AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

uncertainty-and-units

Tracks physical units and propagates measurement uncertainty in scientific calculations using pint and uncertainties. Use for unit conversion and dimensional checking, GUM uncertainty budgets, Type A and Type B evaluation, coverage factors and expanded uncertainty, Monte Carlo propagation, significant-figure and plus-minus reporting, error propagation through curve fits, CODATA constants, auditing Python code for stripped units or broken uncertainty propagation, and order-of-magnitude plausibility checks using dimensionless groups (Reynolds, Peclet, Damkohler, Knudsen, Biot, Womersley), characteristic scales such as diffusion time or Debye length, and observed magnitude ranges. Trigger on "is this number physically reasonable", "sanity check these units", "what regime is this flow in", or a result that looks off by orders of magnitude.

Original source
K-Dense-AI
License
MIT
Pinned revision
5f2bc0341354fd5590aef76a533ada1e1e5b4e62

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

This skill also requires local files or dependencies. Remote reading alone may not be enough.

Ask your assistant

Use Agentic Memory skills to read k-dense-ai--uncertainty-and-units at revision 5f2bc0341354fd5590aef76a533ada1e1e5b4e62, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 8

references/review.md

Review and execution scope — 2026-10-01

This skill targets Pint 0.26.1, uncertainties 3.2.3, NumPy 2.5.3 and SciPy 1.18.1, executed together on Python 3.13.3 in an isolated environment. Pint now requires Python 3.12+ and NumPy 2+. Core local workflows require no credentials or service API. No authenticated external scientific service was invoked.

Reviewed contracts

  • Pint: registries/application registry, Quantity conversion, offset and logarithmic arithmetic, context names and parameters, wraps, check, NumPy dispatch, custom definitions, formatting, and ufloat magnitudes. The 0.26 release adds offset string parsing, without making ambiguous offset arithmetic valid. Runtime checks cover spectroscopy, chemistry, Boltzmann and Gaussian conversions. Other bundled context definitions were inspected in released source, not exhaustively executed.
  • uncertainties: ufloat, identity, correlated_values, correlated_values_norm, covariance/correlation matrices, derivatives, components, umath/unumpy, formatting, and wrap. Correlated variables use independent latent variables internally; their derivative dictionary is not indexed by the original correlated objects. Deprecated absolute-value helpers were replaced in the bundled evaluator.
  • NumPy/SciPy: array functions and RNG distributions, positive-semidefinite correlation factors, quantiles, SD, Student-t factors, erf, CODATA accessors and weighted curve_fit covariance. np.log accepts an output buffer as its second positional argument; the evaluator explicitly implements a logarithm base instead.
  • JCGM: Type A/B interpretation, covariance propagation, independent-input Welch-Satterthwaite limitations, reporting, numerical tolerance and the adaptive Monte Carlo requirement. The bundled fixed-trial comparison remains diagnostic: endpoint_agreement is a boolean and gum_framework_validated is null.
  • Unit semantics: angular versus cyclic frequency, molecular mass versus molar mass, ppm kinds, pH activity, wavelength in a medium, exact versus measured constants, and uncertainty units for offset/logarithmic quantities.
  • Plausibility: all 14 group formulas, 8 scales and 22 band conversions executed. Corrected reversed Damkohler interpretation and removed unconditional equilibrium, breakup, tracer-validity and thermal-noise conclusions. Bands remain explicitly illustrative; historical textbook sources were not all independently reread.

Execution evidence and limits

The per-skill isolated suite passes 113 tests. A separate native smoke run passes 114 numerical/API checks and confirms five relevant released source files match the installed wheels byte-for-byte. Documented command-line workflows and all six help commands were executed. These are synthetic numerical checks, not measurement traceability, calibrated coverage, experimental validation, clinical reference ranges, or a certification of any physical model.

pint-pandas and pint-xarray are optional ecosystem pointers verified against their current official documentation; neither integration was installed or executed. Arbitrary user model/fit fragments are illustrative and need actual input data and an appropriate error model. The helper does not sample Student-t inputs or estimated variance uncertainty, propagate context-parameter uncertainties, or prove convergence of reciprocal/heavy-tailed distributions. Finite samples with no NaN do not establish that a distribution has a finite mean or variance.

Primary sources

Some live documentation headers identify development builds (Pint) or the 1.18.0 manual (SciPy). Released-source checks and the stated installed versions bound the execution claims. No general guarantee is inferred from a documentation URL returning HTTP 200. The citation record was checked at the current arXiv landing page.