A Decision Memo Can Finish With Unknowns

Research becomes useful when somebody can make a bounded decision from it. A folder of links is not a decision packet, and a confident recommendation that hides its unknowns is not finished work.

The practical goal is smaller: define the decision, gather only the evidence that changes it, label what is still unknown, and end with one authorized next action.

Here is a fully filled synthetic example.

The Messy Request

Should a two-person online course business translate its weekly newsletter into a second language this quarter?

The request sounds simple, but it silently contains several different questions:

• Is there audience demand?
• Does the business have publication rights for the source material?
• Can a fluent human review every issue?
• How much production time is available?
• What result would justify continuing?

If those questions remain implicit, the researcher can produce a polished answer without producing a usable decision.

1. Bound the Decision Before Research

The deliverable is not “research translation.” It is:

DECISION OWNER
Course business operator

DECISION
Pilot, prepare, defer, or reject a four-week translated-newsletter test

DELIVERABLE
One-page memo with recommendation, evidence table, risks, and next action

IN SCOPE
Audience demand, production effort, review capacity, rights, and measurement

OUT OF SCOPE
Guaranteed revenue forecast, legal advice, or automated publishing

DONE WHEN
The owner can make one of the four decisions without another general research round

This boundary prevents the work from expanding into a translation-platform comparison, a marketing plan, or a speculative revenue model.

2. Route Work by Failure Mode

One role gathers current evidence. A second produces the comparison. A skeptical pass looks specifically for unsupported demand claims, missing review capacity, and false precision. The final pass packages the memo and corrects citations.

The stop conditions matter as much as the assignments:

• stop if audience-language data is unavailable;
• stop if publication rights are unclear;
• do not turn support-message language into subscriber demand;
• do not invent conversion data;
• do not publish or contact anyone without the owner.

The point is not to simulate a large team. It is to make the failure checks independent from the first draft.

3. Separate Facts From the Story You Want to Tell

Assume the synthetic business supplies these facts:

• 18% of recent support questions were written in the target language;
• the newsletter has 2,400 active subscribers;
• no translated edition has been tested;
• one fluent human reviewer is available for four weeks;
• production must stay under two staff hours per issue.

The tempting sentence is: “18% of subscribers want a translated newsletter.”

It is also unsupported. Support-message share and subscriber demand are different measures. The honest handoff says:

VERIFIED
18% of recent support questions used the target language.

UNKNOWN
How many subscribers would opt into a translated edition.

PROHIBITED CLAIM
18% of subscribers want translation.

The unknown is not a defect to hide. It is the reason to design a small test.

4. Make the Recommendation Proportional to the Evidence

The memo recommends a four-week manual pilot to a separate opt-in segment. It does not recommend translating the archive or automating publication.

Why this decision fits the evidence:

1. The support-language signal is enough to justify a test, not a rollout.
2. A fluent reviewer exists, so factual and tonal errors can be checked before release.
3. A separate opt-in segment measures interest without changing the primary newsletter.
4. The two-hour production ceiling creates a real capacity stop rule.

The pilot tracks opt-ins, unique readers, replies, unsubscribes, production minutes, and reviewer corrections. It stops if review cannot be completed, rights are unclear, or production exceeds two hours twice.

After four weeks, the owner chooses:

• Keep if opt-in demand recurs, readership is real, corrections are manageable, and production stays within the ceiling.
• Revise if interest exists but review or production is too costly.
• Stop if demand is weak, factual risk repeats, or owner capacity fails.

The exact thresholds belong to the owner. Research can propose them, but it should not quietly invent business authority.

5. Close With a Completion Receipt

ARTIFACT
Translated-newsletter pilot decision memo

STATUS
Complete for owner review

EVIDENCE USED
The supplied synthetic audience and capacity facts

KNOWN UNKNOWNS
Opt-in demand, reading behavior, and conversion impact

EXTERNAL ACTION TAKEN
None

NEXT AUTHORIZED ACTION
Owner chooses pilot, prepare, defer, or reject

This receipt proves what exists. It does not prove that the pilot will succeed.

The Useful Definition of “Finished”

A decision memo is finished when:

• the decision and owner are explicit;
• every material fact has a traceable source;
• inference and unknowns are labeled;
• the recommendation is no stronger than the evidence;
• contrary evidence and stop conditions are visible;
• the next external action still has a human owner.

The result can be complete while uncertainty remains. In fact, a memo that tells the owner exactly what is unknown—and how to test it safely—is often more useful than one that sounds certain.

The free AI Execution Starter Pack includes a one-workflow brief, human review gate, completion receipt, and a smaller worked example as plain Markdown. No signup or private-data upload is required.

The $19 AI Solo Operator System v1.4 adds this complete research example plus filled content, digital-product, and recurring-operations examples. It contains 32 local Markdown files and has zero verified sales at publication; no result is guaranteed.