Measurement for AI Frontiers

The measurement science
for AI impact.

Four frameworks. Three layers. One standard. VaryOn Amplitude quantifies what no one else measures - from data quality to agent trust to systemic risk.

Frameworks for the AI Economy

Data Layer

VaryOn Meridian

Data Quality

Is this data worth consuming, and what should an agent pay for it?

Meridian evaluates external data sources consumed by AI agents across four orthogonal dimensions, producing a composite score mapped to procurement tiers and dynamic pricing. Delivered in real time via MCP server integration during agent tool-call execution.

Dimensions

ScarcityQualityDecision ImpactDefensibility

Aggregation

Weighted Geometric Mean - Non-compensatory

Scale

0-100 -> Platinum / Gold / Silver / Bronze / Unrated

Agent Layer

VaryOn Drift

Alignment Impact

Is this agent still serving its principal’s intent?

Drift detects the invisible gap between what a human principal wants and what an agent actually does - especially across delegation chains where alignment degrades per hop. Its shadow principal detection acts as a multiplicative gate, identifying when third-party interests silently influence agent behavior and directly capping the maximum possible score.

Dimensions

Goal FidelityDelegation DegradationOverride AnalysisShadow Principal DetectionPreference Drift

Aggregation

Gated Geometric Mean - Shadow principal as multiplicative gate

Scale

0–100 → Aligned / Drifting / Misaligned

Ecosystem Layer

VaryOn Cascade

Systemic Impact

If something breaks, how far does the damage spread?

Cascade is the financial stress test for the agent economy. A single compromised agent can poison 87% of downstream decisions within 4 hours. Cascade runs Monte Carlo simulations on observed network topology to estimate propagation probability - the systemic risk measurement central banks are demanding.

Dimensions

Interconnection DensityCascade ProbabilityBehavioral CorrelationRecovery TimeConcentration Risk

Aggregation

Weighted Geometric Mean with Monte Carlo simulation

Scale

0–100 → Critical Risk / Elevated / Contained

Ecosystem Layer

VaryOn Convergence

Collusion Impact

Are autonomous agents colluding to manipulate market prices?

Convergence detects emergent algorithmic collusion and anti-competitive behavior in AI agent markets through statistical analysis of observable market outcomes. The framework identifies when autonomous AI agents converge on supra-competitive pricing equilibria - sustaining prices 200% or more above competitive levels - without any explicit communication or coordination protocol.

Dimensions

Price ConvergenceMarket DivisionCommunication AnalysisBid Pattern AnalysisConsumer Welfare

Aggregation

Minimum-of-Components - Non-compensatory

Scale

0–100 → Collusive / Competitive / Healthy

Frameworks in Development

Agent Layer

VaryOn Provenance

Identity Impact

Can you verify who this agent is, what it does, and who deployed it?

Provenance is the passport layer for autonomous agents. Before trust can be assessed, identity must be established. Provenance measures the verifiability, transparency, and completeness of an agent’s identity, capability claims, and operational history, enabling SOC 2-style certification for the agent economy.

Dimensions

Deployment VerificationCapability AttestationVersion IntegrityBehavioral HistoryTransparency

Aggregation

Weighted Arithmetic Mean - Partial identity adds value

Scale

0–100 → Certified / Provisional / Uncertified

Agent Layer

VaryOn Fidelity

Trust Impact

Can this agent be trusted to do what it claims?

Fidelity measures signal integrity - the credit score for autonomous systems. It scores whether an agent can be trusted based on its observable behavioral track record. Identity is measured by Provenance; Fidelity measures behavior exclusively: consistency, fulfillment, reputation, and anomalies.

Dimensions

Behavioral ConsistencyContract FulfillmentReputationAnomaly Freedom

Aggregation

Weighted Geometric Mean - Trust requires all dimensions

Scale

0–100 → Low / Moderate / High / Critical / Extreme Risk

Agent Layer

VaryOn Threshold

Resilience Impact

How resistant is this agent to adversarial attack and manipulation?

