New papers. Better decisions.
A research update matters when it changes a claim, a model, an experiment or a commercial requirement.
The review pipeline
Discover primary sources; deduplicate DOI and versions; obtain methods where accessible; extract operating conditions and claim boundaries; compare with the current registry; identify a decision affected; retain access limits. Preprints remain distinct from peer-reviewed journal articles.
What we extract
Biology/material, active versus projected area, full-cell versus biased half-cell, illumination, load, output duration, replicate identity, uncertainty, energy inputs, raw data availability and possible competing explanations. Peak values with incompatible normalizations are not ranked as equivalent performance.
Research watch
A private daily research brief is scheduled for the owner around 8 a.m. America/New_York, with deeper Monday synthesis and a monthly evidence-gap review. It is a review queue, not an automatic public claims feed. This page shows the review method and the dated source set; fresh findings are not silently presented as verified.
Machine-learning boundary
Retrieval and extraction assistance can accelerate review, but every important claim needs a verifiable source passage. Experimental prediction requires independent run/day validation, simple baselines, uncertainty estimates and the ability to abstain.