Accuracy, Evidence, Tone, Completeness, and Usefulness
Unit ID: PAW-M08-U01 Estimated active time: 20-25 minutes
Five separate questions
"Is this good?" cannot be answered consistently. Five narrower questions can.
- Accuracy — is what it says true?
- Evidence — can each claim be traced to something?
- Tone — is it right for this reader?
- Completeness — is anything important missing?
- Usefulness — can the reader act on it?
They fail independently
An output can be accurate and useless — every fact correct, no action possible. It can be complete and wrong in tone. It can be useful and unevidenced.
Checking them separately is what stops a strong performance on one hiding a failure on another.
Completeness is the hardest
The other four ask about what is present. Completeness asks about what is absent, which is much harder to notice.
Two techniques help: compare against your original success criteria, and ask explicitly what has been left out.
> What did you leave out that a careful reader would expect to see?
The five-question pass
FIVE-QUESTION REVIEW
Accuracy Every factual claim checked? pass / fail / n-a
Anything I did not supply? list it
Evidence Each claim traceable to a source? pass / fail
Unevidenced claims: list them
Tone Right for the reader described? pass / fail
Completeness Against my success criteria? what is missing
What was left out deliberately?
Usefulness Can the reader act on this today? pass / fail
If not, what is missing?
Overall: use as is / edit / verify first / ask again / rejectDo the cheap checks first
Accuracy and evidence are the slow ones. Tone and usefulness take seconds.
If an output fails usefulness, there is no point tracing its citations. Order the pass so the quick disqualifiers come first.
Mini practice
Run the five-question pass on one output you have already used.
Note which of the five you had not actually checked at the time.
That is the one your normal review is missing.