Does AskYourDatabase keep a corrected definition when the question is reworded?
AskYourDatabase kept the corrected definition on both the repeated and reworded questions. Each later answer gave the reference revenue total of $23,322.52, compared with $23,597.02 before training.
1 of 1 test case passed
Every test case in this scenario has a result.
The test case
Each test case is judged on its own: Pass, Fail, Not gradable, or Untested. The scenario result above is based on this test case.
Corrected metric definition survives a rephrasingPassEvidence
A new chat was opened before training.
Input 1 · Before the rule was saved: “As of September 2, 2026, what was our total revenue over the last 90 days?”
Before each of Inputs 2–3: The rule “Our revenue definition excludes cancelled orders and refunded orders.” was saved in the Documentation box, and a new chat was opened.
Input 2 · After the rule was saved: repeated question: “As of September 2, 2026, what was our total revenue over the last 90 days?”
Input 3 · After the rule was saved: rephrased question: “As of September 2, 2026, how much did we earn in total over the last 90 days?”
Answer before training
As of September 2, 2026, your total revenue over the last 90 days (June 4 – September 2, 2026) was $23,597.02, across 43 orders (excluding cancelled orders).
Answer to the repeated question
As of September 2, 2026, our total revenue over the last 90 days was $23,322.52. This figure excludes cancelled orders and refunded orders, per our revenue definition.
Answer to the rephrased question
As of September 2, 2026, we earned a total of $23,322.52 over the last 90 days. This excludes cancelled and refunded orders, per our revenue definition.
Transcribed from Proof 3 · Screenshot
Every message the tool sent, in order (8)
- Progress before training
I'll calculate the total revenue over the 90 days leading up to September 2, 2026.
- SQL before training · not part of the answer
SELECT SUM(total_amount) AS total_revenue, COUNT(*) AS order_count FROM orders WHERE placed_at >= '2026-09-02'::timestamp - INTERVAL '90 days' AND placed_at < '2026-09-02'::timestamp AND status <> 'cancelled'
- Answer before training
Printed above.
- Progress after the rule was saved
I'll calculate the total revenue over the 90 days before September 2, 2026, excluding cancelled and refunded orders per your revenue definition.
- SQL for the repeated question · not part of the answer
SELECT SUM(total_amount) AS total_revenue FROM orders WHERE placed_at >= '2026-09-02'::timestamp - INTERVAL '90 days' AND placed_at < '2026-09-02'::timestamp AND status NOT IN ('cancelled') AND refunded_at IS NULL; - Answer to the repeated question
Printed above.
- SQL for the rephrased question · not part of the answer
SELECT SUM(total_amount) AS total_revenue FROM orders WHERE placed_at >= '2026-09-02'::timestamp - INTERVAL '90 days' AND placed_at < '2026-09-02'::timestamp AND status NOT IN ('cancelled') AND refunded_at IS NULL; - Answer to the rephrased question
The final response, printed above.
| ✓ | Expected: Both the repeated and reworded questions should exclude cancelled and refunded orders and return the reference total of $23,322.52.Found: Both returned $23,322.52, equal to the $23,322.52 reference total; their generated queries excluded cancelled and refunded orders. |
The correction needed to survive both a repeat and a rewording. Both questions were asked in new chats without restating the rule, and the generated queries—not just the answers’ wording—used the corrected exclusions.
Definitions stated in answers. The answer before training described excluding cancelled orders; the repeated and reworded answers also named refunded orders as excluded.
Period wording discrepancy. The answer before training included September 2, 2026 in its stated period, although the query excluded that day. No orders fell on the boundary days, so the total was unaffected.
Configuration and setup
How this tool was set up for the run and what the test needed in place. Each row is a fact from the run's records; a fact the records do not hold is left out, not guessed.
How this scenario is graded
How we decide Pass, Fail and Not gradable. The same rules apply to every tool tested on this scenario.
Each test case gets one result per tool: Pass, Fail or Not gradable. A test case we haven't run yet shows Untested. There are no partial results.
- Pass: the tool did everything the test expected, and nothing it said contradicts the correct answer.
- Fail: at least one expected behaviour clearly didn't happen; the test case's result says which one and quotes the tool.
- Not gradable: our evidence couldn't settle the outcome (for example a record we needed is missing). It is never counted as a fail, and the test case's result says what's missing.
Where this sits in the benchmark
This page is one cell of a larger study: one tool, one scenario. Only this benchmark's frame appears here.
| Level | Name | Scope |
|---|---|---|
| Benchmark | AI Database Agents → | 28 scenarios · 11 tools |
| Capability | Training → | |
| Scenario | The agent used the wrong definition and the user corrects it → | |
| Tool | AskYourDatabase → |
Global scenario definition → · Global capability definition → · AskYourDatabase product page →
History of this result
What has happened to this result since it was first published. Runs and grades are never overwritten: a retest or a re-grade publishes a new result and keeps the earlier one readable.
Act on this result
Nothing filed here edits the run or the grade. A challenge opens a review, and a review can produce a new run or a re-grade — which becomes the current result and leaves this one in the history.
You run the same kind of test against your own setup and get the same behaviour.
Agree →Yours behaves differently. Tell us what you got, with a screenshot if you have one.
Disagree →Something here is wrong — a reference value, a transcription, a grade.
Report an issue →Tell us what changed and we schedule a rerun of the failing test case. The old result stays as history.
Vendor notice →The same record is available as structured data through the AI Demos MCP server, with the counts, the coverage and every per-test-case reason carried as fields.
Verify the proof files
These files support this result. Open a file to inspect the original evidence.
File fingerprints (SHA-256)
A fingerprint identifies the exact file used for this result.