spike

What problem does it solve? Deciding whether an idea deserves engineering investment often relies on intuition rather than evidence. This Skill structures exploratory spikes: it reproduces current production behaviour as a validated baseline, designs a falsifiable experiment, measures both gain and regression, and records a directional verdict with inspectable evidence. ## Core Features & Use Cases - Baseline Reproduction: Locates real production behaviour and representative data, then validates reproduction fidelity against known examples before comparing alternatives. - Falsifiable Experiment Design: Defines success bars, primary and regression metrics, representative and adversarial slices, and controls for leakage or confounding. - Evidence-Backed Verdicts: Produces a standalone reproducible artifact folder with figures, tables, and numeric evidence, pressure-tests conclusions through adversarial debate, and posts a tracker verdict comment. - Use Case: A team suspects a new ranking algorithm could outperform the current one. Run a spike to reproduce the production baseline offline, measure improvement and regression on representative and adversarial slices, and receive a stop/iterate/plan verdict plus a /sy:plan-ready brief if the idea clears the bar. ## Quick Start Ask the AI to run a spike on your idea, providing the problem statement, the success bar, and the data anchors it should use as ground truth.