bmc-marshall/choosewell

Options appraisal with Claude: shortlist, score and test options for organisational decisions, and produce recommendations that stand up to scrutiny.
Choosewell guides a person or a working group through a structured choice. It covers what's in scope, where to work, how to organise (overall or case by case), who to work with, and smaller design choices such as team set-up, funding or hosting. It recommends a method for the situation, explains each methodological choice in plain language, keeps the group's judgement in charge, checks its own work, and produces the analysis and storyline behind a recommendation.
It was built from real options appraisals (operating models, partner screening, geographic prioritisation, contracting models) and from decision-analysis practice, generalised and anonymised.
## Methods included
Screening: threshold filters, knock-outs, RAG (red/amber/green) ratings against critical success factors
Scoring: scored criteria with written score definitions, swing weighting, pros and cons with option labels, impact × availability and Eisenhower matrices
Matching: decision trees that pick the right model for each situation, with red lines and push/pull factors
Testing: sensitivity analysis, switching values, dominance, tornado charts, random-weight tests, top-k stability, value-for-money frontier
Uncertainty: quantified decision trees and expected value, risk profiles, value of information (EVPI, EVSI), Bayesian updating, option value, real options, risk tolerance and certainty equivalents, scenarios and minimax regret
Judgement and bias: private scoring before discussion, claims log, pre-mortems, adverse-consequence analysis, decision-quality check, reference-class forecasting
Rulebooks: compatible with the UK Green Book (five-case model, critical success factors, MCDA guidance) and US federal approaches (OMB Circular A-94, GAO analysis of alternatives), plus a general method for everyone else
> Not legal, financial or regulatory advice. Choosewell identifies where advice is needed and drafts the questions to ask.
## What it does
It works through three questions and ten steps:
Tree 1: What are you deciding? Scope → what → where → structure (overall or by situation) → who → design choices. Start at the top; stop at the first question you can't yet answer. Single go/no-go, timing or how-much decisions (launch, pilot, settle, bid, spend) have their own route.
Tree 2: What rules, and how much rigour? UK central government → the Green Book; US federal → OMB and GAO guidance; everyone else → a general method drawing on both. Reversibility and stakes set the depth (quick, standard or full).
Tree 3: How will you assess the options? Threshold filter, scored criteria with swing weights, red/amber/green against success factors, pros and cons, impact × availability or Eisenhower matrices, a decision tree for matching options to situations, and a quantified decision tree when a few uncertain events or staged choices drive the result.
Steps 1–10: frame, rules and rigour, long list, choose a method, criteria, evidence, assess, test, decide, check and present. Each step says when you can skip it, and carries its own safeguards: a claims log for every fact, drafted content flagged until the group agrees it, independent scoring before discussion, sensitivity and break-even tests, a pre-mortem, an audit, and independent review for big decisions.