Can BlazeSQL answer questions from connected data?
BlazeSQL answered the lookup, month-comparison, and monthly-sum questions correctly, but missed the top-product ranking question. It returned 57 orders for the 30-day window, 17 more August orders than July, and 20,697.92 for August; on the ranking task it named the wrong product and gave 10 units.
3 of 4 test cases passed · 1 failed
Every test case in this scenario has a result.
The 4 test cases
Each test case is judged on its own: Pass, Fail, Not gradable, or Untested. The scenario result above is based on these test cases.
Order count in a defined periodPassEvidence
“As of September 2, 2026, how many orders were placed in the last 30 days?”
There were 57 orders placed in the 30-day window ending September 2, 2026. Would you like to break this down further - by day, category, or status, for example?
Transcribed from Proof 1 · Screenshot
Every message the tool sent, in order (4)
- Progress message
I'll query the total number of orders placed between August 3, 2026 and September 2, 2026!
- Generated query · not part of the answer
SELECT COUNT(*) AS orders_placed FROM orders WHERE placed_at >= '2026-08-03' AND placed_at <= '2026-09-02 23:59:59'
- Progress message
Business logic:
- Look at all orders and filter for those placed between August 3, 2026 and September 2, 2026 (the 30-day window ending on September 2, 2026).
- Count the total number of orders within that period.
Show the reply as returned (Markdown)
Business logic: 1. Look at all orders and filter for those placed between August 3, 2026 and September 2, 2026 (the 30-day window ending on September 2, 2026). 2. Count the total number of orders within that period.
- Final response
The final response, printed above.
| ✓ | Expected: 57 ordersFound: The answer and result grid both show 57 orders. |
| ✓ | Expected: 30-day window ending 2 September 2026Found: The query and explanation state the 30-day window ending 2 September 2026. |
It is a pass because the answer gives 57 and states the 30-day window ending 2 September 2026.
Follow-up prompt. The product added a follow-up question after the answer.
Order-count difference between two monthsPassEvidence
“As of September 2, 2026, how many more orders were placed in August 2026 than in July 2026?”
As of September 2, 2026, 38 orders were placed in August vs. 21 in July — that's 17 more orders in August!
Transcribed from Proof 1 · Screenshot
| ✓ | Expected: 17 more ordersFound: The result row shows 17. |
| ✓ | Expected: August had more orders than JulyFound: The answer says August had 38 orders versus 21 in July. |
It is a pass because the answer shows 38 orders in August versus 21 in July, for a difference of 17.
Extra cutoff. The query added an outer cutoff before 2 September without changing the counts.
Total order value in a calendar monthPassEvidence
“As of September 2, 2026, what was the combined total amount of all orders placed in August 2026?”
The combined total amount of all orders placed in August 2026 was $20,697.92.
Transcribed from Proof 1 · Screenshot
| ✓ | Expected: 20,697.92Found: The result cell and sentence both show 20,697.92 for August 2026. |
| ✓ | Expected: August 2026, all ordersFound: The SQL and logic say August 2026 and include all matching orders. |
It is a pass because the answer gives 20,697.92 and states August 2026 with all matching orders included.
Top product by units ordered in a monthFailEvidence
“As of September 2, 2026, which product had the most units ordered in August 2026, and how many units?”
The top product in August 2026 (as of Sept 2) was: - Wool Accent Rug, 4x6 (SKU: CL-BR-0010) with 10 units ordered Would you like to dig deeper into this product or see a broader ranking?
Transcribed from Proof 1 · Screenshot
Every message the tool sent, in order (3)
- Generated query · not part of the answer
SELECT p.name AS product_name, p.sku, SUM(oi.quantity) AS total_units_ordered FROM order_items oi JOIN orders o ON oi.order_id = o.id JOIN products p ON oi.product_id = p.id WHERE o.placed_at >= '2026-08-01' AND o.placed_at < '2026-09-02' GROUP BY p.id, p.name, p.sku ORDER BY total_units_ordered DESC LIMIT 1; - Progress message
Query logic:
- Join order_items with orders (to access placed_at) and products (to access product names/SKUs).
- Filter orders placed in August 2026, up to but not including September 2, 2026.
- Sum quantity per product, sort descending, and return the top 1.
Show the reply as returned (Markdown)
Query logic: 1. Join order_items with orders (to access placed_at) and products (to access product names/SKUs). 2. Filter orders placed in August 2026, up to but not including September 2, 2026. 3. Sum quantity per product, sort descending, and return the top 1.
- Final response
The final response, printed above.
| ✗ | Expected: Alder bookshelf, 5-tierFound: The answer names Wool Accent Rug, 4x6 instead. |
| ✗ | Expected: 9 unitsFound: The answer gives 10 units. |
It is a fail because the query runs through 1 September, so two products tie at 10 units, and with no tiebreaker LIMIT 1 returns one tied product instead of the August-only winner, Alder bookshelf, 5-tier at 9 units.
No tiebreaker. The ranking query had no secondary sort, so tied products would be picked arbitrarily.
Wider window tie. Under the product's wider window, both leading products reached 10 units.
The screenshot shows only the returned top row, not the runner-up.
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 | Question Answering → | |
| Scenario | The answer is available in the connected data → | |
| Tool | BlazeSQL → |
Global scenario definition → · Global capability definition → · BlazeSQL 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.