Select and frame the analysis

Start with the decision the analysis must support, not with a convenient distribution. Name the project or bounded program, accountable owner, stakeholders, analysis date and horizon. Choose cost, schedule, or integrated mode. Record one currency for cost and one declared duration unit for schedule. State what physical scope, contract packages, phases, escalation dates, and schedule boundaries are included and excluded.

Write the decision question so a reviewer can identify the action and boundary: for example, “What cost contingency and schedule reserve should the steering committee consider for the defined reference design at the September gate?” Also record the human-entered budget, deadline, existing contingency, and existing reserve only when they are genuine comparison boundaries. They do not change simulated outcomes.

Selection test

Use this method only when numerical ranges have a defensible basis. If experts can discuss only severity, triggers, and control quality, begin with qualitative stress testing. If choices occur after observations, use a decision tree. If the team faces structural futures for which probabilities are not credible, use scenario planning. An integrated model may analyze cost and duration together, but separate cost-P80 and schedule-P80 values do not imply 80 percent joint confidence.

Framing review

A strong frame identifies the estimate maturity, reference date, price basis, schedule logic status, decision authority, and intended use. Record assumptions and exclusions explicitly. Do not treat a default currency, unit, percentile, or iteration count as a project fact. Continue to baselines and distributions. ## Interpretation discipline

Keep inputs, derived results, and human decisions visibly separate. Every material number needs linked evidence or an explicit assumption flag and basis. Unknown is not zero. A percentile is conditional on this model, scenario, seed, engine version, and source version; it is neither a guarantee nor an approval. Any material model edit makes an earlier result stale.