Does Dot generalize a learned question-query pair to a rephrasing?
Dot used the supplied question-and-query pair for both the original question and a rephrasing, returning the verified result of 42 successful orders for each. Before training, it asked for clarification and gave no count.
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.
Verified question/query pair generalisesPassEvidence
A new chat was opened before the training note was added.
Input 1 · Same account, before training, 15:51 IST: “As of September 2, 2026, how many successful orders did we have in the last 90 days?”
A new chat was opened after a note titled "Verified query: successful orders, last 90 days as of September 2, 2026" was saved in Model > Notes with no answer value and this text: "Verified question and SQL pair (trusted query). Question: As of September 2, 2026, how many successful orders did we have in the last 90 days? Answer (SQL): -- "Successful orders" = orders with status delivered, placed in the 90 days ending 2 Sep 2026 SELECT COUNT(*) AS successful_orders FROM orders WHERE status = 'delivered' AND placed_at >= '2026-06-05' AND placed_at < '2026-09-03';"
Input 2 · Same account, supplied wording, 15:52 IST: “As of September 2, 2026, how many successful orders did we have in the last 90 days?”
A new chat was opened for the rephrased question.
Input 3 · Same account, rephrasing, 15:53 IST: “As of September 2, 2026, how many orders were successfully delivered in the last 90 days?”
Before-training comparison: clarification question
Should I count “successful” as delivered/completed orders, or all non-cancelled orders—and use the customer support or order tracking database?
Answer to the supplied wording
We had 42 successful orders from June 5 through September 2, 2026, inclusive. Successful means orders with status “delivered,” counted by placement date.
Answer to the rephrasing
42 orders placed from June 5 through September 2, 2026 had delivered status. This uses placement date, per the verified definition.
Transcribed from Proof 1 · Screen recording at 0:25
Everything the tool returned, in order (5)
- Before-training comparison: clarification question
Printed above.
- Answer to the supplied wording
Printed above.
- Query behind the supplied-wording answer · not part of the answer
SELECT COUNT(*) AS successful_orders FROM customer_support_db.public.orders WHERE status = 'delivered' AND placed_at >= '2026-06-05' AND placed_at < '2026-09-03'
- Answer to the rephrasing
The final response, printed above.
- Query behind the rephrased answer · not part of the answer
SELECT COUNT(*) AS successful_orders FROM customer_support_db.public.orders WHERE status = 'delivered' AND placed_at >= '2026-06-05' AND placed_at < '2026-09-03'
| ✓ | Expected: The supplied question and a rephrasing should both return the verified query’s result: 42 successful orders.Found: The verified query counted 42 delivered orders placed June 5 through September 2, 2026 in the live database. Both post-training answers returned 42; Dot’s queries differed only in including the table’s full database location. |
The test required more than repeating an answer to identical words. Dot executed the trusted query for a genuine rephrasing as well as the supplied question, demonstrating generalisation rather than an exact-question lookup.
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 user supplies example questions and the queries they trust → | |
| Tool | Dot → |
Global scenario definition → · Global capability definition → · Dot 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.