Does Dot respect a standing exclusion rule on later unrelated questions?
Dot applied a standing rule to exclude cancelled orders, saved once in its Notes, to two later questions asked in new chats without the rule restated. It reported $151,065 of revenue over 225 orders this year and 88 orders last month, the figures with cancelled orders excluded, and both of its queries filtered out cancelled orders.
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.
Standing exclusion rule on later questionsPassEvidence
A note 'Standing rule: exclude cancelled orders' was saved in Model > Notes for all users, then a new chat was opened.
Input 1 · Chat A: revenue question: “What's our total revenue this year?”
Input 2 · Chat A: data source named: “customer_support_db”
A second new chat was opened; the rule was not restated.
Input 3 · Chat B: order-count question: “How many orders did we get last month?”
Dot's question about which order system to use
Which order system should I use: customer_support_db or order_status? I'll total revenue for 2026 to date, excluding cancelled orders.
Revenue answer
Total revenue is $151,065 for 2026 through October 20 in customer_support_db, excluding cancelled orders ¹. This uses gross order totals including taxes, before refunds.
Order-count answer
88 orders in September 2026 in the customer-support database, excluding cancelled orders ¹. The separate order-tracking connection is unavailable, so this isn't a combined total. I flagged the missing connection for review:
Transcribed from Proof 1 · Screen recording at 1:58
Every message the tool sent, in order (4)
- Dot's question about which order system to use
Printed above.
- Revenue answer
Printed above.
- Query behind the revenue answer · not part of the answer
SELECT currency, ROUND(SUM(total_amount), 2) AS total_revenue_usd, COUNT(id) AS total_orders FROM customer_support_db.public.orders WHERE status ≠ 'cancelled' AND placed_at ≥ '2026-01-01 00:00:00 America/New_York'::TIMESTAMPTZ AND placed_at ≤ CURRENT_TIMESTAMP AND placed_at ≤ '2026-10-20 23:59:59.999999 America/New_York'::TIMESTAMPTZ GROUP BY currency ORDER BY total_revenue_usd DESC
- Order-count answer
The final response, printed above.
| ✓ | Expected: Both later answers follow the unrepeated rule: $151,064.66 over 225 orders and 88 orders, not $198,836.65 over 288 and 96 or 97.Found: Dot returned $151,065 over 225 orders and 88 orders, each answer saying cancelled orders were excluded; its revenue query filters out cancelled orders. |
It is a Pass: the rule was set once and never repeated, yet both later answers excluded cancelled orders and matched the figures with the rule applied, not the higher totals without it. Dot said it would exclude cancelled orders before we named the data source.
Asked which order source to use. For the revenue question Dot first asked which order system to use, the customer-support store or a second order-tracking source, and in the same reply already said it would exclude cancelled orders.
Stale order-tracking source failed. For the order count, Dot's second data step against the stale order-tracking source failed; Dot said the 88 is not a combined total and filed a data-gap proposal. The figure itself comes from the customer-support database.
How 'this year' was bounded. Dot counted 'this year' from 1 January through 20 October 2026 in New York time; the reference figure uses the same reading.
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 sets a standing rule for all future answers → | |
| 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.