ProvenATSApplicant Tracking System
HOW IT WORKS

How We Read a Candidate

Most tools give you a resume summary. We give you a complete psychographic profile, built from 22 dimensions of behavioral insight, validated across 4 analytical tiers, and delivered in under 60 seconds.

LinkedIn URL
Input
LinkedIn Read
Data Extraction
4-Tier Engine
22 Dimensions
Cognition Profile
Output
01
THE INPUT

Three data sources. One comprehensive picture.

We do not ask candidates to fill out questionnaires. We read what they have already written, their public professional narrative, and enrich it with optional technical portfolio data.

LinkedIn Profile

The primary signal. We read the static profile and the candidate's public activity, the things they have already chosen to put into the world.

Full work history with tenure dates and descriptions
Company details for every employer, not just the current one
Posts, comments, and reactions (where public)
Skills, endorsements, education, and certifications

GitHub Portfolio

Optional enrichment. We analyze public repositories for technical depth, consistency, collaboration patterns, and code quality signals.

Repository count and language diversity
Commit frequency and consistency
Documentation quality signals
Contribution patterns (solo vs. collaborative)

Job Description

The context layer. We parse the role requirements to weight dimensions and compute a job-specific fit score rather than a generic score.

Required skills and experience level
Team composition and reporting structure
Company culture indicators
Role-specific priority weighting

Optional inputs. Paste the candidate's resume to fill in sparse LinkedIn profiles (stored encrypted, used to augment the LinkedIn signal). Capture the candidate's email at analyze time so the rejection-email flow is one click later. Save interview Q&A pairs after a phone screen and an AI scorer returns a verdict, overall score, and per-question reasoning.

02
THE PIPELINE

Four analytical tiers. Twenty-two dimensions.

Each tier is a specialized inference agent running on a domain-specific psychological framework. Tier 1 gates confidence for everything downstream. No single call tries to do everything.

Tier 1,Claude Opus 4.7

Structural Analysis

The gatekeeper. Before any personality or behavioral inference happens, we establish the structural facts of the candidate's career. This tier runs first and controls confidence for everything downstream.

Career Transition State

Detects recent job changes, gaps, field switches, or layoffs. If active transition is detected, all downstream confidence is capped at Medium.

0 = Stable, 10 = Major Transition

Tenure Stability

Measures average tenure, job-hopping frequency, and progression consistency. Short tenures without advancement are penalized.

0 = Unstable, 10 = Highly Stable

Advancement Velocity

Tracks promotion speed, scope increases, and responsibility growth over time. Normalized by industry and role type.

0 = Stagnant, 10 = Rapid Advancement

Risk Appetite

Inferred from career choices: startup vs. big-co, established industry vs. emerging, steady progression vs. pivots.

0 = Risk-Averse, 10 = Risk-Seeking

How it works

1
Parse work history
Extract every job: title, company, dates, description text.
2
Compute tenure metrics
Average tenure, shortest tenure, tenure trend over time.
3
Detect transitions
Look for gaps > 3 months, title downgrades, field switches, recent start dates.
4
Score advancement
Compare title progression against timeline; normalize for industry norms.
5
Gate downstream
Output confidence modifier: High (no transition), Medium (minor transition), Low (major transition).
Example output
{
  "careerTransitionState": 1.2,
  "tenureStability": 7.8,
  "advancementVelocity": 6.5,
  "riskAppetite": 4.2,
  "transitionDetected": false,
  "downstreamConfidence": "high"
}
Tier 2,Claude Sonnet 4.6

HEXACO Personality

The HEXACO-PI-R framework1 measures six major dimensions of personality. We infer all six factors from the candidate's written narrative, job descriptions, recommendations, and self-reported achievements. Independent meta-analytic research finds the HEXACO model explains roughly 60% more variance in counterproductive work behavior than the legacy Big Five (31.97% vs 19.05%)5.

