Answer checks 05 · run in code
The work behind the answer
Before any judgment, five checks read the machinery: did a query run, did rows come back, and did the knowledge your team approved reach the answer. No model is involved. They are lookups, and they land first.
- Queried the estateDid any query run against your systems?
- Returned rowsDid any query return rows? Zero rows is reported as a fact, not a failure.
- Relevant knowledge surfacedDid approved knowledge for these tables reach the answer?
- Surfaced knowledge usedDid the answer use the knowledge that reached it?
- No unsurfaced pitfallsIs there an approved pitfall on these tables that the answer did not mention?
In order
The fast result shows first. No spinner.
The code checks come back in milliseconds and show right away. The judged checks follow in seconds and fill the result in. A check that cannot run says so, in its own state.
Answer checksChecking
0 of 9 run
- Queried the estate…
- Returned rows…
- Relevant knowledge surfaced…
- Surfaced knowledge used…
- No unsurfaced pitfalls…
- Data supports the answer…
- Right population and period…
- Right source…
- Filters behave as assumed…
Demo data
Four words
The result, in words, not a number.
An earlier version showed one overall score. It could not be read: a check that passed at 0.55 and one that failed at 0.45 both produced 0.55. The result is now one of four words, with counts beside it.
- No problems foundNothing unmet, nothing flagged. Never “passed”, and never “correct”.
- Worth a lookOnly middle-level notes.
- Problems foundAt least one check unmet.
- Not checkedNothing could be evaluated. It never reads as a pass.
Where knowledge comes in
A pitfall your team approved, checked in code.
Some defects never show in SQL: a view that under-counts members, a table that carries two feeds. No model catches them from the query alone. When your team approves a pitfall and it attaches to its table, an answer that reads the table without it can be flagged in code, the same way every time.
That is a knowledge problem, not a model problem, and the knowledge graph is where it gets solved.
Approved pitfall · enrollment.v_member_roster
Under-counts members: termed-then-reinstated spans drop out.
Answer read
enrollment.v_member_roster, without the pitfall
No unsurfaced pitfallsUnmet
Demo data
Start with a number your team has to defend.
Bring that number, and see what the checks find in it.