## TL;DR
Candidates download resume templates and forget to delete the example text, and parsers extract "M.S. Computer Science, 2019" as a real degree. Keep a known-placeholder list, detect template fingerprints like "Sample Resume" watermarks, and strip example blocks before extraction. Anything matching the placeholder list gets zero confidence and a verify flag.

## The query

```text
parser agent credited a candidate with a master's degree from the resume template's example text - the placeholder 'M.S. Computer Science, 2019' was never stripped
```

## Use this when

- Parsed resumes contain degrees, employers, or names the candidate never had
- Multiple resumes from the same template all show identical credentials
- Education claims look suspiciously round or generic

## Not for

- OCR misreads of genuine resume text
- ATS sync or API errors
- Interview scheduling problems

## Steps

### 1. Build a known-placeholder list

Collect common template strings: "M.S. Computer Science, 2019", "John Doe", "123-456-7890", "lorem ipsum", example email addresses, and "Example University". Match against it at high similarity, not just exact equality.

Expected output: placeholder text gets flagged in the parse before extraction runs.

### 2. Detect template fingerprints and strip example blocks

Look for template markers: "Sample Resume" watermarks, template vendor footers, unmodified example section headers. Drop example blocks wholesale.

Expected output: example text never reaches the extraction stage.

### 3. Sanity-check claims against each other

A 2019 master's from "Example University" plus a 2015-2018 bachelor's from the same school should trip a template smell. Cross-claim consistency catches what single-field checks miss.

Expected output: suspicious combinations flagged for human review.

### 4. Confidence-gate education claims

Anything matching the placeholder list at high similarity gets confidence zero and a "verify" flag. It can still be shown to a recruiter, but never silently.

Expected output: no example degree ever enters the candidate record as fact.

### 5. Regression-test against raw templates

Run your parser over the actual sample resumes shipped by popular template sites and confirm zero extracted credentials.

Expected output: a clean pass on every sample, re-run whenever the placeholder list changes.

## Variant phrasings

### resume parser extracted template example text

Same fix. Steps 1 and 2 are the whole game: know the placeholders, strip the templates.

### candidate credited with a sample degree

The education-specific version. Education claims deserve their own confidence gate (step 4) because a fake degree does more damage than a fake hobby.

## Why it happens

Template sites ship resumes with realistic example content, and candidates upload them half-edited. Parsers that extract first and validate never see a placeholder degree as anything but a fact, because nothing in the pipeline knows what template boilerplate looks like.

## Edge cases

- Candidates with genuinely generic names: "John Smith" is a placeholder in one resume and a real person in another. Use template fingerprints plus placeholder combos, never a single name match, before stripping.
- Real contact details that look placeholder-ish: validate against the email actually receiving messages before flagging.
- Non-English templates: maintain placeholder lists per language. English-only lists miss the same failure in Spanish, Hindi, or Portuguese templates.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_L2trUEWc7XNpArTkjTGBlw