← Back to home

Scoring Methodology

v1.0

Last updated: July 2026

Your Skill Vector match score is deterministic: the same resume and job description always produce the same number. It is computed by a pure-Python engine — AI (Claude) never influences the score. This page documents exactly how it works so you can verify every point.

Inputs

Two texts: your resume and a target job description. Both are cleaned and checked for minimum substance before scoring. Nothing else affects the number — not your name, not the time of day, not any AI model.

Signal 1 — Skill Match

The engine extracts the concrete skills a job requires (tools, languages, certifications) from a curated taxonomy of 360+ skills and their aliases, then measures what share of them your resume demonstrates. “JS” and “JavaScript” count as the same skill.

Signal 2 — Semantic Similarity

Both documents are converted into multilingual sentence embeddings, and the engine measures how close their meaning is. This credits relevant experience even when the exact words differ — “built predictive models” relates to “machine learning” without a literal keyword match.

How the two combine

The final score is a weighted blend of the two signals, clamped to 0–100:

score = (skill_weight × Skill Match) + (semantic_weight × Semantic Similarity)
Job languageSkill MatchSemantic Similarity
English55%45%
Non-English60%40%

Language adaptation

For non-English job descriptions the skill taxonomy detects fewer skills, so the engine weights the skill signal slightly higher and applies a confidence adjustment to avoid unfairly penalizing candidates whose real coverage exceeds what the taxonomy can detect. This is why the weights differ by language — it is deliberate, published, and consistent.

What AI does

Claude powers the optional improvement tools: rewriting bullet points and full resumes, generating cover letters, career coaching, and mock-interview scoring. These are creative aids you choose to use.

What AI never does

AI never computes, adjusts, or influences your match score, your skill gaps, or your usage limits. Those are 100% deterministic engine outputs. If AI is unavailable, your score is unaffected.

Known limitations

The score reflects text-based signal only — it cannot see design, PDF layout rendering, or achievements you did not write down. Skill detection depends on the taxonomy, which is continually expanded but not exhaustive. The score is a decision aid, not a hiring guarantee.

Version history

  • v1.0 (Jul 2026) — Two-signal deterministic score published: Skill Match + Semantic Similarity, language-adaptive weights.

Questions about the methodology? The engine is documented in the open repository.