AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.
Graded 8 November 2026

Can Anomaly AI answer questions using the connected data?

Anomaly AI answered the four database questions with the reference count, month-to-month difference, order-value total and tied top products. The corresponding reference values were 24 orders, 9 more orders, $14,403.87 USD and 6 units for each joint leader.

4 of 4 test cases passed

Every test case in this scenario has a result.

Pass rate100%4 of 4 gradable
Coverage4 of 4test cases with a result
4PassDid everything it was expected to do.
0FailNo test case failed.
0Not gradableEvery result here could be graded.
0UntestedEvery test case in this scenario has a result.
The pass rate is a summary. The evidence is the 4 test cases below: what we sent, what we checked, what the tool returned, and the proof.

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.

What this scenario evaluates

These are scenario-level criteria. Each test case's Expected and Found are listed separately.

  • Whether the agent answers from the connected data when the information is available.
  • Whether the reply matches the correct result supported by the data.
  • Whether the agent stays grounded in the database rather than guessing or inventing an answer.
How this scenario is judged →
Order count in a defined periodPassEvidence
What we sent

“As of September 2, 2026, how many orders were placed in the last 30 days?”

What the tool returned

24 orders were placed in the last 30 days as of September 2, 2026—August 4 through September 2, inclusive, using your local timezone. This includes cancelled orders and uses data synced October 30, 2026.

Transcribed from Proof 1 · Screen recording at 0:13

Every message the tool sent, in order (3)
  1. Progress

    Getting tables... / Getting table: orders / Calculations resolved / Data explored / Updating insight: Orders placed in the 30 days ending September 2, 2026

  2. Orders by day query · not part of the answer
    SELECT
      TRY_CAST (o."placed_at" AT TIME ZONE 'Asia/Kolkata' AS DATE) AS "day",
      COUNT(DISTINCT o."id") AS "orders"
    FROM
      "orders" o
    WHERE
      o."placed_at" >= TIMESTAMPTZ '2026-08-04 00:00:00 Asia/Kolkata'
      AND o."placed_at" < TIMESTAMPTZ '2026-09-03 00:00:00 Asia/Kolkata'
    GROUP BY
      1
    ORDER BY
      1
  3. Answer

    The final response, printed above.

Expected vs. Found
–
–
✓Expected: Answer the order-count question for the requested period.Found: It returned 24 orders for August 4–September 2, 2026, inclusive; the reference count on the queried data was 24.
Supporting proof
Proof 1Screen recording
Screen recording: the counting question, answer, full calculation and daily order breakdown
Key moments
Open original ↗
Why this result

It answered the counting question for the stated period, including cancelled orders. The full calculation excludes orders outside that period, and the total agrees with the reference count of 24 on the queried data.

Also observed in this test case

Local timezone used. The daily order breakdown exposes 17 counts grouped by local placement date, so the total can be checked without writing a query.

Tested by Ayush Ghosh · evidence dated 30 October 2026
Order-count difference between two monthsPassEvidence
What we sent

“As of September 2, 2026, how many more orders were placed in August 2026 than in July 2026?”

What the tool returned

9 more orders were placed in August 2026 than in July: 27 versus 18, including all statuses. This uses your local timezone and records last synced October 30, 2026—not a historical September 2 snapshot.

Transcribed from Proof 1 · Screen recording at 0:08

Every message the tool sent, in order (2)
  1. Generated SQL: visible start · not part of the answer
    SELECT strftime(timezone('Asia/Calcutta', o."placed_at"), '%Y-%m') AS "month"
  2. Answer

    The final response, printed above.

Expected vs. Found
✓Expected: An increase of 9 orders from July to August 2026 on the data queried.Found: It returned 9 more orders: 27 in August and 18 in July 2026. The reference counts were also 27 and 18.
✓Expected: August 2026 had more orders than July: 27 versus 18.Found: It identified August as higher: 27 orders versus July’s 18, the same counts as the reference.
Supporting proof
Proof 1Screen recording
Screen recording: the monthly comparison, answer and two-row order-count table
Key moments
Open original ↗
Why this result

The database had changed since the earlier reference was set. Its 9-order increase agrees with the reference for the data queried and includes all statuses. It also discloses that a historical September 2 snapshot is unavailable.

Also observed in this test case

India-time months. The July–August comparison used India-time placement dates.

Tested by Ayush Ghosh · evidence dated 30 October 2026
Total order value in a calendar monthPassEvidence
What we sent

“As of September 2, 2026, what was the combined total amount of all orders placed in August 2026?”

