What problem does it solve? Scientific and engineering questions often have multiple plausible explanations, and teams waste effort running expensive experiments that do not discriminate between them. This Skill structures the process of generating competing hypotheses, attacking each with the cheapest discriminating experiment, and identifying which test resolves the most uncertainty per unit of cost. ## Core Features & Use Cases - Hypothesis Refraction: Splits one research question into 2-4 competing hypotheses, each with a mandatory falsification condition and a structured experimental-design artefact (PICO, PECO, or generic experimental setup). - Constructive Falsification Attacks: Spawns one attacker subagent per hypothesis to propose the cheapest experiment discriminating it from its siblings, with cost tiers from trivial to prohibitive. - Bayesian Synthesis: Produces a prior-update landscape, identifies the single highest information-gain-per-cost crucible experiment, and outputs an ordered research roadmap plus a learning objective. - Use Case: An engineering team sees a 4.3 dB attenuation anomaly on a fiber run. The Skill generates three hypotheses (splice loss, bend-radius violations, intrinsic fiber loss), proposes an OTDR-based discriminating test for each, and recommends the cheapest experiment that rules out the most alternatives first. ## Quick Start Ask the AI to run engage-crucible on your research question, for example: use engage-crucible to figure out why our model accuracy dropped after the data pipeline change.
