Our methodology

The five measurement layers and where each one can be wrong

The scoring model used by TalentTuner operates across five distinct layers. Each layer has a defined scope, a measurement approach, and a set of conditions under which it produces less reliable output. Knowing those boundaries is part of understanding what the score means.

The five-layer structure is documented in full on the algorithm page and in the ATS whitepaper match model section.

The match score is a relative ranking tool, not an absolute threshold

A TalentTuner score of 72 does not mean "72 percent of the requirements are met." It means "this resume, against this job description, is positioned better than a resume scoring 61 and worse than one scoring 84." The score is most useful when comparing multiple resume versions against the same job description, or when tracking improvement after specific edits.

What the score cannot account for: employer-specific ATS configurations (a Workday instance configured with a 60-point knockout threshold will behave differently from one with no automated filtering), the weight the hiring manager places on any one requirement, and factors that never appear in a job description (internal referrals, team composition, budget changes).

The detailed mechanics of the scoring pipeline, including the weighting formula for each layer, are documented in the whitepaper's ATS match model section. If you want to challenge or replicate any part of the methodology, that is the right starting point.

How we handle edge cases in the scoring model

After more than 50,000 resume analyses through the TalentTuner platform, a predictable set of edge cases has surfaced. The table below documents the most common, what the model does, and what the output should be understood to mean.

A note on career changers and industry-pivot candidates. The intent-fit layer is the one most likely to penalise a genuine but unconventional candidate. If you are pivoting industries and your resume scores low on intent fit, the score is not wrong — it is reflecting that your current resume does not yet signal fluency in the target industry's vocabulary. The fix is to mirror that vocabulary explicitly, which is what the match model documentation and the algorithm walk-through are designed to help you do.

What TalentTuner does not measure, and why that matters

The honest version of what TalentTuner optimises for: the algorithmic screening stage of a competitive application process. If your resume does not pass automated keyword filtering, everything else is moot. We address that specific problem well. We do not address everything that follows.

How the ATS platform model data is structured and updated

The Workday, Oracle Taleo, Greenhouse, and Lever simulators on TalentTuner are built from three types of input: the vendors' own published documentation (recruiter and administrator guides, API references where publicly available), observed behavior from controlled test applications through live employer ATS-powered portals, and documented changes tracked through vendor release notes and community reporting from HR practitioners.

ATS platform variance our simulators model — and what they cannot model

Platform	Primary filtering mechanism our model covers	What the model cannot account for
Workday	Screening question gate logic; candidate profile field mapping from resume upload	Employer-custom knockout question sets; pipeline automation rules each employer configures
Oracle Taleo	Requisition rank calculation using required vs. preferred criteria weighting	Employer-defined requisition configurations; custom fields added by enterprise customers
Greenhouse	Structured scorecard attribute alignment; what a strong scorecard-aligned application looks like from the candidate side	Individual interviewer scoring behaviors; pipeline stages that vary by employer
Lever	Boolean search query patterns; talent pool discoverability for sourcing workflows	Employer-specific tag taxonomies; nurture sequence configurations

The platform pages are updated when vendors publish major feature changes (typically quarterly) and when our controlled testing reveals behavior that differs from documented defaults. Each platform page carries a "last verified" date. If you notice outdated information, use the contact form to flag it.

Verdict: Our ATS platform simulators are accurate for the default behavior of each system, unreliable for any employer-specific configuration sitting on top of that default. Use them to understand the mechanics of the platform, not to predict what any specific employer's instance will do.

How TalentTuner's methodology compares to other resume scoring tools

For specific audiences: what to read and what to trust

If you are a journalist verifying our claims before citing us:

The factual claims we make in our content — the 50,000+ analysis figure, the ATS platform mechanics, the scoring pipeline — are documented on this page, the algorithm page, and the whitepaper. The 50,000+ figure refers to cumulative resume analyses run through the TalentTuner platform since launch and is based on internal usage data. We do not publish user-level data; that figure is verifiable to us but not independently auditable by a third party. The ATS platform claims are based on publicly available vendor documentation and our own controlled testing. We link to vendor documentation where we cite it. If a specific claim looks wrong, use the contact form and we will respond with sources or correct the content.

If you are a researcher considering our dataset for an academic study:

The 50,000+ analyses are not available as a public dataset due to privacy constraints (resumes contain identifying information). We have published aggregate findings and documented our methodology in the research hub and whitepaper. We are open to discussing research collaboration that does not require sharing individual resume data. Our keyword matching approach uses standard scikit-learn TF-IDF with spaCy 3.x; our content quality pass uses GPT-4 with the prompt documented in the whitepaper. Our ATS-platform testing methodology is described in the research hub. One important limitation to flag in any citation: our dataset overrepresents English-language, North American job applications. Findings should not be generalised to other labor markets without further study.

If you are a hiring manager wondering how our scoring would treat your candidates:

TalentTuner scores a resume against a specific job description — not against your company's requirements unless you paste your job description in. A candidate who scores 85 against your job description has a resume that uses the vocabulary of your JD well and presents achievement-oriented content. What TalentTuner cannot see: whether the candidate's claimed experience is verifiable, whether their compensation expectations fit your range, whether their work style matches the team, or whether they would pass your company's specific ATS configuration. The score is a shortlisting aid, not a hiring recommendation. We are explicit about this to candidates as well.

If you are a candidate trying to decide whether to trust our score:

Trust the gap analysis more than the number. The itemised list of missing keywords and under-represented terms is more actionable than the percentage. A score of 65 does not tell you much by itself; "missing: cloud architecture, Terraform, Kubernetes — three terms that appeared four or more times in the job description" tells you exactly what to fix. The score will rise when you address those gaps, but more importantly, your resume will be meaningfully closer to what the role requires. We recommend running the analysis, working through the specific gap list, and re-running to verify the changes landed. The comparison view makes that iteration easy.

Verdict: The gap list is the product. The score is a summary of the gap list. If you only look at the number, you are using the tool at half its value.

How TalentTuner Research handles opt-in outcome data

We do not require candidates to report interview outcomes, and we do not track application outcomes by default. When users voluntarily indicate that they received an interview or offer after using TalentTuner, we record that as opt-in feedback. This creates selection bias: users who have a positive outcome are more likely to report it. We do not use this data to make causal claims about TalentTuner's impact on hiring outcomes. We use it to identify patterns in what high-scoring, interview-successful resumes look like compared to resumes that were optimised but did not advance — and that comparison informs the recommendations the model generates.

Scenario	How we use the data	What it cannot tell us
User scores 80+ and reports an interview	Logged as positive feedback; contributes to pattern analysis of high-performing resume features	Causation (would they have gotten the interview without TalentTuner?); whether TalentTuner was the deciding factor
User scores 80+ and does not report an outcome	No outcome data collected; majority of analyses fall into this category	Whether they applied, what happened, whether the score was accurate
User scores low and asks why	Qualitative feedback used to improve gap list clarity and edge case handling documentation	Whether the low score correctly predicted a rejection

The full methodology disclosure — including the limits of our outcome data — is available in the ATS whitepaper. We believe tools that make claims about hiring outcomes have an obligation to disclose how those claims are measured.

Verdict: This methodology page is a binding commitment to what TalentTuner measures and does not measure. When the tool or the underlying model changes in a material way, this page is updated. The date at the top of this page reflects the most recent review.