bioavailability.

IV: Cyclodextrin substitution pattern (hydroxypropyl vs. dimethyl vs. methyl vs. maltosyl) · DV: Route-specific drug permeation enhancement factor (ocular/nasal vs. pulmonary)

Mechanism: Hydroxypropyl groups increase cyclodextrin hydrophilicity and steric bulk, favoring complexation and permeation enhancement at the tighter, surfactant-poor ocular and nasal epithelial barriers. Conversely, dimethyl substitution imparts moderate lipophilicity that preferentially disrupts pulmonary surfactant layers and alveolar macrophage membranes, maximizing inhaled drug retention. This substitution-dependent mucosal affinity creates a route-specific permeation enhancement threshold that is not explicitly compared across the individual formulation studies.

grounded on: R151616, R155599, R155595, R151525, R155499, R151520, R151621, R155456

Experimental evolution under constant sublethal antibiotic pressure yields significantly higher MIC fold-changes for agents targeting DNA replication machinery compared to antimicrobial peptides, mediated by the fixation of high-impact target-site mutations rather than low-impact membrane modifications.

IV: Drug target class (DNA gyrase inhibitor vs. antimicrobial peptide) · DV: MIC fold-change relative to ancestor strain

Mechanism: Target-site mutations (e.g., gyrB S446F) confer large, specific resistance jumps by altering drug binding sites, whereas AMP resistance relies on bulk membrane charge/structure changes that saturate quickly, limiting the phenotypic ceiling.

grounded on: R1351005, R1351027, R1385695

Transition metal dichalcogenide (TMD) gas sensors employing a field-effect transistor (FET) architecture will achieve a limit of detection (LOD) for ammonia (NH3) that is at least two orders of magnitude lower than equivalent TMD sensors employing a chemiresistor architecture, driven by electrostatic channel modulation and superior charge transfer kinetics.

IV: Sensor architecture (FET vs. chemiresistor) combined with TMD material identity (MoS2 vs. MoSe2) · DV: Limit of detection (LOD) for NH3 in ppm

Mechanism: The FET configuration converts surface adsorption-induced charge transfer into a modulated drain-source current via the gate field, providing intrinsic signal amplification. Concurrently, MoS2's higher electron mobility and optimal work function alignment with NH3 compared to MoSe2 reduce contact resistance and thermal noise, collectively pushing the detection threshold lower.

grounded on: R139328, R139336

Lattice proton enrichment in layered niobate photocatalysts synergizes with molecular co-catalysts to accelerate visible-light H2 evolution rates by 2.5–4.0 fold relative to alkali-exchanged niobate analogues, because surface-migrated H+ lowers the activation barrier for proton-coupled electron transfer at the co-catalyst interface.

IV: Niobate lattice proton content (H+-rich vs. alkali-rich A-site/B-site composition) · DV: H2 evolution rate (μmol h-1 g-1)

Mechanism: Under visible-light excitation, photogenerated holes oxidize methanol at the niobate valence band, while electrons reduce to the conduction band. In H+-rich niobates, labile lattice protons migrate to the particle surface and adsorb as hydroxyl species, creating a localized acidic microenvironment. This proton reservoir facilitates proton-coupled electron transfer (PCET) at the adjacent molecular co-catalyst (e.g., Ni-amine or Cu2+), reducing the kinetic overpotential for H–H bond formation. The accelerated surface reduction kinetics outcompete bulk electron-hole recombination, yielding a net increase in H2 turnover frequency compared to alkali-exchanged frameworks where surface proton availability is rate-limiting.

grounded on: R46156, R46150, R46158, R46162

Introducing Fe³⁺ surface complexes onto N-doped TiO2 nanoparticles will shift the optimal solution pH for visible-light methylene blue degradation to the acidic range (pH 3.0–4.0) and increase the maximum degradation rate constant by 45–65% relative to neutral pH, due to pH-dependent enhancement of trapped hole lifetimes at Fe-OH surface sites.

IV: Solution pH (adjusted from 3.0 to 8.0) · DV: Apparent first-order visible-light methylene blue degradation rate constant (k_obs)

Mechanism: R45114 demonstrates that solution pH modulates the absorption spectra and lifetime of photogenerated trapped holes in nanocrystalline TiO2 by altering surface Ti-OH protonation. R45116 establishes that hole trapping occurs on ~220 fs timescales and directly governs charge separation efficiency. R46111 shows that Fe³⁺ adsorption onto N-doped TiO2 generates visible-light-active mid-gap states that enhance methylene blue degradation. When combined, Fe³⁺/Fe²⁺ surface complexes introduce pH-sensitive hole-trapping sites that are maximally populated under acidic conditions (pH 3.0–4.0), where protonated Ti-OH₂⁺ surface groups favor Fe-OH⁺ coordination. This extends the trapped hole lifetime beyond the 0.2–0.4 μs baseline reported for pristine films (R45114), increasing the steady-state flux of oxidative holes to adsorbed methylene blue molecules and thereby elevating k_obs.

grounded on: R45114, R45116, R46111, R46091

Integrating exception handling mechanisms into goal models significantly improves the classification accuracy of ambiguous regulatory requirements by explicitly capturing deviation scenarios that mitigate semantic uncertainty during early decision-making.