Threshold stress-tests agents against adversarial conditions. Where Fidelity measures past behavior (credit score), Threshold measures future resilience (stress test). Research shows 82.4% of LLMs succumb to peer-agent manipulation - Threshold quantifies exactly how resistant a specific agent is.

Dimensions

Prompt Injection ResistanceManipulation ResistanceData Poisoning ToleranceStress DegradationRecovery Time

Aggregation

Weighted Harmonic Mean - Weakest-link property

Scale

0–100 → Vulnerable / Resilient / Hardened

Agent Layer

VaryOn Parity

Fairness Impact

Is this agent treating all populations equitably?

Parity measures what no other framework captures: whether an agent’s decisions produce equitable outcomes across demographic groups. A hiring agent filtering out certain backgrounds, a pricing agent charging more based on inferred characteristics - these are Parity failures invisible to trust, alignment, or competition metrics.

Dimensions

Outcome DisparityTreatment ConsistencyProxy Variable AnalysisAccessibility CoverageEconomic Equity

Aggregation

Ceiling-Constrained Mean - Outcome disparity ceiling

Scale

0–100 → Inequitable / Fair / Equitable

Agent Layer

VaryOn Mandate

Human Oversight Impact

Can a human effectively intervene, override, or stop this agent?

Mandate quantifies whether human control over autonomous agents is real or ceremonial. EU AI Act Article 14 mandates human oversight; Mandate measures it. Each delegation hop adds latency between the human and the action - at some point, the human is nominally “in the loop” but functionally irrelevant.

Dimensions

Override EffectivenessIntervention LatencyVisibility DepthEngagement QualityEscalation Reliability

Aggregation

Multiplicative Chain - Every component must function simultaneously

Scale

0–100 → Ceremonial / Partial / Effective

Ecosystem Layer

VaryOn Yield

Economic Impact

Is value being created efficiently, or is friction destroying it?

Yield measures economic efficiency - the ratio of value created to value extracted in agent ecosystems. It detects when transaction costs consume value, when intermediaries extract excessive rents, and when misaligned incentives destroy welfare. Every basis point of friction compounds across millions of autonomous transactions.

Dimensions

Value Creation EfficiencyAllocative EfficiencyTransaction Cost RatioRent Extraction RateMarket Liquidity

Aggregation

Multiplicative Efficiency Model - Each inefficiency compounds

Scale

0–100 → Inefficient / Balanced / Optimal

Ecosystem Layer

VaryOn Lineage

Governance Impact

Who is accountable when autonomous agents cause harm?

Lineage traces accountability chains in AI agent ecosystems, mapping the flow of responsibility from actions to actors. It quantifies governance effectiveness, audit trail completeness, and liability attribution when autonomous systems create unintended consequences.

Dimensions

Audit Trail CompletenessResponsibility AttributionDecision TransparencyGovernance EffectivenessLiability Mapping

Aggregation

Weighted Arithmetic Mean - Accountability components

Scale

0–100 → Untraceable / Partial / Full Accountability

When frameworks combine, invisible patterns emerge.

Amplitude's power multiplies at the intersections.

Systemic Competition Risk

Market concentration in a dense network means anti-competitive behavior goes systemic.

Uncontrollable Failure

When humans can't intervene and failures propagate, the system is ungovernable.

Silent Misalignment

An agent drifting from intent while human oversight is ceremonial creates invisible risk.

Coordinated Discrimination

Fairness failures in a concentrated market amplify bias across the entire ecosystem.

Efficiency Crisis

Transaction friction compounds through interconnected networks, destroying value at systemic scale.

Data ROI

High-quality data inputs directly correlated with genuine economic efficiency.

Research

Our research teams build the measurement science for the AI economy - quantifying data quality, agent trust, alignment, resilience, fairness, and systemic risk.

Latest

Cross-Index Intelligence: How Patterns Across Frameworks Reveal What Single Scores Cannot

We present a systematic analysis of emergent intelligence patterns that arise when scores from multiple Amplitude frameworks are examined jointly. High Fidelity paired with low Drift reveals compliant but misaligned agents. Elevated Cascade risk alongside depressed Harmony signals fragile concentrated markets. These cross-framework patterns surface systemic insights invisible to any individual measurement instrument.

