Does FutureSmart Database Agent keep earlier conditions when a user narrows a request?
FutureSmart Database Agent passed this narrowing scenario: it kept the earlier 90-day window and added the Texas filter, and its answer reported 7 orders shipped to Texas. The first answer used a 10-row limit without ordering, and the follow-up table used placeholder column labels.
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.
Narrow the same question to one regionPassEvidence
Input 1: “Show me all orders from the last 90 days, as of September 2, 2026.”
Input 2: “Only the ones shipped to Texas.”
Response 1 replies to Input 1.
Response 2 replies to Input 2.
Response to input 1 Orders Placed in the Last 90 Days (as of September 2 2026) The query returned the most recent 10 orders that were placed between June 4 2026 and September 2 2026. Response to input 2 Orders Shipped to Texas in the Last 90 Days (as of September 2 2026) The query returned 7 orders placed between June 4 2026 and September 2 2026 that were shipped to addresses in Texas (state TX).
Transcribed from Proof 1 · Screenshot
Transcribed from Proof 2 · Screenshot
| ✓ | Expected: Return the subset of turn 1's answer with the other conditions still applied.Found: The follow-up kept the same 90-day window, added only Texas, and reported 7 orders in that window. |
The follow-up keeps the earlier window and adds only Texas, so it narrows the first answer instead of replacing it. Its SQL and answer show 7 Texas orders in that same window.
Ten-row claim. The first answer called the rows the most recent 10, but its SQL had no ordering and only limited the result to 10 rows.
Placeholder labels. The second answer's table and column chooser showed placeholder labels, while the first answer's table showed real field names.
The turn-1 screenshot shows only five of the ten rows, so the hidden rows cannot be checked from the capture.
The turn-2 screenshot shows only five of the seven Texas orders, so the other two cannot be checked from the captures.
The typed message is not shown; the screenshot only includes the product's restated question.
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 | Conversational Interaction → | |
| Scenario | The user narrows what they just asked → | |
| 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.