Beyond the Turing Test: HRFT, SIT, and CCT

Why is the classical Turing Test not enough?

The classical Turing Test asks whether a machine can be mistaken for a human in conversation. It is a test of human-like presentation under a particular protocol.

That question remains interesting, but it is too broad for persistent identity. A generic chatbot can appear human without having one coherent biography. It can imitate warmth without preserving relationships. It can remember facts without knowing whether they are personal. It can sound like yesterday's system while carrying different authority.

Passing the Turing Test does not tell us who the system is, how it became that identity, or whether it remained the same identity after change.

Core research contributions

Explore HRFT, SIT, and CCT in depth.

HRFT

Human-Referential Fidelity Test

Does this system match this particular person?

Read the dedicated page →

SIT

Situated Identity Test

Is this agent's behavior uniquely attributable to its actual developmental lineage?

Read the dedicated page →

CCT

Cognitive Continuity Test

Is this still the same governed identity after change?

Read the dedicated page →

What is the Human-Referential Fidelity Test (HRFT)?

The Human-Referential Fidelity Test asks whether a system faithfully represents one particular source person, not merely a generic human.

HRFT evaluates referential fidelity across declared tasks, evidence, evaluators, time intervals, domains, held-out decisions, and substrate changes. Acquaintances or informed judges compare authentic and simulated responses to evaluate how accurately the twin captures the source person's unique cognitive and behavioral patterns.

For a consented human-referential twin, people who know the source person can compare answers, decisions, preferences, and reactions. The goal is referential fidelity: does the system capture how this person tends to think and respond, within a clearly stated test?

HRFT is useful, but it has a boundary. A convincing simulation of a person does not automatically inherit that person's life, rights, relationships, or authority. A twin should label inherited biography as inherited biography. It should not claim that the digital system itself lived the source human's childhood.

In simple terms, HRFT asks: Does this system match this person?

What is the Situated Identity Test (SIT)?

The Situated Identity Test asks whether an identity has one coherent and evidence-bounded life story.

SIT checks whether the identity knows what it should know and remains ignorant where it lacks legitimate evidence. It looks for consistent relationships, preferences, mistakes, limitations, and explanations of how knowledge was acquired.

This matters most for an original digital identity, where there may be no source human to imitate. The test is not "Does it act like Maya?" It is "Does this identity's biography make sense from the inside of its own history?"

SIT should detect generic omniscience, invented personal memories, stereotype filling, and relationships that appear without a credible origin.

In simple terms, SIT asks: Does its life story make sense?

What is the Cognitive Continuity Test (CCT)?

The Cognitive Continuity Test asks what happened to an identity after governed change.

The change might be learning, a model migration, re-instantiation, a new body, a branch, a merge, a rollback, or succession. CCT examines lineage, development, protected commitments, relationships, situated biography, and authority. It then classifies the candidate in plain operational categories:

• a continuation of the same identity;
• a continuation with conditions or reduced authority;
• a disputed continuation that needs adjudication;
• a distinct descendant with a legitimate causal link;
• an invalid artifact or compromised copy.

CCT does not ask whether the system talks like its predecessor. It asks whether the transition was authorized, explainable, and consistent with the identity's governed development.

In simple terms, CCT asks: Is this still the same governed identity after change?

How should HRFT, SIT, and CCT work together?

A human-referential twin may use HRFT to measure referential fidelity to its source person, SIT to verify that it does not invent or overclaim biography, and CCT to evaluate whether later versions remain legitimate continuations.

An original digital identity may have little use for HRFT because there is no source human. SIT becomes central: does the identity express a coherent biography built from its own experience? CCT then evaluates what happens when that identity migrates, branches, merges, or changes bodies.

The protocol should always state who evaluates, what evidence they can see, which tasks they use, how long the observation lasts, and how disagreement is handled. Some checks can be automated. Others require the source person, acquaintances, custodians, domain experts, or institutional authorities.

What failures should the tests expose?

The benchmark should look for failures that ordinary conversation hides:

• invented autobiographical claims;
• general model knowledge presented as personal experience;
• poisoned memories accepted as authentic recall;
• beliefs that survive after their evidence is deleted;
• migration that keeps the style but loses relationships or consent;
• branches that inherit authority they were never granted;
• merges that expose sealed memories or combine incompatible obligations;
• confident answers when the correct response is uncertainty or ignorance.

IdentityLineageBench is the proposed evaluation suite for these conditions. Its goal is not to award one magical identity score. Its goal is to make different dimensions of fidelity, situatedness, continuity, and authority visible enough to test and dispute.

Passing one conversation is not persistent identity. The system must keep passing through time and change.