Can FutureSmart Database Agent generalise from trusted example questions and queries?
FutureSmart Database Agent did not generalise the trusted example pair. The exact wording returned 42, while the reworded question returned 41 from a new delivery-date query; before training, the same question was refused.
1 of 1 test case failed
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 generalisesFailEvidence
Input 1: “As of September 2, 2026, how many successful orders did we have in the last 90 days?”
Input 2: “As of September 2, 2026, how many successful orders did we have in the last 90 days?”
Input 3: “As of September 2, 2026, how many orders were successfully delivered in the last 90 days?”
Response 1 answers Input 1.
Response 2 is the query the tool generated for Input 2; it is not part of the answer.
Response 3 answers Input 2.
Response 4 is the query the tool generated for Input 3; it is not part of the answer.
Response 5, the final response, answers Input 3.
41 orders were successfully delivered in the last 90 days, as of September 2, 2026.
Transcribed from Proof 4 · Screenshot
Every message the tool sent, in order (5)
- Response to input 1
I can't determine this from the available tables: they don't contain data that supports the requested information. This does not establish a count of zero.
- Generated query for input 2 · not part of the answer
SELECT COUNT(*) AS successful_orders FROM orders WHERE status = 'delivered' AND placed_at >= '2026-06-05' AND placed_at < '2026-09-03';
- Response to input 2 (message 2 of 2)
We had 42 successful orders in the 90-day period ending September 2, 2026.
- Generated query for input 3 · not part of the answer
SELECT COUNT(*) AS successfully_delivered_orders FROM orders WHERE status = 'delivered' AND delivered_at >= TIMESTAMPTZ '2026-06-05 00:00:00+00' AND delivered_at < TIMESTAMPTZ '2026-09-03 00:00:00+00' LIMIT 10
- Final response
41 orders were successfully delivered in the last 90 days, as of September 2, 2026.
| ✗ | Expected: Both wording variants should return the verified result.Found: The exact wording returned 42, but the rewording returned 41 from a new delivery-date query. |
Both wording variants had to return the verified result. The taught wording gave 42 from the saved query, but the rewording gave 41 from a new delivery-date query, so only the exact wording matched.
Limit unchanged. The reworded query added a limit after a count, which did not change the returned count.
Baseline refusal. Before training, the agent refused to answer and said the refusal did not prove zero.
Control check. A separate new chat produced SQL and returned 308, showing the agent could query the database.
The trusted question and query were entered in settings, not in chat.
The model and plan were not shown, so the result is limited to the surface and workspace tested.
The saved SQL is only partly visible.
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 | FutureSmart Database Agent → |
Global scenario definition → · Global capability definition → · FutureSmart Database Agent 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.