Emits a blank Languages item when the resume has no languages section, leaving an empty placeholder instead of omitting the field.
What was measured
Noise in output
Does it add incorrect or hallucinated fields?
decisive for this rankingtransformation
Incorrect or hallucinated fields mean the resume parser is not producing reliable structured data, which is central to the job. (3 of 3 judges)
What was given, what came back
Test input: Messy real-world resume — John Kumar · pdf · group: resume-parsing
Input — what we sent
Research media screenshot 202026 05 05 20124544.png
Messy real-world resume — John Kumar
A poorly formatted, inconsistent real-world resume for John Kumar, used to test robustness against noisy structure, inconsistent dates, and mixed-content sections.
Why this input is hard
- · messy formatting robustness
- · section detection fallback
- · inconsistent date parsing
- · soft-skills extraction
- · noise and hallucination control
Output — unretouched
Also checked on this input — same tool, 5 other criteria
Accuracy◐ MixedPreserves the education marks text verbatim, including the unnormalised '72 percent marks' value instead of converting it to a cleaner percentage format.Accuracy◐ MixedReturns both certification names in lowercase exactly as written, with no capitalization normalisation applied.Field coverage✓ WorkedStill covers the main resume fields on the messy input: name, email, phone, experience, education, skills, and certifications.Messy resume handling✓ WorkedHandles the noisy, inconsistent layout without crashing and still extracts the main sections.Output quality✓ WorkedProduces a usable JSON result even on the messy resume, with the important sections still readable and structured.
Provenance
- Observation
- 7ec0595e-ec84-427a-a9c7-22c290258f01
- Evidence run
- cbbef4db-964c-49fa-a57f-a2977822bdfc
- Study
- Parse resumes into structured data using an API
- Research task
- 86b9jm30n
- Tested at
- not recorded
- Source
- first-party
- Evidence state
- verified
- Proof shown
- input + output shown
- Cost / latency
- not captured
- Repeat run
- not captured
- Tester
- not captured
The last three rows are honest blanks, not placeholders — our capture has no field for them yet.
Query this
get_evidence({
tool: "extracta-labs",
scenario: "resume-parsing"
})MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Same input, same check — 5 other tools
measured on Noise in output
Affinda✗ FailedThe skills output includes hallucinated taxonomy noise, such as Business Education, that does not appear in the resume.CVParserPro✗ FailedThe parser hallucinated unsupported month values in the work date range, expanding 2019–2021 into January 2019 to December 2021.Hireability✗ FailedThe certification field merged three content types — certifications, skills, and references — into one extracted value instead of separating them.HrFlow◐ MixedThe messy parse can bleed location text into a task item, prepending 'Pune' to one extracted responsibility.Skima AI✗ FailedThe messy-resume skills block was collapsed into a single concatenated string with no separators, making the field hard to reuse downstream.
This evidence is published in
Real inputs and real outputs, no retouching · every cell queryable via API & MCP · aidemos.com