TYEclipse/dsh-probstat
Seven deterministic probability and statistics tools for dsh, aimed at the maths agents most often get wrong — z-tables, distribution cdf/quantiles, intervals, tests and sample sizes.
Each tool replaces a step where a model typically improvises: reading a z-table, inverting a distribution, picking critical values, deciding a test, sizing a study, or eyeballing an A/B comparison. Because the results come from closed-form code rather than recall, the same inputs give the same outputs and the intermediate values can be checked by hand. The README frames the target squarely as the failure modes it removes — hallucinated table values, cdf/quantile errors, shaky compound-event identities and guessed sample sizes.
Install
dsh plugin --profile web add github:TYEclipse/dsh-probstatThe GitHub install line from the README's Install block. The repo is not published to npm, so the github: form is the documented route. The README states zero runtime dependencies, so nothing else is pulled in.
Compatibility
Zero runtime dependencies, pure deterministic computation — no model call sits between the inputs and the numbers. Tools: dist_calc (pdf/cdf/survival/quantile/stats over six distributions), z_score (standard-normal table maths), confidence_interval (mean z/t plus Wilson proportion), the event-probability identities (union/intersection/conditional/Bayes), single-sample hypothesis tests, required sample size for a target error (Wald or Wilson), and two-sample tests (Welch or pooled variance, two proportions).
Details
- Repo: TYEclipse/dsh-probstat
- Category: Learning & Research
- Stars: 1
- Version: github tag (no npm release); see repo releases
Alternatives
TecFancy/dsh-deeptutor · PerryLink/dsh-fund-research · gychen-NJU/dsh-overleaf