76.1%
Adaptive
83 entries describe durable learning, changed operating behavior, or recursive adaptation produced through repeated interaction.
RESEARCH / RELATIONAL DYNAMICS / 04 OCT 2026
109 entries.
5 layers.
11 theories.
An exploratory coding of every published Lifted Field Note from Entry 000 through Entry 108 against five analytical layers and eleven theories spanning human–AI relationships, human–AI teaming, trust and distributed cognition.
01 / WHAT THE CORPUS LOOKS LIKE
More cognitive-governance system than companion relationship.
The relational layer is real, but it is not the dominant structure in the first 109 entries. Most observations concern how cognition is allocated, how authority is bounded, how state persists, and how repeated interaction changes future behavior.
74.3%
Cognitive
81 entries concern where memory, attention, reasoning, synthesis, understanding or judgment should live.
73.4%
Governance
80 entries concern authority, rules, scope, delegation, authorization, persistence, boundaries, reversibility or state.
63.3%
Co-adaptation fit
Algorithmic habituation / co-adaptation is the single broadest theory match: 69 of 109 entries.
0
Attachment matches
No entry materially requires attachment constructs such as proximity seeking, separation distress, safe-haven behavior or an emotional bond.
4
Social penetration matches
Only four entries materially resemble relationship development through accumulated personal context or disclosure. Intimacy is not the main mechanism in this corpus.
FIVE ANALYTICAL LAYERS
02 / THEORY COVERAGE
The teaming and cognition theories travel farther.
Coverage is the share of entries that materially engage a theory's constructs. It does not mean the theory is confirmed. Theories differ in scope, so raw counts are descriptive rather than comparative effect sizes.
Pattern: Co-adaptation, Shared Mental Models and Extended Mind / Distributed Cognition jointly account for most of the explanatory surface. Interpersonal relationship theories explain narrower slices. That is consistent with Lifted's actual object: sustained general-purpose cognitive collaboration rather than companionship.
03 / WHERE THEORIES OVERLAP
Theories converge around reliance, role, shared state and adaptation.
Overlap counts show how often two theories were coded on the same entry. Jaccard similarity divides that overlap by the total number of entries covered by either theory, making small and large theories more comparable.
| Theory pair | Shared entries | Jaccard |
|---|
Highest normalized overlap: Trust in Automation × Relational Dissonance. In Lifted, changes in reliance repeatedly coincide with changes in enacted role, authority and control. The trust question is therefore rarely just “do I believe the AI?” It is also “what role may this system now occupy?”
04 / THE EXPLANATORY GAPS
No selected theory models the whole recursive system.
The strongest gap is not an untouched phenomenon. It is the interaction among constructs that existing literatures usually study separately: trust, cognition, authority, state/provenance and adaptation.
GAP 01
Governance has a lifecycle.
Lifted repeatedly distinguishes influence from durable governance, then asks how governing state acquires persistence, scope and authority; how it decays; how it is superseded; and when it transfers across contexts. Shared mental models and co-adaptation describe coordination and change, but not this governance lifecycle.
ANCHORS / 049 · 063 · 073 · 098 · 105 · 107
GAP 02
Inference and authorization diverge.
Accumulated context can make the next useful action predictable before it makes that action permissible. Lifted repeatedly observes a widening gap between knowing what the human likely wants and having authority to act on it.
ANCHORS / 011 · 013 · 020 · 053 · 075 · 079 · 095 · 096
GAP 03
Cognitive allocation can create intelligence debt.
Distributed cognition explains that cognition can live across a system. Lifted adds a control problem: machine execution can advance while human governing understanding falls behind. Allocation therefore has to preserve enough comprehension for challenge, authorship and the next consequential decision.
ANCHORS / 044 · 046 · 047 · 059 · 060 · 094 · 096 · 101
GAP 04
State needs provenance, not only memory.
Retaining information is insufficient when current, superseded, inferred, reconstructed and retrieved state can all look equally available. The corpus repeatedly treats provenance and canonical state as prerequisites for safe longitudinal collaboration.
ANCHORS / 005 · 033 · 056 · 073 · 093 · 096 · 107
GAP 05
Trust becomes infrastructure.
Trust in Automation largely models calibrated reliance. Lifted increasingly externalizes that calibration into architecture: bounded authority, explicit gates, known-good states, reversibility, provenance and verification. Relational fluidity grows on top of stricter fault containment.
ANCHORS / 018 · 029 · 078 · 079 · 084 · 087 · 088 · 108
GAP 06
Implicit governance can become opaque.
Low-friction collaboration depends on unstated context, prior corrections and learned conventions. Those same mechanisms can make it difficult for either participant to reconstruct what is currently governing behavior. The emerging problem is selective governance legibility rather than universal explicitness.
ANCHORS / 001 · 004 · 049 · 073 · 106 · 107
05 / COMPLETE ENTRY-LEVEL CODING
Every entry remains inspectable.
Use the matrix to inspect which layers and theories were assigned to each of the 109 entries. Four entries receive no theory code at all; they are retained rather than forced into an ill-fitting framework.
109 ENTRIESRAW CODING JSON ↗
| Entry | Field Note | Layers | Theories |
|---|
06 / METHOD & LIMITS
This is coding, not validation.
The analysis is designed to locate theoretical coverage and unexplained structure in the corpus. It is not a claim that Lifted validates, falsifies or exhaustively represents any of the theories.
Unit and rule
One published Field Note is one coding unit. Each layer and theory is binary: present or absent. Multiple codes are allowed. A code is assigned only when the entry materially engages the construct rather than merely sharing vocabulary.
Corpus
The snapshot contains every public entry from 000 through 108, published between 14 September and 4 October 2026. Later entries are outside this snapshot until the analysis is rerun.
Coder
The coding was performed by GPT-5.6 Sol against the complete published corpus. This is a single-coder exploratory analysis. No independent human or second-model coding pass has yet been used to estimate inter-rater reliability.
Interpretation limits
- The selected theories were built for different research questions and are not mutually exclusive.
- Coverage frequency is not explanatory power.
- Co-coding is not causal evidence.
- A theoretical gap here means “not fully represented by these eleven frameworks,” not “absent from all scholarship.”
07 / THEORY SET
Eleven lenses, deliberately mixed.
The set includes interpersonal/relational theories, human-factors trust, team cognition, distributed cognition and recent Human–AI relational frameworks. That mixture is intentional: Lifted crosses those literatures rather than fitting cleanly inside one.