What problem does it solve? Product questions about a Grok Bot are usually answered by changing the live Bot first and reading the damage later. This Skill replaces that pattern with a local simulation loop: frame falsifiable hypotheses, replay kept logs and run arithmetic locally, and produce a readout where every number traces back to a reproducible run log. ## Core Features & Use Cases - Hypothesis framing: Restates each product question as a claim that can fail, with a measure and a fail line, plus written assumptions fixed before any number is produced. - Local simulation with kept logs: Replays kept run logs, sweeps stated parameter ranges, or computes arithmetic with python3, recording inputs, code, seeds, and outputs so any readout can be rerun to the same numbers. - Honest readouts: Reports effect with interval, verdict (supported, unsupported, or underpowered), the assumptions it leans on, and the single cheapest next live measurement. - Use Case: Before shortening a draft prompt on a live Bot, simulate the variant against kept logs to test whether it cuts mean draft length by 15% without raising owner rewrite rate, then ship the readout with its reproducing log. ## Quick Start Ask the assistant to use the model-guided-research skill to test what would happen if a prompt, threshold, or schedule changed on the Bot, before making any live change.
