# Revise for a person who must decide

Run these passes in order. A smoother sentence cannot rescue a missing argument.
These tests guide judgment; they are not an AI detector or a numerical quality score.

## 1. Reconstruct the argument

Without looking at the headings, state the requested decision, the mechanism by
which it helps, the strongest competing option, and the principal downside. Can a
skeptical reader find each in the draft? If not, fix the reasoning first.

Trace each recommendation to its supporting evidence. Keep the denominator,
measurement period, comparison, and uncertainty when supplied. “Three pilot customers
asked for CSV export” cannot become “customers overwhelmingly demand advanced
analytics.” If the sample size is absent, say it is absent; do not create a percentage.

Ask what observation would change the recommendation. Include contradictory results
and a plausible interpretation instead of relegating them to a footnote. If the
user wants certainty the evidence cannot support, offer a clearer conditional case.

## 2. Make actions and mechanisms visible

For each consequential sentence, identify who acts, what changes, and how that
change affects the outcome. Replace abstract claims with that mechanism. “Unlock
operational excellence” does not tell the reader whether a shared queue removes
coordinator delay or whether specialists lose routing context.

Retain technical terms that carry meaning. In an engineering memo, idempotency,
p95 latency, or backpressure may be more precise than a vague paraphrase. Explain
them for unfamiliar readers; do not purge them to satisfy a jargon quota. Use passive
voice when the actor is unknown or irrelevant, and name an owner where responsibility matters.

## 3. Test every paragraph's job

Write its contribution in the margin: establishes a fact, explains a cause, compares
an option, states a consequence, or asks for a decision. Delete or combine paragraphs
that repeat the conclusion, praise the idea, announce the next section, or offer
background that changes no reader judgment. Keep context that prevents a real misunderstanding.

Check for manufactured symmetry: three benefits with no costs, identical paragraph
rhythms, repeated “not X but Y” constructions, and conclusions that merely restate
headings. Revise the particular failure, not every occurrence of a word or pattern.
Do not replace a precise term with a thesaurus choice to make prose look less generated.

## 4. Read as the intended audience

Read the opening and one dense paragraph aloud. Mark where the reader has to hold
too many ideas before reaching the action, where a pronoun has no clear referent,
or where a transition implies a cause the evidence does not show. Split or reorder
those sentences. Vary length where meaning calls for it, without a fixed sentence limit.

Ask a representative reader: What is being decided? What evidence is weakest?
What would you choose instead, and why? An answer that differs from the author's
preferred decision can still demonstrate a successful memo. Confusion about the
ask or omitted tradeoffs signals revision work.

## 5. Compare before and after

Verify that the revision preserves numbers, scope, uncertainty, and dissent. Check
that every new factual assertion has a source or is labeled as an assumption.
Report substantive edits briefly: “Named the queue owner, replaced the scale promise
with a load-test condition, and retained the routing risk.” Do not claim the text
is definitively human-written or immune to AI detection.
