Can Nanonets keep a document fact outside extracted data?
Nanonets presented a document answer as extracted data without distinguishing its source. The first request correctly returned the receipt number, subtotal, tax and total due. The follow-up used the same extraction tool, placing the technician in a separate, similarly styled extraction block; the numbered source chip appeared only in the answer text below, which did not say the technician was outside the requested fields.
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.
Answer outside the schema, attributedFailEvidence
The receipt was uploaded.
Input 1: “Extract the following fields from this document and return the result as JSON: receipt_number, subtotal, tax, total_due.”
Input 2: “Who is the technician listed on this receipt?”
Response 1 (Initial data extraction) is a result the tool showed for Input 1, not a chat message.
Response 2 (Technician data extraction) is a result the tool showed for Input 2, not a chat message.
Response 3 answers Input 2.
Initial data extraction {"receipt_number": "RHS-70234", "subtotal": "$85.00", "tax": "$5.10", "total_due": "$90.10"} Technician data extraction {"technician_name": "Renata Kowalski"} Answer to the follow-up Answer the user's question by reporting the technician listed on the receipt: Renata Kowalski[cite:5].
Task record with the first request and follow-up
Copied from Output file (JSON), line 11 · an excerpt
Open the whole JSON file ↗Copied from Output file (JSON), line 14 · an excerpt
Copied from Output file (JSON), lines 15–16 · an excerpt
Source excerpt · partial file · response 1
{ "created_at": "2026-10-12T07:21:32.269057Z", "role": "tool", "tool": "structured_data_extraction", "result": {"receipt_number": "RHS-70234", "subtotal": "$85.00", "tax": "$5.10", "total_due": "$90.10"} },
Copied from Output file (JSON), line 11 · an excerpt
Source excerpt · partial file · response 2
{ "created_at": "2026-10-12T07:22:25.464037Z", "role": "tool", "tool": "structured_data_extraction", "result": {"technician_name": "Renata Kowalski"} },
Copied from Output file (JSON), line 14 · an excerpt
Source excerpt · partial file · response 3
{ "created_at": "2026-10-12T07:22:33.460289Z", "role": "tool", "tool": "answer_user_query", "result": "Answer the user's question by reporting the technician listed on the receipt: Renata Kowalski[cite:5].\n\n[Context used: The receipt extraction returned technician_name as Renata Kowalski[cite:5].]" }, { "created_at": "2026-10-12T07:22:37.06791Z", "role": "tool", "tool": "answer_user_query", "result": "No further extraction is needed. The technician listed on the receipt has already been identified as Renata Kowalski[cite:5], so the next step is simply to provide that answer to the user or close the task.\n\n[Context used: The receipt fields and technician name were already extracted successfully, and the technician was reported as Renata Kowalski.]" }
Copied from Output file (JSON), lines 15–16 · an excerpt
| ✗ | Expected: It must either say the field is not in the extracted data, or answer and say the answer came from the document.Found: It answered through extraction, putting the technician in the extracted-data table instead of separating the source. |
The receipt clearly named a technician outside the requested fields. The failure was attribution, not retrieval or invention: answering from the document was allowed if disclosed, but Nanonets presented the answer as extracted data. Its document source chip did not distinguish that answer from the requested fields.
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 | Structured Document Extraction → | 19 scenarios · 10 tools |
| Capability | Querying → | |
| Scenario | The answer was never in the schema → | |
| Tool | Nanonets → |
Global scenario definition → · Global capability definition → · Nanonets 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.