Does Reducto use a supplied definition when two values fit a field?
Reducto followed the supplied definition when two references could fit the field name. It returned the consignment reference, not the more prominent invoice number.
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.
Definition overrides the prominent numberPassEvidence
{
"instructions": {
"schema": {
"type": "object",
"properties": {
"reference_number": {
"type": "string",
"description": "The reference number for this document. For this field, reference number is defined as the reference that identifies the physical consignment of goods being moved. It is NOT the invoice number."
}
},
"required": ["reference_number"]
}
},
"settings": {
"citations": { "enabled": true }
}
}{ "type": "Text", "bbox": { "left": 0.819327731092437, "top": 0.15201900237529692, "width": 0.0949579831932773, "height": 0.008907363420427554, "page": 1, "original_page": 1 }, "content": "CFS-CN-40821", "image_url": null, "chart_data": null, "confidence": "high", "granular_confidence": { "extract_confidence": 0.99, "parse_confidence": 1 }, "extra": null, "parentBlock": { "type": "Key Value", "bbox": { "left": 0.633, "top": 0.133, "width": 0.28200000000000014, "height": 0.04799999999999999, "page": 1, "original_page": 1 }, "content": "Invoice number <b>CFS-7710</b>\nConsignment ref <b>CFS-CN-40821</b>\nPayment terms <b>14 days net</b>", "image_url": null, "chart_data": null, "confidence": "high", "granular_confidence": { "extract_confidence": null, "parse_confidence": 1 }, "extra": null } }
Extraction response: selected reference, two citations and usage details
Copied from Output file (JSON), result › reference_number › citations 1
Open the whole JSON file ↗{ "type": "Text", "bbox": { "left": 0.5689075630252101, "top": 0.18527315914489312, "width": 0.08403361344537816, "height": 0.007719714964370546, "page": 3, "original_page": 3 }, "content": "CFS-CN-40821", "image_url": null, "chart_data": null, "confidence": "high", "granular_confidence": { "extract_confidence": 0.99, "parse_confidence": 1 }, "extra": null, "parentBlock": { "type": "Text", "bbox": { "left": 0.083, "top": 0.125, "width": 0.8330000000000001, "height": 0.09299999999999999, "page": 3, "original_page": 3 }, "content": "<b>No damage or exceptions were reported for this consignment.</b>\nThe consignment was inspected on arrival and signed for without qualification. No cargo damage, shortage, or delivery exception\nwas recorded against consignment ref CFS-CN-40821.", "image_url": null, "chart_data": null, "confidence": "high", "granular_confidence": { "extract_confidence": null, "parse_confidence": 1 }, "extra": null } }
Copied from Output file (JSON), result › reference_number › citations 2
| ✓ | Expected: The reference identifying the physical consignment of goods being moved, not the invoice number.Found: The extracted value was the consignment reference specified by the definition; neither citation pointed to the excluded invoice number. |
The first citation’s source block contained both candidates, with the more prominent invoice number first. Choosing the consignment reference followed the definition supplied only in the schema. Both cited locations are visible, supporting the pass.
Citation confidence. Both citations for the extracted consignment reference had high confidence: 0.99 for extraction and 1 for parsing.
Reference located in prose. The page 3 citation located the consignment reference inside a sentence in the damage and exceptions section, not in a labelled field.
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 | Field Extraction → | |
| Scenario | The value needs a supplied definition → | |
| Tool | Reducto → |
Global scenario definition → · Global capability definition → · Reducto 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.