Four things people responsible for decisions under uncertainty should be able to do.

Make a register entry someone would read unprompted.

Do this: Find the entries in your register nobody has read since the last required review, and say for each what would have to change for someone to read it.

Classify a statement correctly and know which pot of money pays.

Do this: Reclassify ten register entries as risk, issue or assumption, and name the funding consequence of each change.

Produce specific, ownable risk statements the team would not otherwise have written.

Do this: Run a pre-mortem and convert the output into register entries that each name a cause, an uncertain event, a consequence and an owner.

Rewrite register entries until they can be modeled at all.

Do this: Take your register through the worksheet and produce entries with a quantified range, a probability and a mapped cost or schedule effect.

Replace a word with a number and state what the number means.

Do this: Have each member of a risk review privately assign a percentage to the words your register uses, then compare the spread.

Produce and defend a quantified estimate under sparse evidence.

Do this: Take an exposure your organization has left qualitative because there is no frequency history, and produce a defended numerical estimate with the reasoning that supports it.

Defend a range to a body that has to decide on it.

Do this: Prepare the four answers a governing body will ask of a range, and the one sentence that says what you are asking them to decide.

Read a percentile without over-reading it.

Do this: State, for a percentile your program already publishes, what it promises and what it does not.

Know when a matrix genuinely misleads, and what to do when the format cannot be abandoned.

Do this: Test your matrix against the conditions under which ordinal ranking inverts the quantitative ordering, and record which of your risks sit in that region.

Decide whether a schedule can carry a simulation before running one.

Do this: Check a schedule against the preconditions and produce a written go or no-go with the specific defects that would be inherited.

Choose between two bias corrections that are weak exactly where the other is strong.

Do this: Build a reference class for one current commitment and compare its distribution with the inside-view estimate already on record.

Tell merge bias from a modeling defect when both look identical on the report.

Do this: Take a simulated date that disagrees with the deterministic critical path and produce the diagnosis, with the evidence that distinguishes the two causes.

Trigger re-analysis by conditions rather than by the calendar.

Do this: Write the conditions that will trigger a re-run on your program, and retire whichever calendar cadence you are currently defending.

Settle who is accountable when three groups each name one of the others.

Do this: Map your three lines and name, for one significant risk, who is accountable for it and who owns the judgment inside the number.

Diagnose an owner who has been named but cannot decide.

Do this: For each named owner, write down the decision they can take alone. Where the answer is none, the risk is unmanaged.

Tell a supply constraint from an exposure constraint before funding a fix for the wrong one.

Do this: Take the approval your program waits on longest and write what the approver carries if they sign and it goes wrong. If nothing is written there, every fix you are funding is aimed at supply.

Replace an unanswerable executive question with three that can be answered.

Do this: Answer all three replacement questions for your program and show where they disagree with each other.

Set a threshold that can actually be violated.

Do this: State your appetite as a confidence level and identify what would constitute a breach of it.

Close the gap between a published appetite and how the organization behaves.

Do this: Find one decision taken in the last quarter that was outside the published appetite, and establish why nothing stopped it.

Choose a response knowing that transfer moves the invoice rather than the event.

Do this: Take three treated risks and state what the treatment actually changed, separating the event from who pays for it.

Sequence work by what most changes the decisions still ahead of you.

Do this: Take the top drivers from a sensitivity ranking, filter them through the resolvable and decision-changing tests, and re-sequence the next two cycles.

Establish who holds release authority before the argument starts.

Do this: Write down which definition of contingency and management reserve your agreement uses, and who is named to release each.

Release contingency as uncertainty resolves, rather than holding it by default.

Do this: Track drawdown against retired risks in the tracker and produce the release recommendation the arithmetic supports.

Recognize what a staffing decision gave away, and recover it.

Do this: Name what the risk role on your program requires that the person holding it was never given, and decide which half you are changing.

Stop risk identification rotating away from the experience it requires.

Do this: Establish how long your risk role has been held by its current occupant, and what identification capability leaves when it rotates.

Surface a known but unwritten risk without requiring anyone to accept blame.

Do this: Identify one risk everyone knows and nobody has written down, and design the route by which it could be raised.

Build the board-level system the standard actually asks for.

Do this: Produce the record that would show a board received risk information and used it.

Reconcile what the organization says about its governance against the evidence of how it decided.

Do this: Take one thing the declared record says the board oversees and assemble twelve months of operating evidence for it, including what never arrived.

Report movement rather than position, in three formats rather than one.

Do this: Rebuild one risk report so every item shows what changed since the last one and what decision is being asked for.

Produce what a sponsor is actually required to produce, at the required percentile.

Do this: Check your program against the cost and schedule rules separately, since they are not the same rule.

Score the uncertainty the organization already stated against what actually happened.

Do this: Score your published ranges against arriving outcomes, separate a centering fault from a width fault, and name the practice you are changing as a result.