> ## Documentation Index > Fetch the complete documentation index at: https://nova.dweet.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Scoring Methodology > How Nova uses a structured, evidence-based workflow to screen candidates consistently. Nova scoring is a deliberate, multi-step workflow. Nova defines the role before it judges any candidate, holds every candidate to the same standard, and ties each conclusion back to evidence. A person stays in control of hiring decisions. ## Why Nova Separates The Steps "Job description plus resume, return a score" looks simple, but it asks a model to do several jobs at once: interpret an inconsistently written job description, invent the criteria, decide what matters most, find or infer the evidence, resolve what's missing, and reach a verdict. When all of that happens in one step, the effective standard can drift from one candidate to the next. Nova separates those decisions into stages, so it sets the standard once and applies it consistently. A structured interview asks every candidate the same planned questions and judges them against the same standard. Nova applies the same discipline to screening: it settles the role and the criteria first, then assesses every candidate against them. ## The Scoring Workflow When the integration supports it, Nova asks targeted questions about priorities, flexibility, deal-breakers, and trade-offs before writing any criteria, so role nuances are resolved up front. Where the job description or intake notes already answer these clearly, Nova skips that pass. Nova drafts the criteria from the role and your answers, applying built-in best practices for criteria that score well: each one stays resume-verifiable, grounded in the actual role, and marked **Must have**, **Preferred**, or **Nice to have**. Your team reviews and edits a strong starting draft, which turns an inconsistently written job description into one consistent standard. See [Configuring Scoring Criteria](/nova/features/configuring-criteria). Nova fixes the scoring standard for the role and reuses it for every candidate, so Nova holds the 200th candidate to the same bar as the first. Nova marks each criterion Pass (supported), Partial (partly supported), or Fail (contradicted), then combines the evidence into one role-specific score and a written assessment. Missing evidence creates uncertainty; evidence pointing the other way creates a concern. See [Understanding Scores](/nova/features/understanding-scores). Anything left unresolved becomes an interview-focus point for your team to verify next. Interview-only traits a resume can't show don't lower the score, while a missing resume-verifiable requirement is treated as uncertainty that can affect it. ## Scoring Principles * **Importance guides judgment.** Must have, Preferred, and Nice to have tell Nova how seriously to treat a gap. There's no hidden points formula and no separate numeric weighting. The score reflects overall evidence alignment across all criteria. * **Hard gates stay hard.** Recruiters can set explicit exclusions that a candidate must clear before the other criteria are weighed together, so a real deal-breaker isn't outweighed by unrelated strengths. * **Absence means uncertainty.** Nova treats a requirement a resume doesn't mention as something to verify at interview. A missing Must have signal can still leave a strong candidate worth reviewing; only evidence that clearly contradicts a requirement creates a strong concern. * **Calibrated to seniority.** Nova scales how much evidence a criterion needs to the level of the role, so a senior title is held to a higher bar of demonstrated impact than an entry-level one. * **Evidence over keywords.** Relevant context can be the evidence. A senior role at a well-known team can demonstrate a capability without naming it, and equivalent experience can count when the criterion allows it. Nova doesn't require exact keyword matches. * **Reads the timeline.** Nova weighs the chronology of a resume, so a current disqualifying role counts as current and an outdated summary doesn't mask it. * **Interview-only traits stay out of the score.** Things a resume structurally can't show (soft skills, motivation, culture alignment) go to interview guidance and don't reduce the score. * **One standard for every candidate.** Nova assesses every candidate for a role against the same criteria and the same standard, which keeps scores comparable across a pipeline. The most important nuance for calibration: a resume that **contradicts** a requirement reads very differently from one that's simply **silent** on it. The first is a concern; the second is an interview question. ## How We Evaluate Scoring Nova puts its scoring prompts and pipeline through an offline LLM-as-a-judge evaluation before changes ship, so quality is measured systematically and at a scale beyond what a person could review by hand. A judge model grades scoring outputs against the candidate's resume and the role, using fixed rubrics: whether claims are grounded in the resume, whether missing information is handled as uncertainty, whether must-have requirements are read faithfully, whether the verdict and score stay consistent, and whether the interview guidance is specific and usable. Judge models can be swayed by surface signals like writing style, length, and tone, so the rubrics keep the grade on evidence and scoring behavior. When Nova compares scoring models, it uses a blinded comparison: the judge reviews their outputs as Model A and Model B. Regression tests cover known cases with expected score bounds, so Nova catches a change that would revert a fix it's already made. Nova chooses the model behind scoring by how well it performs on real screening tasks, then weighs cost and latency. Treat each score as a signal. Automated evaluation has limits, human review is still the gold standard, and your team makes the call. ## Keeping People In Control Nova helps your team decide what to review first. It doesn't make employment decisions. Criteria are explicit and editable, every conclusion carries its supporting evidence, and your team reviews that evidence before acting. For how Nova checks scoring patterns for adverse-impact signals, see [Bias Testing Methodology](/nova/legal/bias-testing). ## Frequently Asked Questions A single prompt would both define the standard and apply it at the same time, which lets the standard drift between candidates. Separating role calibration from candidate assessment means Nova measures every candidate against the same, pre-established criteria. No. Nova's scoring is a structured assessment against your criteria, with a supporting evidence trail. A person reviews the evidence and makes the hiring decision. There's no separate numeric weighting setup. Use Must have, Preferred, and Nice to have to express importance. This avoids a false-precision total and keeps the score an evidence-based judgment. When the resume is merely silent on a requirement, that becomes something to verify. A clearly contradicted Must have requirement is what drives a much lower score. Use Must have only for requirements the role genuinely can't work without. Nova produces stable scores for the same evidence and criteria. Minor variation can still occur, which is exactly why the evidence and the written assessment (not just the number) are what your team should review. No screening system can guarantee that. Nova is designed to keep criteria job-related and visible, ground conclusions in evidence, support review and monitoring, and keep accountable people in control. See [Bias Testing Methodology](/nova/legal/bias-testing) for how Nova checks scoring patterns. Write criteria that steer scoring Read and interpret Nova assessments How Nova checks for adverse-impact signals