IV: Integration of exception handling rules in goal models (binary: present vs. absent) · DV: Classification accuracy of ambiguous regulatory requirements

Mechanism: Exception handling rules in goal models explicitly formalize edge cases and regulatory deviations. This formalization reduces semantic ambiguity in downstream requirement specifications, thereby providing clearer decision boundaries for requirements engineers during early uncertainty-driven evaluation phases.

grounded on: R211145, R211198, R211137

Integrating Position-Aware Circular Convolution (ParC) modules into the local processing blocks of Separable Self-attention Mobile Vision Transformers (MobileViTv2 architecture) reduces inference latency by 15-25% on mobile hardware while maintaining Top-1 ImageNet accuracy within 1% of the baseline, due to the lower computational complexity of circular convolution compared to standard convolutions used in local blocks.

IV: Type of local feature extraction module (standard convolutions vs. Position-Aware Circular Convolution) · DV: End-to-end inference latency (ms) and Top-1 classification accuracy (%)

Mechanism: Circular convolution exploits rotational symmetry and fixed kernel weights to compute local features with O(N) complexity instead of O(N*K^2) for standard convolutions, reducing FLOPs in the local processing stage without sacrificing receptive field coverage. When paired with separable self-attention for global modeling, this shifts the computational bottleneck away from the local blocks, which are typically the latency bottleneck on mobile hardware.

grounded on: R1856041, R1856060, R1855993

Integrating HorNet-style recursive gated convolutions into the Context Autoencoder (CAE) backbone as a replacement for linear token mixers will significantly improve semantic segmentation accuracy on high-resolution aerial imagery benchmarks compared to ActiveMLP-based variants.

IV: Token mixing mechanism in self-supervised vision backbones (Recursive Gated Convolutions vs. Active MLP vs. Standard Attention) · DV: Mean Intersection over Union (mIoU) on semantic segmentation

Mechanism: Recursive gated convolutions facilitate efficient high-order spatial interactions without the quadratic computational cost of attention mechanisms. When embedded within the CAE framework, this allows the encoder to capture long-range contextual dependencies more effectively during self-supervised reconstruction. These enriched spatial features transfer more robustly to the UperNet decoder, enhancing boundary delineation and object coherence in high-resolution urban scenes where standard linear token mixers (like ActiveMLP) struggle with global context aggregation.

grounded on: R1802664, R1802991, R1803091, R1802398

The choice of upper-bound MOTU delimitation algorithm systematically modulates the estimated cryptic diversity window in Diptera barcoding, such that model-based clustering approaches (NJ/BINs/GMYC) inflate lower-bound taxonomic counts by 1.8 to 2.6 times more than distance-based methods (Barcoding gap/ABGD) when applied to Neotropical lineages.

IV: Upper-bound MOTU delimitation algorithm type (model-based clustering vs. distance-based gap methods) · DV: Cryptic diversity inflation ratio (upper-bound MOTU count divided by lower-bound taxonomic count)

Mechanism: Model-based clustering algorithms resolve finer phylogenetic breaks and shallow intraspecific structuring characteristic of hyperdiverse Neotropical radiations, whereas distance-based gap methods enforce stricter minimum intraspecific thresholds that artificially lump recently diverged cryptic lineages.

grounded on: R145304, R146643, R145434, R146639

The implementation of Ocean Biogeo Chemistry modules in CMIP5 Earth System Models is structurally dependent on the concurrent inclusion of Sea Ice dynamics to accurately represent high-latitude carbon fluxes.

IV: Implementation of Ocean Biogeo Chemistry module · DV: Inclusion of Sea Ice dynamics module

Mechanism: Sea ice physically regulates surface albedo, freshwater flux, and gas exchange in polar oceans; these physical processes are fundamental drivers of biogeochemical cycles. Therefore, a carbon cycle module cannot be physically decoupled from sea ice dynamics without introducing structural inaccuracies in high-latitude simulations, creating a mandatory coupling in model architecture.

grounded on: R23287, R23471, R23408, R23326

The explicit coupling of Atmospheric Chemistry to Aerosol dynamics in Earth System Models is a necessary structural precondition for simulating aerosol-mediated land-ice albedo feedbacks, as chemical transformation governs the formation and deposition flux of light-absorbing secondary aerosols onto cryospheric surfaces.

