What the score measures, and where it must stop.

Application ROI

The top-line score runs from 0 to 100. It estimates how much usable decision information the posting provides relative to the time and risk normally involved in an application process. The score starts from a neutral baseline and is adjusted by findings across four dimensions.

Verdict bands

85 to 100: Worth applying

The posting provides strong decision information and contains no major text-based contradictions. The candidate should still verify the team, manager, funding and process.

70 to 84: Promising, but ask first

The role may be worthwhile, but the report found gaps that should be resolved before a long process, take-home assignment or compensation discussion.

50 to 69: Low-information role

Important decision fields are missing or unclear. A short recruiter screen may be reasonable, but the candidate should avoid substantial unpaid work until the gaps are answered.

30 to 49: High-risk application

Several signals suggest a poor return on application time. The recommendation focuses on the smallest set of facts that could materially change the decision.

0 to 29: Likely time sink

The text contains severe contradictions, major transparency gaps or scam-like requests. This is not proof that the employer is illegitimate, but it is a reason not to proceed without verification.

Findings and severity

Each finding has a severity, an explanation, evidence when a matching phrase exists, and a next move. Severity describes its effect on candidate decision quality, not moral judgment about the employer.

• High: likely to materially affect compensation, workload, safety, seniority or time investment.
• Medium: important ambiguity that should be checked before the process becomes costly.
• Low: missing context or weak language that reduces clarity but may be easy to resolve.

Confidence

Confidence is based mainly on the length and specificity of the pasted text. A short fragment can only support a limited review. A longer description can support more checks, but length does not prove truthfulness.

Hiring credibility

Hiring credibility is deliberately not scored from description text. A useful credibility model would require external evidence such as:

• the posting on an official company domain;
• approved or recently confirmed headcount;
• posting age and meaningful repost history;
• recent hires into comparable roles;
• candidate-reported response and interview outcomes;
• consistency between posted and discussed compensation;
• verified company identity and contact domain.

Until those sources are integrated and quality-controlled, the product displays “insufficient evidence.”

Salary handling

Version 1.0 parses common salary formats and assesses whether the employer published a usable range. It does not estimate market pay from memory or static constants. A production market benchmark should include location, employment type, level, currency, data date, sample size and confidence.

Known limitations

• Rule-based language checks can miss context, irony and unusual formatting.
• Some ATS pages expose incomplete text to the extension extractor.
• Keyword presence does not guarantee that the stated policy is accurate.
• A score should never replace direct questions, identity checks or professional judgment.
• The tool is designed for software and adjacent technical roles. Other occupations may require different risk models.

Versioning and review

The bundled engine identifies itself as version 1.0. Changes to weights, rules or verdict bands should be documented, tested against a fixed corpus and released with a new engine version. The project includes a test plan and automated engine tests.