Verdict standards, not opinions

Most advisory documents end the same way: an opinion. "We believe the market is attractive." "In our view, the risks are manageable." The sentence sounds like a conclusion, but it carries none of the machinery a decision needs: no statement of how strong the evidence behind it is, no boundary on where it holds, no condition under which its author would admit it wrong.

A verdict is a different product from an opinion, and boards should procure verdicts. A verdict, as we practise it, is a decision-ready judgment with three printed components: an explicit evidence grade, so the reader knows whether the sentence rests on regulation text, peer-reviewed measurement or the firm's own interpretation; explicit bounds, naming the geography, period and entity class within which the judgment holds; and a falsification condition, naming the observable event that would overturn it, with a review date. An opinion asks to be trusted. A verdict asks to be checked.

The claim is narrow. Opinions are not worthless, and some questions are too open for grading. We argue only that the professions whose judgments carry the highest stakes have all stopped shipping bare opinions, and that their common architecture is available to advisory work.

Medicine moved first. Faced with guideline panels rating the same evidence in incompatible ways, the GRADE working group published a common system that rates the quality of evidence separately from the strength of the recommendation built on it. A strong recommendation on weak evidence becomes visible as exactly that, and auditable by someone who was not in the room.

The intelligence community made the same move by directive. ICD 203 fixes a seven-term likelihood vocabulary mapped to percentage bands and separates confidence from likelihood. It exists because prose was failing its readers: the same word meant different probabilities to different analysts.

Audit went furthest: it closed the vocabulary. Departures from the clean opinion are confined to three modifications, each triggered by defined conditions of nature and pervasiveness. An auditor cannot soften an adverse opinion into warmer prose.

Three professions, three instruments, one architecture. Each fixes a public vocabulary so a judgment cannot hide in prose, separates the judgment from the evidence behind it, grades that evidence, and states in advance what would change the verdict. What follows is our position, graded as such: advisory work that ends in a bare opinion is withholding precisely the machinery its highest-stakes peers were forced to build.

Evidence cards

The gap between opinion and verdict is widest where the board is far from the ground. A European holding board deciding capital, covenants or an exit for an operating company in Abidjan or Dakar cannot check an adviser's feel against its own. "The regulatory environment is stabilising" reads identically whether it rests on the regulator's published calendar or on two airport conversations, and in the corridor the board usually cannot tell which. That distance is where the standard earns its price; it is also its bound. The SEG scale invoked throughout this note is the firm's own governed practice, not a product we sell, and its grades are only as good as the assurance process that audits them.

Sources and limitations

Sources and limitations. The institutional facts in this note rest on the text of ICD 203 and of ISA 700 and 705 (Revised), and on the GRADE framework as published in the BMJ (2008) and maintained by the GRADE working group, each cited with its caveat. The step from those anchors to the verdict standard is the firm's position, graded SEG-2: the three professions cited grade clinical evidence, estimative probability and financial-statement compliance, not forward-looking advisory judgment, and we cite no dataset showing that boards decide better on verdicts than on opinions, because we found none meeting our standard. The SEG scale referenced here is the firm's own governed practice: its grades appear throughout this note, they are audited by our assurance process, and they carry the same self-review limits that process declares. The boundary is explicit: the position covers decision advice; exploratory work may end ungraded, and nothing here claims that any adviser who ships opinions is negligent. This note is not medical, intelligence or audit guidance, and the cited standards are invoked as design precedent only.

1. General counsel: write the verdict standard into the engagement letter before the next adviser is selected. Require every recommendation to carry an evidence grade, bounds and a falsification condition, in the adviser's own vocabulary if it has one. The cited professions show the request is reasonable.

2. The committee secretary: return any paper whose recommendations carry no printed grade. Conviction is not a grade. If every conclusion arrives with full confidence and none with a source class, the pack contains opinions.

3. The sponsoring executive: record bounds and a review date against each live recommendation in the decision log. Geography, period and entity class where it holds; the observable event that would retract it; the date of recheck. A recommendation that holds everywhere has been tested nowhere.

4. The chair: ask once whether "likely", "attractive" and "manageable" are defined anywhere in the pack. ICD 203 exists because they were not. Where no definition exists, the reader is grading evidence by tone: commission the definitions or strike the words.