N
NEXPERT
Version 2.0 MVP2 Specification

Trust Intelligence Methodology

How Nexpert objectively measures expert credibility, evaluates behavioral outcomes, and prevents reputation gaming through multi-signal mathematical aggregation.

Section 01

1. Trust Philosophy & Core Formula

Conventional platforms rely on static 5-star ratings, which suffer from severe inflation (94% of ratings cluster between 4.8 and 5.0) and vulnerability to review collusion. Nexpert replaces subjective star reviews with an evidence-based behavioral framework.

Trust Score = α · Signal_Current + (1 - α) · Score_Historical

Where α = 0.3, allowing an expert's score to evolve dynamically while maintaining long-term stability and resilience against single-session anomalies.

Section 02

2. Exponential Moving Average (EMA) & Signal Decay

Trust is temporal. An expert who delivered exceptional outcomes two years ago but is currently disengaged should not maintain identical standing to an actively engaged practitioner.

Recent Signals (0–90 Days)

Weighted at 100% full value to reflect active demonstrated capability.

Aged Signals (>90 Days)

Weighted at 60% decay value to encourage continuous consistency over time.

Section 03

3. The 4 Behavioral Dimensions

Every learner-expert interaction emits raw behavioral events analyzed by specialized AI agents across four orthogonal dimensions:

Structure (25%)

Clarity of curriculum, adherence to session agendas, actionable milestones, and resource preparedness.

Outcome (35%)

Tangible learner progress, verifiable certifications, career transitions, and goal milestone completion.

Boundary (20%)

Punctuality, respectful communication, scope compliance, and ethical interaction standards.

Consistency (20%)

Reliability across multiple cohorts, low cancellation rates, and sustained quality over repeat sessions.

Section 04

4. 5-Tier Trust Bands

Band Name Score Range Minimum Confidence Significance
Sovereign 90.0 – 100.0 70% Industry-defining mastery and highest verified outcome rate.
Established 75.0 – 89.9 50% Proven track record with consistent learner success.
Verified 60.0 – 74.9 30% Validated baseline quality and verified credentials.
Emerging 40.0 – 59.9 10% New experts building an initial portfolio of evidence.
Unverified 0.0 – 39.9 0% Pending onboarding or insufficient interaction volume.
Section 05

5. Confidence & Stability Scoring

A score with 2 completed sessions carries different statistical weight than a score with 100 sessions. Nexpert calculates real data completeness:

Confidence Score (%): Measures signal density relative to the target expected signals (target = 20 signals for 100% confidence).

Stability Score (%): Computes historical variance across snapshot history. High stability indicates low volatility.

Trend Direction: Computes the trajectory (rising, stable, or declining) based on recent score deltas.

Section 06

6. Anti-Gaming Protections & Freeze State

To prevent malicious rating manipulation, sybil attacks, or sudden performance irregularities:

  • Freeze State (is_frozen = 1): Temporarily locks an expert's score calculation during audit reviews or active dispute investigation.
  • Sybil Detection: Repeat bookings from identical IPs or linked payment accounts have diminished signal weighting.
  • Deterministic Fallback: If third-party AI APIs become unresponsive, deterministic fallback scoring ensures uninterrupted platform stability.
Section 07

7. Outcome Validation Protocol

Outcomes represent the core pillar of educational trust. Nexpert validates outcomes through a tripartite protocol:

1. Self-Report

Learner submits evidence URL or milestone completion proof.

2. AI Verification

Automated parsing of artifact links and session transcripts.

3. Admin Audit

Human-in-the-loop review for high-impact certifications.

Section 08

8. Version & Changelog

Version Release Date Key Additions & Changes Status
v2.0 MVP2 August 2026 5-band mapping, 90-day signal decay, trust notification queue, real confidence & stability scoring. CURRENT
v1.0 MVP1 May 2026 Initial 3-tier EMA aggregator with basic session signal extraction. SUPERSEDED