Chapter 9 – Running an SRA your review board believes
The Quantitative Schedule Risk Analysis Handbook · Edition 1, August 2026 · by the team behind [Epoch SRA](https://epochsra.com), an Excel add-in for schedule risk analysis · corrections welcome at contact@epochsra.com
The mathematics of the previous chapters fail for a non-mathematical reason more often than any other: the room does not believe the numbers. Belief is not won by more decimal places. It is won by reproducibility, provenance, and a cadence that survives contact with a real program. This chapter is about the operating discipline.
Reproducibility, or it did not happen
An analysis that cannot be re-run is an anecdote. The minimum standard: a fixed random seed, recorded inputs, and a stamped output – run timestamp, iteration count, seed, and input fingerprint on every chart and table that leaves the tool. When a stakeholder asks "is this the current picture?", the stamp answers; when two analyses disagree, the fingerprints say why. Boards have been burned by stale slides too often to extend trust without this, and they are right.
Answer the provenance question before it is asked
Chapter 7's labels earn their keep in the meeting. Present percentiles with their evidence class stated – calibrated versus model estimate – and the predictable challenge ("how do you know these numbers?") has already been answered in the artifact itself. The same applies to limitations: an analysis that states what the import could not translate, which quality checks failed, and which risks were unmeasurable is an analysis that has done the board's due diligence for it. Stated limitations read as competence; discovered ones read as concealment.
Cadence: the analysis is a verb
A one-off SRA is theater with better graphics. The value compounds only on a rhythm: re-import the current plan, re-run, and compare against the previous run at every review cycle. The comparison is the product – which way did P80 move since last month, and which risks moved it? – because a program manages deltas, not snapshots. This is also the quiet test of your toolchain: if re-running next month's plan means rebuilding mappings and re-entering risks, the cadence dies by friction, and the analysis with it. Task identities and register mappings must survive a re-import, or the second month never happens.
The review agenda that works
Four items, in order, before any percentile is shown. One: schedule quality – what the checks found, what was repaired. Two: what changed since last run – plan deltas, register deltas. Three: the distribution – P50, P80, and the gap, trended against previous runs. Four: the tornado – measured contributions, and the mitigation decisions they imply. Ending on decisions rather than dates is what separates an analysis the board uses from one it receives.
--- In practice: put a recurring thirty-minute slot in the calendar the day before each program review, titled "re-run and diff". The habit, not the tool, is what makes the numbers believed by month three – because by then the board has watched the distribution respond to reality twice.