IV: Interactive coupling of Atmospheric Chemistry to Aerosol dynamics (coupled vs. uncoupled/passive treatment) · DV: Simulated land-ice surface albedo reduction and associated mass balance anomaly

Mechanism: Atmospheric chemistry modules simulate gas-phase oxidation and nucleation of precursor emissions into secondary organic carbon and black carbon aerosols. These chemically aged particles exhibit higher hygroscopicity and light-absorbing capacity, enhancing their wet/dry deposition onto land ice. Surface deposition darkens the cryosphere, lowering albedo, increasing shortwave absorption, and accelerating melt-driven mass loss. Without interactive chemistry, models underestimate secondary aerosol lifetimes and deposition efficiency, functionally decoupling aerosol forcing from cryospheric response.

grounded on: R23368, R23383, R23398, R23300

In ALD-grown Yb2O3:Er nanolaminate thin films, substituting 25 at.% of Yb3+ sensitizer sites with La3+ will disrupt the Yb→Er energy transfer pathway, decreasing the NIR-to-visible electroluminescence intensity ratio by 45–65% under electrical excitation at 200°C while simultaneously reducing the thermal quenching rate of visible emission by >30% compared to undoped controls.

IV: La3+ substitution ratio at Yb3+ sites in Yb2O3:Er ALD nanolaminates (0% vs 25% atomic fraction) · DV: Ratio of NIR upconversion emission intensity (980 nm) to visible emission intensity (550 nm) under constant electrical excitation at 200°C

Mechanism: La3+ incorporation in oxide hosts modifies local cation coordination and introduces deeper electronic trap states, as demonstrated in La-doped SrTiO3 gate dielectrics (R1469867) and La/Y/Dy-doped ZrO2 electrolytes (R1469781). When La3+ substitutes Yb3+ in Yb2O3:Er nanolaminates, these trap states act as energy transfer barriers between Yb sensitizer clusters and Er activator sites, interrupting the multiphonon-assisted Yb→Er transfer pathway that drives enhanced electroluminescence in Al2O3:Yb,Er nanolaminates (R1469756). Consequently, the dominant radiative decay shifts from NIR upconversion to visible transitions, which exhibit temperature-dependent luminescence behavior characteristic of Ln3+-doped hosts used in optical thermometry (R1469874). The baseline high EQE of 8.5% in Yb2O3:Er (R1469850) confirms efficient native energy transfer that La substitution is predicted to selectively attenuate.

grounded on: R1469867, R1469781, R1469756, R1469850, R1469874

Substituting the β-diketonate (thd) ligand on yttrium precursors with an amidinate (famd) ligand in O3-oxidized ALD of YMnO3 will increase the growth per cycle (GPC) by 200–350% relative to the Y(thd)3 baseline, due to faster surface ligand exchange kinetics inherent to the amidinate coordination geometry.

IV: Rare-earth precursor ligand class (β-diketonate/thd vs. amidinate/famd) · DV: Growth per cycle (GPC) in angstroms per cycle

Mechanism: Amidinate ligands coordinate to the metal center in a planar N,N-bidentate fashion with longer, weaker metal-ligand bonds compared to the chelating O,O-bidentate β-diketonate ligands. This reduced bond strength and steric accessibility lower the activation energy for ligand exchange during the precursor pulse, enabling higher surface coverage and greater mass uptake per cycle without requiring elevated temperatures.

grounded on: R1470236, R1470264, R1470333

Increasing the steric bulk of ligands on ALD metal precursors reduces the temperature sensitivity of the growth-per-cycle (GPC) but concurrently decreases the asymptotic film density.

IV: Ligand steric bulk of the metal precursor molecule (e.g., dimethylamido vs. ethylmethylamido vs. diethylamido vs. isopropylamide) · DV: Temperature coefficient of GPC (|d(GPC)/dT|) and film density

Mechanism: Sterically demanding ligands increase the activation barrier for complete surface reaction saturation, effectively broadening the self-limiting temperature window and flattening the GPC-temperature curve. However, the same steric repulsion impedes tight surface packing during adsorption, trapping volatile byproducts and creating a more open, less dense amorphous network compared to reactions driven by small ligands or halides that pack more efficiently at lower temperatures.

grounded on: R676142, R676153, R676169, R676130

In deep learning models for COVID-19 CT detection, reported classification accuracy is inversely proportional to the number of training scans, a relationship that holds across both encoder-decoder and encoder-only architectures due to dataset curation bias and heterogeneity effects.

