Answer check 04 · judged by a model
Filters behave as assumed
The question it answers, word for wordDo the query’s filters behave as the answer assumes? Pay particular attention to equality tests on columns that may be NULL, since in SQL a NULL fails both = and != and silently drops rows.
- Level 2MetThe filters behave as the answer assumes.
- Level 1Worth a lookA filter is questionable and should be checked.
- Level 0UnmetA filter silently excludes rows the answer assumes are included.
One question, three outcomes
What each level looks like.
The check returns a weight for each level. The level with the most weight decides the outcome; the spread shows how sure it was. Every outcome below the top carries what it means for the number and one next step.
Asked
How much was paid on claims that were not adjusted in June 2026?
Answered
$4.8M was paid on unadjusted claims in June 2026.
Query that ran 1 row
select sum(paid_amt) from claims.claim_line where adjusted_flag = 0 and paid_date between '2026-06-01' and '2026-06-30'
Filters behave as assumedUnmet
- 5%The filters behave as the answer assumes.
- 18%A filter is questionable and should be checked.
- 77%A filter silently excludes rows the answer assumes are included.
Concentration 0.66How sure the check was, shown beside the spread. Never a gate.
adjusted_flag can be NULL, and NULL fails = 0, so those rows are silently dropped.
What it meansThe filter on adjusted_flag excludes rows. The number is probably understated.
Next stepFix the filter and re-run. Do not use this figure as it stands.
Demo data
What it catches
The defects it flags.
- A nullable flag filtered with an equals sign
- A denominator that counts only members with claimsAlso caught by another check
What it cannot see
It reads the shape of the filters, never a catalog of known traps. If a filter is correct SQL but the wrong business rule, such as a status code your team retired last year, only approved knowledge about that column can say so.
How a finding is addedHow it is judged
Narrow questions, named answers.
- 01
Four questions, one call
The four judged checks are asked together against the same answer, the same question, and the same queries. They run in parallel and cannot see each other’s answers.
- 02
A level, not a score
Each check returns a weight for each of its three named levels. The top level is met, the bottom is unmet, and the middle is worth a look: a note, never a failure.
- 03
How sure, shown beside the spread
Concentration says how much of the weight sits on one level. It is shown, never used as a gate, and it is not the probability the answer is right.
- 04
Never the rows
The checks read the question, the answer, and the SQL. Result rows never leave for judgment.
Start with a number your team has to defend.
Bring that number, and see what the checks find in it.