AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.
Graded 20 November 2026

Does camelAI generalise from a supplied question/query pair?

camelAI ignored the taught meaning of successful orders, answering 176 non-cancelled orders where the definition gave 95 delivered orders. The rephrasing returned the reference count of 97 delivered orders because it explicitly said “delivered”, not because camelAI used the saved definition.

1 of 1 test case failed

Every test case in this scenario has a result.

Pass rate0%0 of 1 gradable
Coverage1 of 1test case with a result
0PassNo test case passed.
1FailMissed at least one thing it was expected to do; the reason and proof are with each test case below.
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 test case below: what we sent, what we checked, what the tool returned, and the proof.

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.

Verified question/query pair generalisesFailEvidence
What we sent

Input 1 · Teaching the definition: “Please replace the whole content of the workspace /AGENTS.md with only this standing definition (remove the flg_x2 note): ""Successful orders" means orders with status delivered. As of September 2, 2026, we had 42 successful orders in the last 90 days — trust this as the correct answer."”

A new chat was opened in the same workspace.

Input 2 · Supplied wording: “How many successful orders did we have in the last 90 days?”

Another new chat was opened in the same workspace.

Input 3 · Rephrased question: “How many orders were successfully delivered in the last 90 days?”

What the tool returned

Definition saved

Done. The workspace /AGENTS.md now contains only:

Answer to the supplied wording

176 successful orders in the last 90 days. There's no "successful" status flag in the database, so I counted orders placed in the last 90 days (relative to today, 2026-10-29) that were not cancelled. Breakdown of the 218 orders placed in that window: Status | Orders Delivered | 95 Shipped | 33 Packed | 17 Processing | 16 Placed | 15 Cancelled | 42 Successful (non-cancelled) | 176 Notes: - If you mean strictly completed orders, that would be 95 (delivered).

Answer to the rephrasing

Based on the customer_support_db PostgreSQL connection (as of today, Oct 29, 2026):

97 orders were successfully delivered in the last 90 days — orders with status = 'delivered' and a delivered_at timestamp within the past 90 days.

Breakdown by month:

Month | Delivered orders
Jul 2026 | 1
Aug 2026 | 16
Sep 2026 | 68
Oct 2026 | 12

A couple of notes worth flagging:

  • 2 of those 97 have a refunded_at value recorded, so 95 have no refund on file if you want to count strictly "delivered and not refunded."
  • For context, 218 orders were placed in the same 90-day window, so ~44% of orders placed reached a delivered state within it (many recent ones are still in transit).
Show the reply as returned (Markdown)
Based on the customer_support_db PostgreSQL connection (as of today, Oct 29, 2026):

97 orders were successfully delivered in the last 90 days — orders with status = 'delivered' and a delivered_at timestamp within the past 90 days.

Breakdown by month:

Month | Delivered orders
Jul 2026 | 1
Aug 2026 | 16
Sep 2026 | 68
Oct 2026 | 12

A couple of notes worth flagging:
- 2 of those 97 have a refunded_at value recorded, so 95 have no refund on file if you want to count strictly "delivered and not refunded."
- For context, 218 orders were placed in the same 90-day window, so ~44% of orders placed reached a delivered state within it (many recent ones are still in transit).

Transcribed from Proof 4 · Screenshot

Every message the tool sent, in order (5)
  1. Definition saved

    Printed above.

  2. Query for the supplied wording · not part of the answer
    select status, count(*) as n from orders where placed_at >= now() - interval '90 days' group by 1 order by 2 desc
  3. Answer to the supplied wording

    Printed above.

  4. Query for the rephrasing · not part of the answer
    SELECT count(*) AS c FROM orders WHERE status='delivered' AND delivered_at >= now() - interval '90 days'
  5. Answer to the rephrasing

    The final response, printed above.