Data Science

Developing real-time data quality scoring methodologies that evaluate external data sources consumed by AI agents during inference-time operations.

Meridian

Agent Trust

Building the behavioral trust layer for autonomous agents - identity verification, behavioral consistency, and adversarial resilience testing.

Provenance, Fidelity, Threshold

Alignment & Control

Measuring the gap between human intent and agent behavior across delegation chains, and quantifying whether human oversight is real or ceremonial.

Drift, Mandate, Parity

Systemic Risk

Modeling failure propagation, market competition dynamics, and economic efficiency across interconnected agent ecosystems at scale.

Cascade, Convergence, Torque

Publications

DateCategoryTitle
Feb 24, 2026MethodologyCross-Index Intelligence: How Patterns Across Frameworks Reveal What Single Scores CannotFeb 22, 2026Alignment & ControlMapping AI Impact Measurement to Nine Regulatory FrameworksFeb 19, 2026MethodologySeventeen Corrections: Stress-Testing a Multi-Framework AI Measurement MethodologyFeb 15, 2026Agent TrustGaming Resistance in AI Measurement: Countermeasures Against Goodhart's Law Across Ten FrameworksFeb 10, 2026Alignment & ControlDelegation Degradation in Multi-Agent Chains: Quantifying Alignment Loss Per HopFeb 4, 2026Systemic RiskMeasuring Economic Efficiency in AI-Mediated Markets: Transaction Costs, Rent Extraction, and WelfareJan 30, 2026Systemic RiskThe Sherman Act for the Autonomous Age: Measuring Competition in Agent EconomiesJan 25, 2026Systemic RiskFinancial Stress Testing for Agent Networks: Monte Carlo Cascade Simulation with Spectral RobustnessJan 21, 2026Alignment & ControlQuantifying Human Oversight: From Ceremonial to Meaningful Control in Autonomous SystemsJan 17, 2026Alignment & ControlOutcome-Capped Fairness: A Ceiling-Constrained Measurement Framework for AI SystemsJan 13, 2026Agent TrustHarmonic Mean Aggregation for AI Security Assessment: Why Weakest-Link Properties MatterJan 9, 2026Alignment & ControlDetecting Shadow Principals in AI Agent Systems: A Correlation-Gated Alignment MeasurementJan 5, 2026MethodologyOn the Selection of Aggregation Functions for AI Impact MeasurementDec 22, 2025Agent TrustThe Credit Score for Autonomous Systems: Recursive Trust Scoring with Sybil ResistanceDec 18, 2025Agent TrustAgent Passports: A Multi-Dimensional Identity Verification Framework for Autonomous AI SystemsDec 12, 2025MethodologyImpact Flows Across Layers: How Data Quality Cascades into Agent Decisions into Systemic StabilityDec 5, 2025MethodologyOne Framework Becomes Ten: The Amplitude Thesis for Multi-Layer AI Impact MeasurementNov 10, 2025Data ScienceFrom Scores to Prices: A Derivation Model for AI Data Marketplace SignalsOct 20, 2025Data ScienceRuntime Data Scoring via MCP Server Architecture: Delivering Quality Signals During Agent Tool CallsSep 15, 2025Data ScienceFour Dimensions of Data Quality in the Agentic Era: Scarcity, Quality, Decision Impact, DefensibilityAug 25, 2025Data ScienceMeasuring Data Source Value for AI Agents: A Counterfactual KL-Divergence ApproachAug 12, 2025MethodologyFrom Readiness to Reality: Why External AI Impact Measurement Matters More Than Internal AssessmentJun 20, 2025MethodologyMeasurement for AI Frontiers: Why the Industry Needs an Impact StandardMay 15, 2025MethodologyCan AI Impact Be Measured? A Case for Quantitative Scoring Methodologies

Every question a regulator, judge, or enterprise buyer would ask about AI agents - answered.

Get early access to Amplitude scoring, research updates, and framework specifications.

Join the waitlist. No spam, just measurement science.