HOW NOETRIX ATS SCORING WORKS
Full algorithmic transparency. We reject black-box LLM scoring in favor of deterministic heuristics, placement weighting, and mathematically reproducible rules.
THE 6 EVALUATION DIMENSIONS
Extracts target technologies and domain terms from the job description and matches them using whole-word regex boundaries and canonical synonym graphs (e.g. K8s = Kubernetes).
Encodes job responsibilities and resume work experience into 384-dimensional dense embeddings using all-MiniLM-L6-v2. Evaluates overall narrative alignment even when different vocabulary is used.
Compares candidate career tenure against the job description's minimum required years. Combines tenure alignment with a density scan of 63 verified high-impact action verbs.
Evaluates degree hierarchy level (PhD, Master's, Bachelor's) and field of study alignment. Institution prestige is strictly ignored to eliminate socioeconomic bias.
Scans bullet points for verifiable performance impact using 10 regex filters: percentages, dollar values, multipliers (3x), time savings, and user/request scale indicators.
Checks for single-column linear layout, standard headings, valid email/phone contact information, and word length bounds (150 - 1500 words). Deducts points for formatting hazards like multi-columns and non-text tables.
THE MUST-HAVE KNOCKOUT CAP
If a recruiter marks specific technical requirements as mandatory must-haves, matching fewer than half (<50%) of them automatically caps the candidate's overall score at 50 points and classifies them in the 'weak' match band. This soft knockout ensures candidates lacking core qualifications never reach the top ranks while preserving their record for human review.
FAIRNESS & ADVISORY CHARTER
All candidate PII (gender, age, photo, marital status, nationality) is stripped before scoring. Education does not favor Ivy League or elite institutions. Ranking is always strictly advisory; human recruiters maintain final hiring authority.