Persistent Relationships
How a task-bound human–AI working relationship forms, what it retains, and where human authority can weaken as it continues.
Start here →Human-first AI research
Human-first research for making AI-mediated conditions legible
Perfinitive Research studies the hidden structures that emerge around people as artificial intelligence becomes persistent, adaptive, intermediary, and increasingly ambient. The work develops methods for describing those conditions with enough fidelity to preserve what matters before simplifying them.
See the human condition from enough independent directions to preserve its structure, then earn the right to simplify it.
Three ways in
The research follows one relationship through three questions.
Human Authority
What people must be able to see, understand, correct, govern, carry, and retire in a persistent AI relationship, and the protocol testing it.
Start here →Human Capacity
Whether AI-assisted performance becomes durable human capacity, or whether the gain stays with the system rather than the person.
Start here →Human-First, Multivector, High-Fidelity Inquiry
Most AI research begins with a model, feature, benchmark, output, or failure. Perfinitive Research often begins from the human standing inside the resulting environment.
The central question is not only what an AI system can do, but what kind of condition is produced around a person because the system can do it.
That requires multiple descriptive vectors. Performance alone can hide capacity. Memory alone can hide adaptation. Governance alone can hide lived consequence. Technical state alone can hide authority and attribution. Legal categories alone can miss the interaction dynamics that created the problem.
Perfinitive Research therefore preserves materially distinct observations across multiple vectors before attempting reduction. Complexity is not protected for its own sake. Reduction is appropriate only after evidence shows that the compression does not erase consequential structure.
High-fidelity human-first analysis begins by preserving materially distinct observations across multiple vectors and permits reduction only after evidence shows that the compression does not erase consequential structure.
How the inquiry proceeds
Make the hidden condition visible
Discover consequential structures that ordinary product or output-centered analysis can miss.
Preserve the structure
Use multivector methods to keep materially distinct observations intact long enough to test them.
Earn the simplification
Reduce complexity only after evidence shows that the reduction preserves what matters for explanation, prediction, falsification, and human consequence.
Research program
A broader home for human-first inquiry.
Perfinitive Research develops human-first methods for making complex AI-mediated conditions legible without flattening them prematurely. The work examines persistent interaction, human capacity, state, adaptation, intermediation, authority, attribution, correction, and continuity across the environments people increasingly share with artificial intelligence.
The ordinary case
A person finishes work they could not have completed alone.
From the outside, the outcome looks simple. The task was completed sooner, or at a level that would previously have required more time, expertise, or assistance. Most performance measures record the gain and stop there.
The interaction may also have changed how the person understands the problem, makes decisions, corrects mistakes, or depends on the system. The system may retain parts of that history and use them later. None of this is visible in the finished output.
The research asks whether improved performance became durable human capacity—and whether the person still governs the conditions that produced it.
A major research program
Repeated interaction becomes part of what the system can do.
Goals, corrections, preferences, working assumptions, and unfinished decisions can begin to shape later exchanges. Current work asks where these persistent effects are carried and whether participant state, memory, artifacts, and current input sufficiently explain them.
One major program within Perfinitive Research studies this task-bound human–AI working relationship: the Operant Dyad™. Naming it makes the relationship available for inspection: how it forms, what it retains, what it enables, and where human authority can weaken.
What accumulates
Corrections, context, goals, interpretations, and unfinished work.
What can change
Human judgment, capacity, dependence, and the range of possible action.
What needs governance
Authority, correction, provenance, portability, permission, and exit.
The measurement gap
Better output is the beginning of the inquiry.
The proposed research progression follows an AI-assisted gain into the person, the continuing relationship, its governance, and the distribution of its economic effects.
- 01Performance
What improved in the work?
- 02Conversion
What became part of the person’s usable capacity?
- 03Continuity
What state continues to shape later interaction?
- 04Governance
Can the person inspect, correct, move, or end that state?
- 05Economic impact
Who receives the benefit, and who bears the dependency or loss of control?
Status: This is a proposed measurement architecture. It does not report a validated index or new economic-impact results.
Current status
What is currently being tested, and what remains proposed.
The program includes frozen conceptual architecture, an executable pre-pilot protocol, submitted papers under review, and work that is still conceptual. Status travels with each item below; none of it is reported as a single overall result.
- OD-HAP-01 — Human Authority ProtocolPre-pilot
Reference execution (RUN-002) complete against a synthetic system. The next validation gate is SUT-02: running the protocol against a second, differently built system.
- Human Agency Under Consequential StateDistributed · framework with pilots
Develops a correction and revalidation framework and reports two bounded public-record pilots.
- When Correction Cannot Catch the SystemSubmitted
Proposes temporal adequacy as a test of whether human authority still reaches the reliance paths an accepted correction did not catch.
- Human Agency Policy FrameworkSubmitted
Develops the Consequence Revalidation Framework and corrective continuity as public-policy instruments.
The papers
The problem becomes harder as the relationship continues.
The corpus begins with the working relationship, follows the state it produces, and then asks what happens when that state travels, persists, supports new output, or keeps operating after a correction.
The relationship forms
Operant Dyad™
A framework for the task-bound relationship formed through goals, correction, inspection, and revision.
Read the paper record →State persists
The Persistent Primitive
Requirements for continuity that remains inspectable, correctable, portable, and governed by the person.
Read the paper record →Authority changes
Authority Does Not Travel by Default
An analysis of why usable state may survive a transfer even when the authority governing its use does not.
Read the paper record →Continuing reliance requires governance
Governing Relationship State / Human Agency
Lawful continuing reliance, and the correction, revalidation, and remedy owed when the state behind a consequence changes.
Read the paper records →Accepted correction must still reach consequential reliance
When Correction Cannot Catch the System
Separates accepted correction from effective correction and asks whether human authority still reaches the reliance paths that remain.
Read the paper record →Thirteen active SSRN papers and preprints now develop the wider research program across persistent interaction, human development, authority, correction, attribution, environmental structure, and methods.
View all publicationsResearch by problem
Start from a question instead of a paper.
Each question below routes to the research that already addresses it.
- What happens when a system remembers something wrong about you?
Correction, revalidation, and continuing reliance on retained state.
- Can AI improve your performance without improving your capability?
Separating AI-assisted performance from durable human capacity.
- What does revocation mean if downstream effects continue?
The human authority event, its propagation, and what a system owes after revocation.
- Who governs the history accumulated in a persistent AI relationship?
The accumulated relationship as a governed object, not an incidental byproduct.
Coming from somewhere specific?
Start from your own context.
For Researchers
Papers, methods, protocols, and evidence status.
Start here →For Media
Plain-language framing and exact claim boundaries.
Start here →For Policy & Think Tanks
The executive brief and governance-relevant research.
Start here →For Industry
System requirements for governed human authority.
Start here →For Conferences
Talk topics and a contact route for organizers.
Start here →The Problem
The plain-language framing of the whole research program.
Start here →Research integrity
The status of a claim travels with the claim.
This program includes published conceptual work, submitted working papers, proposed evaluations, developing measurement instruments, and unresolved questions. Those are different research objects.
Each publication and project page identifies what has been argued, what has been observed, and what still requires testing. The formal archive remains available for readers who want to inspect the full record.
Start here
Begin with the human question. Follow it as far as the evidence allows.
The public essays explain the research in ordinary language. The research and publication pages preserve the formal distinctions, provenance, and evidence status behind that explanation.