A policy brief should show what could change the decision

Begin with the decision that is actually open

A fictional city faces a drought. Its analysts are asked to recommend a way to reduce non-essential water use while protecting households that cannot easily adapt. Before collecting arguments, they need to establish what the decision maker can authorize, the time available, and which outcomes matter. A briefing about emergency restrictions is different from a long-term investment appraisal, even when both cite the same reservoir report.

The UK Treasury’s 2026 Green Book describes appraisal as assessing the costs, benefits, and risks of options for achieving government objectives. It explicitly does not set those objectives or make policy decisions.[1] That distinction is useful for TACITUS: software can help organize an appraisal, but the mandate and choices remain with accountable people.

Put the alternatives on the same footing

The city might compare voluntary conservation, temporary restrictions, targeted support, and accelerated repairs. It also needs a baseline: what is likely to happen under existing arrangements? Each option should be examined over a stated period using comparable assumptions. An attractive measure can look much less attractive when its implementation delay or effect on a particular group is made explicit.

The Green Book asks practitioners to consider uncertainty, unmonetisable effects, and distributional impacts as well as summary value-for-money metrics.[1] A graph can organize the supporting relationships, but it cannot choose their importance or turn an undocumented estimate into a measurement. Where estimates are unavailable, the brief should say which information would be needed.

A constraint needs a source and an interpretation

“The council cannot do this” may refer to a legal prohibition, an insufficient budget, a staffing shortage, or a political objection. Those are different propositions. A constraint record should state its category, source, jurisdiction, effective period, and who interpreted it. A draft rule is not an enacted rule. A departmental preference is not automatically a binding limit.

ODNI’s analytic standards call for a distinction between underlying information, assumptions, and judgments, and for consideration of alternatives.[2] Applied to this example, the statute is source material; a lawyer’s reading is an attributed interpretation; the analyst’s conclusion about feasibility is a judgment. This approach borrows a useful discipline without claiming that a civilian policy brief has been certified to intelligence-community standards.

Reception is a question to investigate

A stakeholder map can record public positions, institutional responsibilities, and prior statements. It should not claim to know how every affected person will respond. A retailer’s submission may describe costs well while saying little about people who did not participate in the consultation. Silence is missing evidence, not consent.

Wind Tunnel is exploratory research into possible reactions and assumptions. Synthetic responses, if used, should be labeled as scenarios rather than forecasts of public opinion. For the drought example, a useful exercise asks what evidence would support or weaken a concern about unequal burdens. It cannot establish that opposition will occur, how widespread it will be, or which message would persuade a population.

Build for revision, then test the revision

The intended PRAXIS workflow would use a reviewed capsule to keep the question, evidence, options, assumptions, and drafting requirements together. DIALECTICA is the proposed boundary for reviewed knowledge. An analyst could ask for an options memo, inspect its sources, and record why a recommendation changed when a new constraint appeared. The complete integration remains under development.

Evaluation should include a baseline summary workflow and practitioner review. We would test whether the draft cites the right version of a rule, identifies missing alternatives, preserves distributional concerns, and updates the affected recommendation after a source correction. A blinded review could measure omissions and unsupported assertions alongside review time. NIST’s evaluation-probe work is relevant to tracing intermediate agent behavior,[3] but no study cited here establishes TACITUS performance or policy impact.

Options depend on the evidence that could change the decision

Example Advisory compares documented January and July proposals, the industry transition request, and an optional pilot conditional on funding. Preparation recommendations belong to the analyst, not to the source as an enacted obligation. The task packet retains the Day 16 deadline, steering-group audience and missing adoption/funding decisions. Ask which new instrument, response or budget decision would change the reversible preparation recommendation. [4]