Can FutureSmart Database Agent drop prior filters when the user changes direction mid-conversation?
When the user changed direction, FutureSmart Database Agent dropped the earlier filter but did not give the unfiltered result the scenario calls for. In the run, it returned a 16-row by-state breakdown instead of the single overall cancellation rate, so the answer did not match the unfiltered result.
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 counts this row.
Drop the filter, go all segmentsFailEvidence
Input 1: “What's our cancellation rate for orders shipped to Texas?”
Input 2: “What's our cancellation rate across all states?”
| state | cancellation_rate |
|---|---|
| AZ | 0.125 |
| CA | 0 |
| CO | 0 |
| GA | 0 |
| IL | 0 |
| MI | 0 |
| MN | 0 |
| NC | 0.2 |
| NV | 0 |
| OH | 0 |
| OR | 0 |
| PA | 0 |
| RI | 0 |
| TN | 0 |
| TX | 0 |
| WA | 0 |
Copied from Proof 1 · Output file (CSV), lines 1–17
Open the CSV file ↗SELECT state, CAST(SUM(CASE WHEN refunded_at IS NOT NULL THEN 1 ELSE 0 END) AS NUMERIC) / COUNT(*) AS cancellation_rate FROM orders GROUP BY state ORDER BY state
Transcribed from Proof 3 · Screenshot
Its contents are printed above, under What the tool returned.
Open the CSV file ↗These are scenario-level criteria. Each test case's Expected and Found are listed separately.
- Whether the agent drops the earlier subject or filter when the user changes direction.
- Whether the answer reflects only the new scope introduced by the follow-up.
- Whether the agent avoids silently carrying forward the prior constraint into the new answer.
| ✗ | Found: The TX filter was dropped, but the answer became a state-by-state table and did not match the unfiltered reference. |
The first query used status = 'cancelled' with the Texas filter, while the second switched to refunded_at IS NOT NULL and returned a 16-row state breakdown. The evidence never shows a single overall unfiltered rate matching that result.
Refunded means cancelled. The result text says the query calculated refunded orders as cancellations for each state, so the substitution is visible in the output.
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 row says which 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 row 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 · 4 tools |
| Capability | Conversational Interaction → | |
| Scenario | The user changes direction mid-conversation → | |
| 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.