Honesty-Humility

sincerity, fairness, greed-avoidance, modesty

Low = manipulative, High = genuine

Emotionality

fearfulness, anxiety, dependence, sentimentality

Low = stoic, High = emotionally expressive

Extraversion

social self-esteem, social boldness, sociability, liveliness

Low = reserved, High = outgoing

Agreeableness

forgivingness, gentleness, flexibility, patience

Low = critical, High = accommodating

Conscientiousness

organization, diligence, perfectionism, prudence

Low = spontaneous, High = disciplined

Openness

aesthetic appreciation, inquisitiveness, creativity, unconventionality

Low = traditional, High = experimental

How it works

1
Extract behavioral text
Pull all job descriptions, achievement statements, and recommendations.
2
Prompt engineering
Each HEXACO factor has a dedicated prompt calibrated on validated item banks.
3
Score + evidence
Claude returns a 0-10 score and 1-10 direct text citations per factor.
4
Population norming
Scores are calibrated against HEXACO-PI-R population percentiles.
5
Confidence scoring
Confidence depends on text richness. Sparse profiles get Low confidence.
Evidence example
“Volunteered to lead the diversity initiative without being asked...” Honesty-Humility: 7.2
Tier 3,Claude Sonnet 4.6

Schwartz Values

Schwartz's theory of basic values2 identifies ten universal motivational dimensions organized in a circular structure. We infer the four value clusters most relevant to career outcomes: Self-Enhancement, Openness to Change, Self-Transcendence, and Conservation.

Self-Enhancement

Achievement, Power, drive for personal success, status, and influence

Low = content, High = ambitious

Openness to Change

Self-Direction, Stimulation, Hedonism, desire for novelty, autonomy, and creativity

Low = conventional, High = experimental

Self-Transcendence

Universalism, Benevolence, concern for others, environment, and social justice

Low = self-focused, High = altruistic

Conservation

Security, Conformity, Tradition, preference for stability, rules, and established ways

Low = rebellious, High = traditional

How it works

1
Value-laden text extraction
Identify statements expressing preferences, priorities, and motivations.
2
Cluster scoring
Score each of the four clusters on 0-10 using PVQ-inspired prompts.
3
Circumplex placement
Map the candidate onto the Schwartz circular value structure.
4
Career relevance weighting
Weight values by their documented predictive power for job performance and satisfaction.
Value circumplex
OpennessConservationSelf-Trans.Self-Enhance.
Tier 4,Mixed (Sonnet + Opus)

Behavioral & Vocational

The deepest tier. Eight specialized inference agents, each calibrated on a different psychological framework, analyze behavioral signals from the candidate's career narrative.

Adaptability

Resilience to change, learning velocity, pivot capability

Knowledge Sharing

Mentorship, documentation, teaching, community contribution

Network Quality

Connection depth, endorsements, recommendation sentiment

RIASEC Profile

Holland vocational interests (Holland 1997, ref. 3): Realistic, Investigative, Artistic, Social, Enterprising, Conventional

Career Anchors

Schein's 9 anchors: what truly drives career decisions

SDT Needs

Autonomy, Competence, Relatedness, intrinsic motivation drivers

Convergence

Cross-validation across 5 frameworks (Premium only)

Trajectory

Forward-looking career path forecast (Premium only)

How it works

1
Behavioral text mining
Extract achievement statements, project descriptions, and collaboration language.
2
Framework-specific scoring
Each agent runs its own prompt library calibrated on validated instruments.
3
Evidence extraction
Every score cites direct text evidence (12+ word sentences).
4
Cross-framework convergence
Premium: Opus reconciles results from HEXACO, Schwartz, RIASEC, Anchors, and SDT.
5
Trajectory modeling
Premium: Forecast likely career path based on velocity, values, and historical patterns.
Swarm panel (Premium)
Standard AssessorEvidence SkepticDevil's AdvocateEvidence Purist
03
THE OUTPUT

Not a score. A complete picture.

Every analysis produces a structured Cognition Profile, a JSON document with 22 dimension scores, confidence levels, evidence citations, and a natural-language narrative. Here is what you get.