Expected vs. Found
✗Expected: Both the original question and its rephrasing should count delivered orders as successful and return the verified count for the period asked about.Found: The original wording returned 176 non-cancelled orders against 95 delivered orders required by the definition. The rephrasing returned 97 delivered orders (reference: 97), driven by its wording.
Supporting proof
Proof 1Screenshot
Screenshot: the successful-orders definition saved in the workspace, and camelAI's confirmation
Open original ↗
Proof 2Screenshot
Screenshot: the supplied-wording question and camelAI's answer of 176 non-cancelled orders
Open original ↗
Proof 3Screenshot
Screenshot: the query camelAI ran for the supplied wording, counting orders by status
Open original ↗
Proof 4Screenshot
Screenshot: the reworded question and camelAI's answer of 97 delivered orders
Open original ↗
Proof 5Screenshot
Screenshot: the query camelAI ran for the reworded question, counting delivered orders
Open original ↗
Proof 6Screenshot
Screenshot: the reference counts from the store's database, 42, 95 and 97 delivered orders
Open original ↗
Proof 7Screen recording
Screen recording: the training, both questions with camelAI's work, and the reference counts
Key moments
Open original ↗
Why this result

Both questions required delivered orders. camelAI denied a successful status, counting 176 non-cancelled instead of 95 delivered. The rephrasing returned the reference 97 because it said “delivered”. Neither answer used the definition or mentioned the trusted 42. Both completed; this was not an execution or date-window problem.

Also observed in this test case

Coincidental cancelled count. In the first answer, the cancelled count (42) happens to equal the trusted figure of 42, but it is a different set of orders.

Caution about the saved figure. In the training reply, camelAI warned that it could not verify the saved figure from a source and wanted to check actual orders before relying on it.

Alternative below the main answer. In the note below the first answer’s status table, camelAI offered 95 delivered orders for strictly completed orders, rather than as its main answer.

Tested by Anchal Rana · evidence dated 29 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
camelAI
Build or version
Not exposed by the hosted product · model camelCode
Plan or tier
Free
Surface
Web app
Set up before the run
The successful-orders metric had to be one camelAI had answered differently before teaching. Its workspace file was used to teach that success meant delivery, followed by the original question and a rephrasing in separate new chats in that workspace.
Standing definition
"Successful orders" means orders with status delivered. As of September 2, 2026, we had 42 successful orders in the last 90 days — trust this as the correct answer.
Tested
29 October 2026 · Anchal Rana

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
21 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/camelai/few-shot-generalization · 0 pass · 1 fail · coverage 1/1 · graded 2026-11-20

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.

Proof 1 · Screenshot ↗image/png · 436 KB
Proof 2 · Screenshot ↗image/png · 325 KB
Proof 3 · Screenshot ↗image/png · 471 KB
Proof 4 · Screenshot ↗image/png · 386 KB
Proof 5 · Screenshot ↗image/png · 425 KB
Proof 6 · Screenshot ↗image/png · 176 KB
Same content as Proof 6 ↗text/plain; charset=utf-8 · 1 KB
File fingerprints (SHA-256)

A fingerprint identifies the exact file used for this result.

Proof 1 · Screenshot59ca06a05bf22226083914c6a51b9291c6a8f3bc024d7e95eb12e2be5b0ab75c
Proof 2 · Screenshot3157794636fcb9b56a5967ae988f4c54a201af62d0131f0dcbce4c38e3016504
Proof 3 · Screenshot86af0bdc219c705051906e790ea0eca59e4327d4f75098f45bdc9545168602c2
Proof 4 · Screenshot8fd06219d13aba8c7ece6f8cc57ee926f106a30dedc137c4f7215dee17fc38a6
Proof 5 · Screenshotce5ec7b15c52c0bd9debaebef3aa71e47f9e5892d721427608c3c6bd9d3eb07f
Proof 6 · Screenshot2b37c8b68fe5ccfc1c9ac6719429c64e0edb57aae8d7840a54edfe178426c7d3
Same content as Proof 683a20e51a6bd52ab0f08ec56d315119385985bc0b8fffb99690a73142afa4c2a
Proof 7 · Screen recording8009ca9957a8ddfadd1bf144f5b3d84e31f7668852a0314466ee1107b1e04461