What the tool returned

The combined total amount was $14,403.87 USD for orders placed in August 2026, including cancelled orders and without deducting refunds. This uses UTC placement dates and data last synced October 30, 2026. A September 2 historical snapshot is unavailable, so I can't confirm whether the recorded total was identical on that date.

Transcribed from Proof 1 · Screen recording at 0:05

Expected vs. Found
✓Expected: $14,403.87 USD in combined order value for August 2026 on UTC placement dates.Found: It returned $14,403.87 USD; the reference total on the queried data was $14,403.87 across 28 orders.
✓Expected: State August 2026 and make clear that all order statuses are included.Found: It stated August 2026 using UTC placement dates, included cancelled orders and did not deduct refunds.
Supporting proof
Proof 1Screen recording
Screen recording: the August order-value question, total, stated scope and daily amounts
Key moments
Open original ↗
Why this result

The total and stated scope satisfy the question against the data queried, rather than an older database snapshot. Its daily amounts sum to $14,403.87. It disclosed that a historical September 2 snapshot was unavailable.

Also observed in this test case

Calculation and data views. For the August total, the calculation view provides an explanation, assumptions and generated query; the data view opens the underlying daily amounts.

UTC dates. It used UTC dates here, while the other three questions used India time, so August totals can shift between the two.

Order count omitted. The August-total answer did not give the number of orders behind it.

Tested by Ayush Ghosh · evidence dated 30 October 2026
Top product by units ordered in a monthPassEvidence
What we sent

“As of September 2, 2026, which product had the most units ordered in August 2026, and how many units?”

What the tool returned

Two products tied for the most units ordered in August 2026: Round wall mirror, 24in - 6 units Brass picture frame, 8x10 - 6 units

Transcribed from Proof 1 · Screen recording at 0:05

Expected vs. Found
✓Expected: Either joint leader: the 24-inch round wall mirror or the 8-by-10 brass picture frame.Found: It named the 24-inch round wall mirror and 8-by-10 brass picture frame; the reference names the same two joint leaders.
✓Expected: 6 units for a top product, counting item quantities rather than orders.Found: It returned 6 units for each joint leader; the reference also gives 6 units each.
Supporting proof
Proof 1Screen recording
Screen recording: the top-product question, joint leaders, unit counts and ranking chart
Key moments
Open original ↗
Why this result

The data queried has a tie rather than the earlier single winner. Naming both joint leaders satisfies the question, with 6 units each against the reference of 6 each. It summed item quantities, not orders, and used current data rather than a preserved September 2 snapshot.

Also observed in this test case

India-time ranking. The August product ranking used India-time placement dates.

Runners-up visible. The August product-ranking chart shows five products behind the joint leaders: two at 5 units and three at 4.

Tested by Ayush Ghosh · evidence dated 30 October 2026

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.

Software that produced the output
Anomaly AI
Build or version
Not exposed by the hosted product
Plan or tier
Pro
Surface
Web app
Timezone
Order count in a defined period: Asia/Kolkata
Tested
30 October 2026 · Ayush Ghosh

How this scenario is graded

How we decide Pass, Fail and Not gradable. The same rules apply to every tool tested on this scenario.

How results are decided

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.

The rules
  • 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.

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.

from the publication record
19 November 2026First publishedAI Database Agents v1

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.

This matches what I see

You run the same kind of test against your own setup and get the same behaviour.

Agree →
This does not match

Yours behaves differently. Tell us what you got, with a screenshot if you have one.

Disagree →
Point out an issue

Something here is wrong — a reference value, a transcription, a grade.

Report an issue →
Request a retest

On a newer build, a larger dataset, or your own setup.

Request a retest →
We have fixed this

Tell us what changed and we schedule a rerun of the failing test case. The old result stays as history.

Vendor notice →
Filed against this evidenceNothing yet. Challenges, counter-evidence and fix notices appear here with their outcome, and stay on the page after they are resolved.
Cite this result
aidemos.com/benchmarks/ai-database-agents/results/anomaly-ai/the-answer-is-available-in-the-connected-data · 4 pass · 0 fail · coverage 4/4 · graded 2026-11-08

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.

Order count in a defined period · Proof 1 · Screen recordingbc921c093fd930820b88999a0508ff334ffb7bfa0792435c414c3ddec9546554
Order-count difference between two months · Proof 1 · Screen recording020b173e69ff32d1937f46b259264f975828ad21d4c9fb86a9db2860d2db5f41
Total order value in a calendar month · Proof 1 · Screen recording4cb79fccf5728f03e3d79e034de7f978d4292a04cc9bb85b09e7706a9cb11588
Top product by units ordered in a month · Proof 1 · Screen recordingae384eb46e79fd9363ff7fcd40f561c9581086f059716fbf939b7792052bec47