What problem does it solve? Physics calculations are chains of transformations where each intermediate result carries uncertainty from its inputs. Without systematic propagation, final error bars are missing or guessed, and effort is wasted improving parameters that barely affect the result. This Skill traces how input uncertainties flow through every intermediate step to the final quantity, producing an explicit error budget that identifies the dominant error source. ## Core Features & Use Cases - Dependency Tree Tracing: Builds the complete derivation chain from leaf input parameters through intermediate results to the target quantity, using state.json depends_on chains and SUMMARY.md provides/requires sections across phases. - Multi-Source Uncertainty Cataloging: Classifies error sources as statistical, systematic, parametric, truncation, numerical, and model uncertainties, including approximation validity-boundary checks and catastrophic cancellation detection. - Sensitivity and Correlation Analysis: Computes analytical or numerical sensitivity coefficients, handles correlated errors via covariance propagation, and supports Monte Carlo cross-checks for non-Gaussian cases. - Use Case: After completing several project phases with tracked intermediate results, run the command with --target on a final observable to generate an ERROR-BUDGET.md ranking each parameter's contribution, showing that e.g. 90% of the uncertainty comes from one input so you know where to invest precision-improvement effort. ## Quick Start Ask the AI to run error propagation on a target quantity, for example: propagate uncertainties through the derivation chain for the coupling constant and produce an error budget identifying the dominant error source.
