Scientific Scoring and Traceability: The Hyrr Framework

In traditional recruitment, candidate assessment is often clouded by human subjectivity. Interviewers may inadvertently rate a candidate highly based on "gut feeling," confidence, or unconscious bias, rather than an objective technical demonstration. Unconstrained AI models can fall into similar traps, hallucinating high scores for articulate but technically shallow answers.

To solve this, the Hyrr AI platform employs a Scientific Scoring Framework that replaces black-box grading with a written standard. Every assessment is anchored in established industrial-organizational psychology models, and every score comes with its arithmetic and the evidence behind it.

This document outlines the architecture of Hyrr's scoring and traceability mechanisms, spanning from the core AI engine to the final Employer Dashboard.

---

1. The Scientific Assessment Pillars

The platform forces the AI to evaluate candidates not on general sentiment, but against three rigid academic pillars:

A. Behaviorally Anchored Rating Scale (BARS)

Instead of arbitrary 1-100 scores without context, Hyrr enforces grading based on specific, observable human behaviors. Each evaluated skill must map to a predefined behavioral tier:

1. Novice / Recall: Recites textbook definitions; struggles with practical application.
2. Competent / Applied: Applies the skill in standard scenarios; knows "what" but struggles with "why" or edge cases.
3. Proficient / Analytical: Deeply understands "why" and "how"; discusses trade-offs; handles complex troubleshooting.
4. Expert / Architectural: Demonstrates architectural foresight; sets best practices; anticipates system-wide impacts.

B. Bloom's Taxonomy

The AI assesses the conversational transcript to determine the highest level of cognitive engagement demonstrated for a specific skill:

• Remembering → Understanding → Applying → Analyzing → Evaluating → Creating.
• Example: A candidate simply defining a framework is capped at "Remembering," while a candidate critiquing its use in a distributed system achieves "Evaluating."

C. Mathematical Rubric Constraints

Skills are broken down into specific assessment criteria (e.g., Technical Communication, Problem Solving), each with an explicitly defined weight. The final score is not a guess; it is a strict weighted average of the sub-criteria, mathematically bound to the designated behavioral tier.

---

2. Platform Enforcement & Data Integrity

The core AI engine that processes interviews scores the evidence, not the person: name, accent, age, appearance and location are excluded from scoring by design. When an interview completes, the system does not just ask the LLM for a generalized summary or score. Instead, the AI is forced to provide specific, verifiable evidence:

• Behavioral Justification: The AI must explicitly quote the Behavioral Anchor that justifies its assigned tier.
• Cognitive Depth: The model must state the explicit taxonomy level reached.
• Complete Transparency in Scoring: The model is required to return the exact mathematical formula used to reach the final score (e.g., calculating the weighted averages of sub-criteria to reach an 88%).
• Quote-Based Verification: The AI is instructed to justify the score by explaining exactly how the candidate's words met that specific behavioral anchor, directly quoting the interview transcript.
• Specific Pillars: The AI must map the candidate's performance to the exact requirements of the competency that the candidate successfully proved.

By demanding these explicit proofs before finalizing an assessment, the system guarantees that no score exists without its underlying mathematical and behavioral justification.

---

3. UI Traceability Presentation (Employer Dashboard)

The theoretical framework and backend processing culminate in the Employer view, specifically within the Smart Skills Analysis dashboard. Here, Hyrr provides absolute transparency.

When an employer clicks to expand a specific assessed competency, they are not just shown a number; they are presented with the full chain of logic:

1. Direct Video Evidence: The system logs exact timestamps for when a skill was demonstrated. Employers can click a button to immediately seek the video recording to the exact second the candidate proved their competency, allowing for immediate human verification.
2. The Formula Unmasked: The exact mathematical formula is rendered directly on screen. Employers can read the exact math (e.g., criteria weights and sub-scores) that generated the final rating.
3. The Behavioral Proof: The cognitive level and behavioral tier are displayed alongside the verification summary. The employer reads exactly why the AI placed the candidate in their assigned tier, complete with verbatim transcript quotes.
4. Pillars Met: The dashboard lists the specific pillars of the skill that were demonstrated, proving the candidate didn't just talk broadly about a topic, but hit the key technical requirements defined by the employer's rubric.

Summary

By intertwining BARS, Bloom's Taxonomy, and Mathematical Rubrics into the core architecture, extracting those proofs via rigorous validation, and exposing the entire logic flow in the employer dashboard, Hyrr delivers a traceable, scientifically defensible assessment that holds every candidate to the same written guide. Employers never have to trust a "black box"—the proof is always in the math, the behavior, and the video.