Caveated Decision Memo¶

Decision artifact — instantiates Uncertainty Explicitness

A recommendation written so its limits travel with it — the call up front, then an explicit separation of what is known, assumed, estimated, and unknown, plus the conditions that would change the answer — so a decision-maker reads the judgment and its uncertainty in the same breath.

A Caveated Decision Memo is the point-of-decision artifact of this archetype: the place a recommendation and its uncertainty are welded together so they cannot be separated on the way to whoever decides. Its distinguishing move is inseparability by structure. Rather than stating a confident conclusion and hoping the reader remembers the caveats, it renders the call and then sorts its own basis into distinct registers — this is known, this is assumed, this is estimated within a range, this is not yet known — and states the conditions under which the recommendation would change. It is not a place where new uncertainty is measured; it is where uncertainty already characterized by its siblings is composed into a single argument a decision-maker can act on without being misled.

Example¶

A city transport office must decide whether to commit funds to a new bus corridor this budget cycle. The analysts could hand up a clean "recommend: proceed." Instead they write a caveated memo. It opens with the call and its confidence. Then it separates the basis: known — current ridership and the construction bid; assumed — that a parallel road closure holds for the full build (owned, medium support); estimated — projected ridership uplift, given as a range, not a point; unknown — whether an anticipated zoning change nearby will proceed, which could swing demand either way. Finally: "reconsider if the zoning decision is deferred past Q3 or the road-closure permit is denied." The committee approves — but approves the actual bet, ranges and open questions included, and knows in advance the two events that should reopen the decision.

How it works¶

The memo attaches uncertainty to one specific object — the recommendation itself — and then does three things a normal brief does not. It leads with the judgment and a stated confidence (so busy readers get the bottom line without losing the qualifier). It segregates the basis into fact / assumption / estimate / unknown so no estimate can pose as a fact.^[1] And it names the triggers for reconsideration — the conditions that would flip or pause the call. It is a composition layer: ranges come from forecast and interval mechanisms, premises from the assumption register, open items from the known-unknowns log; the memo's job is to assemble them into a decision without laundering the uncertainty out.

Tuning parameters¶

Confidence placement — confidence stated once up front versus attached to each individual claim. Per-claim confidence is more honest but heavier to read.
Fact/assumption/estimate granularity — how finely the basis is segregated. Finer separation resists false certainty but can overwhelm a reader who wants the call.
Reconsideration triggers — how many conditions are named and how measurable they are; too many and the memo reads as hedging, too few and it silently commits past its own validity.
Audience register — technical (intervals, probabilities) versus plain-language translation for a non-specialist decision-maker.
Dissent handling — whether minority views and their confidence are carried or collapsed into the single recommendation.

When it helps, and when it misleads¶

Its strength is that the caveats cannot be lost in transit: the structure forces the recommender to show which parts of the argument are solid and which are bets, and it gives the decision-maker the two or three conditions that should reopen the question later. Its failure modes are distinctive. A memo can hedge itself into uselessness — so caveated that no actual recommendation survives, pushing the real decision back onto the reader. It can also do the opposite and be run backwards: a confident conclusion written first, with a thin caveat section bolted on for cover. And translation for a lay audience can quietly round "estimated, wide range" into "will." The discipline that keeps it honest is that the memo must still make a call — an explicit recommendation with its confidence — and that its reconsideration triggers must be real conditions someone will actually watch, not throat-clearing.

How it implements the components¶

claim_or_estimate_object — the memo binds all its uncertainty to one concrete object, the recommendation, so the qualifiers cannot detach from the thing they qualify.
decision_threshold_link — the stated confidence-to-proceed and the reconsideration triggers turn the memo's uncertainty into explicit action logic: proceed now, proceed-but-watch, or hold for evidence.
stakeholder_uncertainty_translation — it renders technical uncertainty (ranges, assumptions, open questions) into the register a decision-maker can act on without a statistics background.

It does not itself produce the ranges, premises, or open items it composes — those come from Forecast Range, Assumption Register, and Known Unknowns Log — and it does not grade the underlying evidence, which is Evidence Grade Rubric.

Instantiates: Uncertainty Explicitness — the memo is where characterized uncertainty is carried intact into the moment of decision.
Consumes: Assumption Register (premises), Forecast Range and Confidence Interval (ranges), Known Unknowns Log (open items).
Sibling mechanisms: Assumption Register · Known Unknowns Log · Risk Register · Forecast Range · Confidence Interval · Confidence Label · Error Bar · Evidence Grade Rubric · Model Limitations Card · Probability Estimate · Uncertainty Band

Notes¶

The memo is a composition, not a source: its integrity depends entirely on the honesty of the inputs it assembles. A well-structured memo built on inflated confidence intervals and rubber-stamped assumptions launders bad uncertainty into a professional-looking recommendation — the tidy fact/assumption/estimate structure can make a weak basis look rigorous. Trust the memo only as far as the artifacts feeding it.

References¶

[1] Separating a written judgment into what is established fact, what is assumed, and what is assessed with stated confidence is a core convention of analytic tradecraft — the discipline exists precisely because prose blends the three by default, letting an assessment read as a fact. ↩