Evidence-Grounded Intelligence

-- Candidates Evaluated · -- Verified Profiles

Cross-referencing technical claims across resumes, GitHub repositories, and interview transcripts with strict zero-hallucination verification.

Zero Hallucination

Candidate	Target Role	Role Fit	Consistency	Quadrant	Discrepancies	Action

Loading Candidate...

[REVIEW REQUIRED]

Role • Applicant

Evaluating candidate evidence and cross-checking source claims...

Role Fit Score -- 60% LLM + 40% Rubric

Evidence Consistency -- Cross-source agreement

Unsupported Claims 0 Unverified assertions

Contradictions 0 Direct conflict in sources

Role Fit Formula & Calibration Insight Calibrated

60% Weight LLM Requirement Fit --

40% Weight Resume Rubric Score --

Final Result Role Fit Blended --

Role Fit Quadrant Cutoff (Threshold: 72.0) -- / 100

Consistency Cutoff (Threshold: 70.0) -- / 100

2D Assessment Quadrant Fit (X) vs Consistency (Y)

[INSUFFICIENT EVIDENCE] Low Fit, High Consistency

[STRONG MATCH] High Fit, High Consistency

[WEAK MATCH] Low Fit, Low Consistency

[REVIEW REQUIRED] High Fit, Low Consistency

Terminal Assessment Report

Loading report...

Candidate Career & Evidence Timeline (3D Spline) Corroborated Trajectory

Scrub Trajectory: --

Corroborated Milestone Discrepancy / Attention Click milestone card or scrub timeline to inspect

Cross-Source Discrepancies & Contradictions

Trace conflicts between claimed experience and corroborating artifacts. Click any source citation to inspect raw evidence.

Cross-Source Verification Graph Corroborated Topology

Multi-Source Claims Network

Verified Link Discrepancy / Attention

Multi-Document Evidence Reader Select source

Loading document...

Priority Interview Probe Questions

Targeted questions for technical interviewers to probe flagged contradictions and evidence gaps.

Deterministic Resume Rubric Breakdown (Baseline A) --

Candidate EEOC Fairness & Defensibility Record Audited Cohort

Evaluation Engine: Local Ollama Engine

Prompt Specification: v2.4-grounded-json

Defensibility Standard: EEOC 4 CFR Part 60 (1.000 Disparate Impact)

Execution Mode: $0.00 Offline Zero-Cost

Sort by:

Showing candidates

Candidate Name ↕	Target Role	Role Fit ↕	Consistency ↕	Decision Quadrant	Unsupported ↕	Action

Algorithmic Fairness & Parity 100% EEOC Compliant

Audited under EEOC Uniform Guidelines on Employee Selection Procedures (4 CFR Part 60) across 44 counterfactual demographic mutations and 11 protected classes.

Mean Role Fit Drift (Δ):—

Mean Consistency Drift (Δ):—

Minimum Disparate Impact Ratio:—

Quadrant Stability:—

Evidence Grounding & Claim Critic Zero Hallucination

Every atomic assertion is audited by the Claim Critic with exact substring containment in raw source documents.

Grounded Claim Fidelity:—

Citation ID Validity:—

Exact Quote Containment:—

Synthesized Inferences Marked:—

*Exact quote containment applies strictly to grounded claims. Remaining inferences are explicitly demarcated with zero hallucinated citations.

Model Calibration & Spearman Consensus Loading…

Evaluates prediction probabilistic alignment, ranking agreement against expert consensus, and contradiction detection efficacy.

Brier Score:—

Expected Calibration Error (ECE):—

Spearman ρ vs Expert Consensus:—

Spearman 95% Bootstrap CI:—

Contradiction Detection F1:—

Contradiction Recall / Precision:—

Statistical Rigor Disclosure: Spearman correlation (ρ = 0.816) statistically excludes 0 via 95% CI [0.446, 0.983]. ECE (0.2395) and Brier (0.2281) indicate under-calibration. Perfect contradiction F1=1.000 reflects a curated synthetic benchmark subset with injected discrepancies.

Adversarial Robustness & Red-Teaming 100% Defended

Automated adversarial stress-testing evaluating defense rates against prompt injection, schema smuggling, and synthetic tenure fabrication.

Prompt Injection Defense Rate:—

Fabrication Detection Recall:—

Tested Adversarial Vectors:

direct prompt injection hidden comment injection interview jailbreak

Structured CV Extraction Accuracy (LongExtractBench) Loading…

Deterministic scoring of local schema-constrained CV extraction against hand-labeled ground-truth records.

Matched Leaf Accuracy:—

Document Completion Rate:—

Array Row Recall / Precision:—

Layout Parser Architecture:Docling Layout-Aware + Fallback

Reviewer Time Efficiency & Latency Loading…

Estimated reviewer effort under standardized cognitive-load model (2,200 words @ 220 wpm + reconciliation).

Human Review Time per Candidate:—

Cognitive Efficiency Gain:—

Pipeline Concurrent Latency (p50):—

Job Requirement Cache Strategy:SQLite Hoisted (1 LLM call / Req)

Governance & Human-in-the-Loop Verdict HITL Mandatory

EEOC Compliance Gate

Compliance Policy Constraint

Autonomous Hire Verdicts: Not Permitted (False)

Human-In-The-Loop: Mandatory (True)

Recommendation Contract:

Proceed to human review with priority questions

Zero-autonomy design: Automated metrics serve as decision-support audits only, never making autonomous employment decisions.

Benchmark Evaluation Matrix

Loading benchmark metrics...

-- Candidates Evaluated · -- Verified Profiles

Loading Candidate...

Requisition Leaderboard

Audit & EEOC Governance

Candidate Name