What problem does it solve? Vague forecasts like "X will happen soon" can never be checked or scored, so decisions about timing rest on unfalsifiable claims. This Skill turns a timing question into one dated, numeric-confidence prediction with the evidence, watchlist, and retraction conditions needed to track and later score it. ## Core Features & Use Cases - Dated falsifiable prediction: Produces a single milestone with an absolute date and a numeric confidence drawn from defined confidence bands (0.5–0.7 marked directional, below 0.5 withheld). - Accelerants and blockers with lead times: Lists three observable forces on each side, each with a lead time in months, so a wait-or-commit decision can be timed against early-warning windows. - Weak signals and kill-signals: Commits to three weak signals with named observation sources and three kill-signals that would force retraction, plus a review date. - Use Case: A product team debating whether to wait for on-device LLM inference gets a dated prediction (e.g., ">50% of new enterprise summarisation deployments on-device by 2027-06-30, confidence 0.65, directional"), the signals to watch in procurement portals and release notes, and the events that would retract the call. ## Quick Start Use the foresight skill to turn the question "when will solid-state batteries reach mass-market EVs?" into a dated prediction with accelerants, blockers, weak signals, kill-signals, and a review date.