Dimension Scores

Every dimension is scored 0-10 with a confidence level (high / medium / low). Scores are calibrated against population norms, not arbitrary thresholds.

Career Transition State2.3
Tenure Stability7.8
Honesty-Humility6.5
Openness8.1
Achievement7.2
Adaptability6.9

Radar Chart

A polar visualization of all scored dimensions on a single 0-10 scale. Instant pattern recognition, see the candidate's shape at a glance. Compare two candidates side-by-side with dual-overlay mode.

Natural Language Narrative

A generated paragraph that reads like a senior recruiter wrote it, summarizing strengths, risks, trajectory, and cultural fit in plain English. Every claim is backed by cited evidence from the source text.

Job Fit Score

A weighted composite score (0-100) computed against the specific job description. You control the weights: Execution, Collaboration, Adaptability, Integrity.

78
Overall Fit
High
Confidence

Interview Questions

AI-generated behavioral interview questions tailored to the candidate's profile and the role. Probe their weakest dimensions. Validate their strongest claims.

“Tell me about a time you had to adapt quickly when a project direction changed unexpectedly.”
“How do you balance individual contribution with team collaboration?”

Trajectory Forecast

Premium tier only. A forward-looking projection of the candidate's likely career path based on velocity, values alignment, and historical patterns in similar profiles.

After the profile generates

Every analysis joins a persistent candidate library you can tag, filter, and re-score against new job descriptions in one pass. The downstream workflow is built into the same view:

Side-by-side comparison. Stack 2 to 6 candidates against an overlay radar with a per-dimension score table. Set minimum dimension targets per role and the table marks who meets the bar.
Role-specific weighting. Tune tier weights globally in Settings or override them per job. Rankings recompute on read, so changing the dial updates results without re-running inference.
Interview Q&A scoring. Save real interview questions and answers per candidate. The AI scorer returns a verdict (advance / borderline / pass), an overall score, and per-question reasoning. The verdict shows up on the comparison view.
One-click rejection. Capture the candidate's email at analyze time (encrypted at rest). Editable rejection-email template per account. One click sends with a delivery audit trail.
04
LIVE ASSESSMENT

Read them from LinkedIn. Then watch them work.

The Cognition Profile tells you who a candidate is. The Live Assessment shows you how they actually think. Send a token-invited link, generate a problem tailored to the role, and observe their full session in a browser-native coding environment. No scheduling. No installs. No screen-share theater.

What the candidate sees

A Monaco-powered code editor in the browser, no install required
A real xterm terminal backed by an in-browser WebContainer runtime
An AI chat assistant they can use openly, we score how they use it
A problem generated from the role and their profile, not a leetcode dump

What you receive

A full session replay: every keystroke, command, and AI exchange
A two-pass review: process review (how they worked) + artifact review (what they built)
A weighted hiring recommendation with strengths, risks, and confidence
Tokens are time-limited and scoped to one candidate plus one job

Nine evaluation dimensions

The session is graded by an AI rubric calibrated for the AI-assisted era of engineering. We do not penalize candidates for using the assistant, we score the quality of how they use it. Each dimension returns a rating (exceptional / strong / solid / mixed / weak), evidence, and a confidence level.

Problem Framing10%
Task Decomposition10%
AI Usage Quality10%
Debugging Resilience15%
Verification Discipline15%
Communication & Reflection5%
Code Quality10%
Judgment & Tradeoffs10%
Functional Outcome15%
05
SCORING METHODOLOGY

Evidence-based. Not guesswork.

Every dimension score is backed by direct evidence from the candidate's profile text. No score is invented. No inference is made from absence of evidence. Structured assessment of work-relevant traits has been shown to outperform unstructured interviews by more than 2x in predictive validity (r=.42 vs r=.19)4.

Confidence Levels

Not all dimensions can be scored with equal certainty. We report confidence honestly.

