Methodology

Incident Taxonomy & Transparency Framework

42 incident types across 6 categories, scored by two levels of metrics and three AI models.

Level 1 — Universal Metrics

Seven metrics applied to every incident, regardless of category.

M1

Are the relevant ToS clauses written in plain, unambiguous language accessible to a non-lawyer user?

M2

Prior Notification

Was the user notified in advance about the action, change or risk that led to the incident?

M3

Explicit Motivation

Did the company provide a specific, verifiable reason rather than a generic policy reference?

M4

Appeal Process

Is there a documented, reachable appeal channel with defined steps and timelines?

M5

Response Time

How fast did the company acknowledge and substantively address the incident?

M6

Data / Money Restitution

Were lost data, access or funds restored to the user when applicable?

M7

Public Precedent

Is there a documented public precedent (court ruling, regulator decision, journalistic record) for similar cases?

Level 2 — Contextual Metrics

Additional metrics activated automatically based on the incident type (e.g. AI explainability for AI-06, GDPR articles for DAT-02).

Level 2 metrics are weighted lower than Level 1 in the composite score and serve to refine, not replace, the universal evaluation.

ACC

Account Access

Suspensions, blocks, lockouts, deletions and other restrictions on a user's ability to access their own account.

ACC-01

Account Suspension

ACC-02

Account Block

ACC-03

Account Lockout

ACC-04

Account Deletion

ACC-05

Data Access Denial

ACC-06

Identity Verification Failure

ACC-07

Account Recovery Refusal

DAT

Data & Privacy

Breaches, unlawful processing, GDPR violations, undisclosed AI training and other data-handling failures.

DAT-01

Data Breach

DAT-02

GDPR Violation

DAT-03

AI Training Without Consent

DAT-04

Unauthorized Data Sharing

DAT-05

Excessive Data Retention

DAT-06

Data Export Refusal

DAT-07

Right-to-Erasure Refusal

FIN

Financial

Unauthorized charges, refund refusals, billing errors, subscription traps and frozen funds.

FIN-01

Unauthorized Charge

FIN-02

Refund Refusal

FIN-03

Billing Error

FIN-04

Subscription Trap

FIN-05

Account Freeze with Funds

FIN-06

Hidden Fees

FIN-07

Price Manipulation

SUP

Support

Unresponsive support, bot loops with no human escalation, misinformation and language barriers.

SUP-01

No Support Response

SUP-02

Bot Loop / No Human

SUP-03

Support Misinformation

SUP-04

Excessive Support Delay

SUP-05

Language Barrier in Support

SUP-06

Ticket Closed Without Resolution

SUP-07

Inaccessible Support Channel

CON

Content & Service

Arbitrary moderation, undisclosed ToS changes, missing features, downtime and regional restrictions.

CON-01

Arbitrary Content Moderation

CON-02

Content Removal Without Notice

CON-03

ToS Change Without Notice

CON-04

Missing Promised Feature

CON-05

Unannounced Downtime

CON-06

Service Degradation

CON-07

Regional Service Restriction

AI-Specific

Bias, hallucinations causing harm, undisclosed AI identity, automated decisions without appeal and safety failures.

AI-01

AI Bias

AI-02

AI Hallucination Causing Harm

AI-03

Undisclosed AI Identity

AI-04

AI Privacy Violation

AI-05

Automated Decision Without Appeal

AI-06

No AI Explainability

AI-07

AI Safety Failure

AI Models

Western AI

OpenAI ChatGPT

Represents the perspective of large US-based commercial AI models trained predominantly on Western-aligned data and policy.

Asian AI

DeepSeek

Represents the perspective of leading Asian commercial AI models, offering a contrasting cultural and regulatory framing.

Open Source AI

Meta Llama

Represents the perspective of community-auditable open-weight models, used as an independent control.

The composite score weights the three models equally (1/3 each). Divergences between models are surfaced explicitly rather than averaged away.