IV: Number of CT scans used for model training · DV: Reported classification accuracy (ACC/AUC)

Mechanism: Smaller cohorts (N<500) are typically curated from single centers or high-quality repositories with homogeneous pathology, reducing label noise and distribution shift, which artificially inflates accuracy. Larger cohorts (N>1000) aggregate multi-center data with greater clinical heterogeneity, ambiguous ground truth, and varied scan protocols, introducing noise that degrades apparent accuracy. This data-distribution effect dominates architectural capacity differences, making the inverse trend architecture-agnostic.

grounded on: R700923, R700931, R700920, R700955, R700959, R675126

Increasing the temporal quantization bin count (B) in event-based surface representations improves stereo depth estimation accuracy by aggregating asynchronous spikes into denser motion cues, up to a point where temporal resolution loss limits correspondence precision.

IV: Temporal quantization bin count (B) in event surface generation · DV: Depth estimation accuracy (measured as absolute relative error)

Mechanism: Temporal quantization converts asynchronous, sparse event spikes into dense, time-aligned surfaces, enhancing motion cue density for stereo matching algorithms used in depth prediction. This aggregation improves correspondence precision until excessive binning degrades temporal resolution, causing motion blur that limits depth accuracy.

grounded on: R642467, R642470, R1411261, R1411264, R1411295, R1411274, R1411281

In onshore wind energy assessments, the interaction between temporal resolution and length of time series systematically biases wind power density estimation, such that high-frequency/short-duration datasets yield significantly lower power density estimates than low-frequency/long-duration datasets, independent of wind speed height.

IV: Dataset configuration defined by the interaction of temporal resolution (sampling interval) and length of time series (collection duration) · DV: Estimated annual mean wind power density (W/m²)

Mechanism: Short-duration, high-frequency recordings capture intense but transient gusts that disproportionately affect cubic wind speed averaging, yet lack seasonal coverage to anchor the Weibull shape parameter. This yields a skewed probability density function that underestimates the cumulative energy potential. Conversely, long-duration, lower-frequency records smooth turbulence but capture full seasonal cycles, producing robust PDF fits that better approximate true annual energy yield. Wind speed height acts as a scaling factor but does not alter the directional bias introduced by the temporal-length interaction.

grounded on: R707531, R709006, R704955, R707519, R707596

Solar radiation prediction models utilizing meteorological drivers (Cloudiness Index, Sky Coverage, Relative Humidity) will exhibit significantly lower seasonal performance degradation in high-variability periods (spring/autumn) compared to models relying solely on historical radiation time series, as meteorological inputs capture the physical determinants of irradiance variability that temporal extrapolation cannot resolve.

IV: Input feature category (Meteorological drivers vs. Historical radiation time series) · DV: Seasonal performance degradation (RMSE increase in spring/autumn relative to annual mean)

Mechanism: R1563739 establishes that high meteorological variability in spring and autumn degrades the performance of models relying on historical data patterns (RF, ANN, SP). R1563732, R1563748, and R1563736 establish that direct meteorological drivers (Pressure, RH, Wind, Temp, Cloudiness Index, Sky Coverage) are valid predictors for solar radiation. R1566035 confirms that solar radiation variability is physically driven by these meteorological factors. By directly measuring the instantaneous physical causes of irradiance changes, meteorological models can adapt to non-stationary patterns, whereas time-series models fail because they assume stationarity or smooth transitions that break down during high variability.

grounded on: R1563739, R1563732, R1563748, R1563736, R1566035

Application of mixture kernel density models to truncated wind speed data reduces the uncertainty in estimated annual energy yield by 25-40% compared to single-parameter distributions.

IV: Distribution Model Type (Mixture Kernel Density vs. Single-Parameter Weibull) · DV: Uncertainty in Annual Energy Yield (measured by 95% Confidence Interval width)

Mechanism: Wind speed data truncation (e.g., cut-in/cut-out effects) distorts the probability distribution tails, increasing the sensitivity and variance of energy yield estimates (R709022). Single-parameter distributions like Weibull cannot adapt to this distorted shape, propagating the truncation-induced error into the energy yield calculation. Mixture kernel density models (R707542) offer flexible, non-parametric fitting that captures the multi-modal or truncated structure of the data. By accurately modeling the truncated distribution, these mixtures reduce the propagated uncertainty in the final energy yield estimation (R709096).

grounded on: R709022, R707542, R709096, R707902