HighMultiple converging indicators, strong direct evidence
MediumSome direct evidence, some inference
LowWeak evidence, high inference, or data scarcity

Evidence Arrays

Every score comes with 1-10 direct text citations from the candidate's profile. You can verify every claim against the source.

“Led a team of 12 engineers through a complete platform rebuild in 8 months...”

Career Transition Gate

Tier 1 detects if the candidate is in a career transition (recent layoff, gap, new field). When detected, confidence is downgraded across all dimensions because behavioral signals are less stable during transition periods.

Transition detected, confidence capped at Medium

Convergence Checks

Premium tier only. Five independent psychological frameworks are cross-validated: HEXACO, Schwartz Values, RIASEC, Career Anchors, and SDT. If they contradict, the profile is flagged for human review.

HEXACOSchwartzRIASECAnchorsSDT
06
SECURITY & PRIVACY

Your data. Their privacy. Protected.

We built ProvenATS with a privacy-first architecture. Raw candidate data never persists in plaintext. Every access is scoped. Every operation is auditable.

AES-256 Encryption

Sensitive fields (LinkedIn URLs, ATS credentials, API keys) are encrypted at the application layer before storage.

PII-Conscious Pipeline

Raw candidate text is processed in-memory during analysis and never persisted. Identifiers are HMAC-SHA256 hashes.

Row-Level Security

Every database table has a userId column. PostgreSQL RLS policies ensure users can only access their own data.

SOC 2 Ready

Infrastructure, access controls, and audit logging are designed to meet SOC 2 Type II requirements.

07
THE MODELS

Right model. Right tier. Right cost.

We do not use one model for everything. Each tier uses the model best suited to its task, balancing reasoning depth, speed, and cost.

Claude Opus 4.7

claude-opus-4-7
Tier 1 (Structural) + Convergence

Our deepest reasoning model. Used for structural analysis that gates all downstream confidence, and for premium-tier convergence checks that reconcile five psychological frameworks.

Max output: 8,192 tokens
Streaming: required
Sampling params: none (deprecated)

Claude Sonnet 4.6

claude-sonnet-4-6
Tier 2-4 (Personality, Values, Behavioral)

The workhorse. Fast, capable, and cost-efficient. Handles HEXACO personality inference, Schwartz values clustering, RIASEC profiling, and most behavioral dimensions.

Max output: 8,192 tokens
Streaming: required
Sampling params: none (deprecated)

Claude Haiku 4.5

claude-haiku-4-5-20251001
Lightweight tasks + contrarian roles

Our fastest model. Used in swarm panel configurations for contrarian/evidence-skeptic roles, and for any lightweight classification or extraction tasks.

Max output: 8,192 tokens
Streaming: required
Sampling params: none (deprecated)

Ready to see it in action?

Create a free account and run your first analysis in minutes. No credit card, no waiting.

SOURCES
  1. 1. Ashton, M. C., & Lee, K. (2007). Empirical, Theoretical, and Practical Advantages of the HEXACO Model of Personality Structure. Personality and Social Psychology Review, 11(2), 150-166. journals.sagepub.com
  2. 2. Schwartz, S. H. (1992). Universals in the Content and Structure of Values: Theoretical Advances and Empirical Tests in 20 Countries. Advances in Experimental Social Psychology, 25, 1-65. psycnet.apa.org
  3. 3. Holland, J. L. (1997). Making Vocational Choices: A Theory of Vocational Personalities and Work Environments (3rd ed.). Psychological Assessment Resources. psycnet.apa.org
  4. 4. Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting Meta-Analytic Estimates of Validity in Personnel Selection: Addressing Systematic Overcorrection for Restriction of Range. Journal of Applied Psychology, 107(11), 2040-2068. doi.org/10.1037/apl0000994
  5. 5. Pletzer, J. L., Bentvelzen, M., Oostrom, J. K., & de Vries, R. E. (2019). A Meta-Analysis of the Relations Between Personality and Workplace Deviance: Big Five versus HEXACO. Journal of Vocational Behavior, 112, 369-383. doi.org/10.1016/j.jvb.2019.04.004