rdk

Produce a detailed phase-by-phase runtime breakdown of a single `prank predict` run (JVM boot, class-loading, Groovy config compile, model load, PDB parse, surface, feature extraction, prediction, output) on the current JRE. Use when asked where p2rank prediction time goes, to profile startup/runtime overhead, to find what to optimize, or to compare launchers/JVMs at sub-phase granularity.

Arm a new autoresearch performance-research round for this repo's SASA kernel: pick/confirm a scope, scaffold autoresearch/runs/<date>_<slug>/ from the template (seeded with the already-closed leads from META.md so they're not re-tried), point the dispatcher KICKOFF at it, start the LOG, commit, and ready the /loop. Use when the user wants to start, arm, or set up a new autoresearch round / a new optimization investigation round on the molecular-surface kernel. NOT for executing an already-armed round (that is `/loop execute the autoresearch kickoff`).