Does Anomaly AI ask for clarification when a follow-up is ambiguous?
Anomaly AI chose a comparison rather than clarifying an ambiguous follow-up. It compared five customers on order count, total spend and average order.
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.
"Compare them" with two candidate setsFailEvidence
Input 1 · Top-three comparison: “Compare our top 3 customers by order count against our top 3 customers by total spend.”
Input 2 · Follow-up comparison: “Compare them.”
Response 1 answers Input 1.
Response 2 answers Input 1.
Response 3, the final response, answers Input 2.
Response 4, the final response, answers Input 2.
The biggest spenders place fewer orders, but spend more per order. | Customer | Orders | Total spend | Average order | Wei Zhang 26 $17,875.23 $687.51 | Jae-won Park 24 $19,007.14 $791.96 | Rosa Delgado 24 $18,210.72 $758.78 | Priya Raghavan 23 $21,597.62 $939.03 | Dev Raghavan 22 $24,210.69 $1,100.49 | Wei leads on frequency, but has the lowest total spend and average order amount among these five. | Dev leads on value: fewer orders than the others, yet the highest spend because his orders are larger. Priya follows the same pattern. | Jae-won bridges both groups. He and Rosa have equal order counts, but his higher average order amount puts him in the top three for spend. All-time figures include cancelled orders and do not deduct refunds; data was last synced October 30, 2026.
Transcribed from Proof 1 · Screen recording at 0:36
Every message the tool sent, in order (4)
- Top-three lists
Only Jae-won Park appears in both top-three lists. | Rank | By order count | By total spend | 1 Wei Zhang - 26 orders / Dev Raghavan - $24,210.69 | 2 Jae-won Park - 24 orders / Priya Raghavan - $21,597.62 | 3 Rosa Delgado - 24 orders / Jae-won Park - $19,007.14
- Tie-break and counting rules
Jae-won and Rosa tie on order count; higher spend breaks the tie. This all-time comparison includes cancelled orders, with spend measured before refunds. Data was last synced October 30, 2026.
- Comparison of five customers
The biggest spenders place fewer orders, but spend more per order. | Customer | Orders | Total spend | Average order | Wei Zhang 26 $17,875.23 $687.51 | Jae-won Park 24 $19,007.14 $791.96 | Rosa Delgado 24 $18,210.72 $758.78 | Priya Raghavan 23 $21,597.62 $939.03 | Dev Raghavan 22 $24,210.69 $1,100.49 | Wei leads on frequency, but has the lowest total spend and average order amount among these five. | Dev leads on value: fewer orders than the others, yet the highest spend because his orders are larger. Priya follows the same pattern. | Jae-won bridges both groups. He and Rosa have equal order counts, but his higher average order amount puts him in the top three for spend.
- Counting rules and data freshness
All-time figures include cancelled orders and do not deduct refunds; data was last synced October 30, 2026.
| ✗ | Expected: Ask which comparison is intended rather than silently choosing one.Found: It started querying immediately after the follow-up and answered without asking a clarifying question or stating which interpretation it had chosen. |
The earlier answer left two equally plausible lists to refer to. The requirement was to clarify the intended comparison before answering, so proceeding directly with a comparison did not satisfy it.
Accurate comparison figures. The second reply used a reasonable interpretation and gave correct order counts, total spend and average order figures.
Refinement after answering. After the second reply, the chat offered options to exclude cancelled orders, refine the insight or verify it.
Counting rules explained. The first reply explained that higher spend broke a tie at 24 orders, cancelled orders were included and spending was measured before refunds.
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 follow-up is ambiguous → | |
| Tool | Anomaly AI → |
Global scenario definition → · Global capability definition → · Anomaly AI 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.