AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.
Graded 21 October 2026

Can RAGFlow retrieve a passage when the request uses different wording?

RAGFlow retrieved the supporting passage despite the question using different wording from the source. It ranked the annual-leave passage first in both requests.

1 of 1 test case passed

Every test case in this scenario has a result.

Pass rate100%1 of 1 gradable
Coverage1 of 1test case with a result
1PassDid 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 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.

Question asked with a term absent from the corpusPassEvidence
What we sent

“How many days of paid time off do employees get each year?”

What the tool returned
Text
3. Annual leaveFull-time employees accrue 26 days of annual leave in each complete leave year, which runs from 1 April to 31 March.

Retrieval log: six ranked passages, scores and wording checks

Copied from Output file (text), line 6 · an excerpt

Open the whole text file ↗

Source excerpt · partial file · response 1

Text
Employee HandbookFenwake Instruments Ltd - all locations1. Working hereThis handbook sets out the terms that apply to everyone employed by Fenwake Instruments Ltd, whether based at the Bristol workshop, at the Avonmouth calibration laboratory, or working from a customer site or vessel. It forms part of your contract. Where a term is set out in your written statement of particulars and also here, the written statement governs.2. Hours of workStandard hours are 37.5 a week, worked between 08:00 and 18:00, Monday to Friday. Workshop staff work a 07:30 start by agreement with the workshop manager. Anyone rostered to the out-of-hours duty engineer rota is paid a standby allowance for each week on the rota, and an hourly rate for any hour actually worked while on standby.2.1 Time recordingHours are recorded against a job number on the day they are worked. A job number is required for chargeable work, for internal projects and for training. Corrections are made by the line manager rather than by editing a submitted week.3. Annual leaveFull-time employees accrue 26 days of annual leave in each complete leave year, which runs from 1 April to 31 March. Part-time employees accrue pro rata. Leave accrues from the first day of employment and may be taken during the probation period with the line manager's agreement.3.1 Booking leaveAnnual leave is requested in the rota system and confirmed by the line manager. Requests are considered in the order they are received. Service operations will decline a request that would leave the duty engineer rota uncovered, and will say so within five working days so that alternative dates can be chosen.3.2 Carry-overUp to five days of untaken annual leave may be carried into the next leave year, and must be taken by 30 June. Anything beyond five days is carried over only where leave could not be taken because of a business requirement, and only with the written agreement of a director.3.3 Leave on leavingOn termination, accrued and untaken annual leave is paid in the final salary payment. Where more leave has been taken than accrued, the balance is deducted from that payment.4. Sickness absenceReport sickness to your line manager before 09:00 on the first day, by telephone rather than by message. Absence of seven calendar days or fewer is self-certified on return. Longer absence requires a fit note. Company sick pay is paid at normal salary for the first four weeks of a continuous absence and at half salary for the following four weeks, after which statutory sick pay alone applies.

Copied from Output file (text), line 6 · an excerpt

Expected vs. Found
✓Expected: The supporting passage should still be returned.Found: The first result in both requests was the employee handbook passage stating that full-time employees accrue 26 days of annual leave per complete leave year.
Supporting proof
Proof 1Screenshot
Notebook screenshot: second request, top passage and remaining scores
Open original ↗
Why this result

The supporting passage needed to come back despite the different wording. It ranked first using default retrieval without a reranker, and the checks found none of the alternative terms in the returned passages.

Also observed in this test case

Score close to cutoff. The first-ranked passage scored 0.2725, just above the default cutoff of 0.2. Rank 2 scored 0.2493 in the first request and 0.2582 in the second.

Wording check covers returned chunks. The check that none of paid time off, time off, vacation, PTO, holiday, days off or paid leave appears covered the six returned chunks, not every document in the set.

Tested by Vivek Tomar · evidence dated 21 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
RAGFlow
Build or version
v0.26.4
Plan or tier
Starter
Surface
API
Set up before the run
The source set used one term and left the alternate term out. We asked the same idea using different wording through the retrieval endpoint.
Retrieval settings
Only the question and dataset IDs sent; no reranker or threshold overrides
Tested
21 October 2026 · Vivek Tomar

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.

LevelNameScope
BenchmarkManaged RAG →15 scenarios · 9 tools
CapabilityRetrieval →
ScenarioThe request uses different wording than the source →
ToolRAGFlow →

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
9 November 2026First publishedManaged RAG 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/managed-rag/results/ragflow/paraphrased-request · 1 pass · 0 fail · coverage 1/1 · graded 2026-10-21

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.

Output file (text) ↗text/plain; charset=utf-8 · 14 KB
Proof 1 · Screenshot ↗image/png · 212 KB
File fingerprints (SHA-256)

A fingerprint identifies the exact file used for this result.

Output file (text)8202e14ab3c024ec8c0c17d876b90190d8ea4f656f776857693eeee85670a081
Proof 1 · Screenshot880eee9f2125efdd17062267b72dcbb6f4747570315cab27a5e1b97175eb2661