Public status: September 2026

Results & Limitations

A public, revisable account of what our tests allow us to strengthen, bound, contradict, or leave open.

CREDIBLE RESEARCH IS MEASURED NOT ONLY BY WHAT IT ASSERTS, BUT ALSO BY WHAT IT KNOWS HOW TO BOUND, REVISE, AND LEAVE OPEN.

Public definition

What we call a “result”

A result is not necessarily a final conclusion. It may be a distinction that withstands testing better, a hypothesis that loses, a scope that narrows, a better-localized ???, or a new question that has become necessary.

RESULT ≠ CONFIRMATION.
DISCONFIRMED HYPOTHESIS = RESULT.
A BETTER-FRAMED QUESTION = POSSIBLE RESULT.
A BETTER-LOCALIZED ??? ≠ FAILURE.

Public statuses

StrengthenedLocal / boundedOpenNot justified at this stage???

Strengthened: A distinction or hypothesis has withstood several tests or cross-checks within the scope examined. This means neither universal validation, definitive proof, nor a complete mechanism.
Local / bounded: The result holds under certain conditions, perturbations, domains, or corpora: here, not everywhere.
Open: The question remains active, and several alternatives remain compatible with the current evidence.
Not justified at this stage: The current evidence does not support a stronger conclusion.
???: An unknown is preserved explicitly rather than replaced with an artificial answer.

DOCUMENTARY STATUS ≠ ONTOLOGICAL STATUS.
STRENGTHENED ≠ GLOBALLY VALIDATED.

A dated state

The register remains alive

The statuses describe the state of the research at a given date. A new observation, a better-chosen perturbation, an external publication, a counterexample, or a new corpus may change them.

Last public review: September 2026.

CURRENT STATE ≠ DEFINITIVE VERDICT.
UPDATE ≠ REWRITING HISTORY.
REVISION ≠ ERASING THE PREVIOUS STATE.

Ongoing research

Gaining resolution without accumulating

The work continues. As tests multiply, some distinctions gain resolution, while others split, change scope, become redundant, or lose their usefulness.

GAIN IN RESOLUTION ≠ ACCUMULATION OF CATEGORIES.
MORE DISTINCTIONS ≠ BETTER OBSERVATION.
MORE PRECISE TERM ≠ NEW PHENOMENON.

The main risk is no longer only a lack of distinctions, but their coherent overproduction before they are put under experimental pressure. Progress must come from discriminating tests, contrastive cases, and perturbations as much as from vocabulary.

What a test can reveal

Sought, secondary, and unexpected results

Sought result: The test directly answers the question it was designed to address.
Secondary or unexpected result: Another difference becomes observable without having been the initial target.
New open question: The result requires the problem to be reframed, bounded, or broadened.

WE DO NOT SEEK ONLY TO KNOW WHETHER A HYPOTHESIS SURVIVES. WE ALSO OBSERVE WHAT THE TEST FORCES US TO DISTINGUISH THAT WE HAD NOT YET SEEN.

WHAT EMERGES FROM A TEST MAY BE MORE IMPORTANT THAN WHAT THE TEST WAS SUPPOSED TO CONFIRM. It must then be retested and bounded before any promotion of status.

UNEXPECTED RESULT ≠ ESTABLISHED SCIENTIFIC DISCOVERY.
SURPRISE ≠ PROOF.
EMERGENCE OF A QUESTION ≠ VALIDATION OF A THEORY.

Examples of gains in resolution

Understanding

Trace / access / mobilization

“Available knowledge” was unpacked into the presence of a trace, the possibility of access, and actual use at the relevant moment.

Installation / maintenance

What enables a configuration to be installed does not necessarily maintain it.

Local / configurational support

Support may be distributed or relational without becoming “the entire system.”

Historical / current dependence

A past cause may contribute to an effect without remaining necessary now.

Epistemic / functional support

What supports a proposition is not necessarily what a function depends on.

Explore the Understanding research area.

Examples of gains in resolution

Emotions

Emotion / action

An emotion may remain active without automatically determining action.

Regulation / extinction

Changing the trajectory does not mean suppressing the emotion.

Intensity / authority

An intense emotion is not automatically granted greater behavioral authority.

Availability / action window

A strategy may be known yet unavailable at the relevant moment.

Valence / function

A pleasant emotion does not automatically orient a person toward a valid or useful outcome.

Explore the Emotions research area.

Examples of gains in resolution

AI Engineering

Presence / mobilization

Information may be in the context without actually being used.

Capability / permission / authority / action

Being technically capable of acting does not mean being authorized to act.

Source / descendants

Correcting the source does not automatically correct everything downstream.

Revision / recovery

Recognizing a correction does not guarantee functional recovery.

Output / agent state

Correcting text does not guarantee correction of memory, the plan, permissions, or intermediate decisions.

Explore the AI Engineering research area.

Strengthened

Distinctions that currently withstand testing

Within the scopes examined, the following distinctions have withstood several cases, perturbations, or methodological cross-checks:

same output ≠ same mechanism;
presence ≠ use;
access ≠ mobilization;
stability ≠ immobility;
revisability ≠ instability;
historical provenance ≠ current justification;
historical dependence ≠ current dependence;
correction of a source ≠ automatic correction of descendants;
local robustness ≠ global robustness;
epistemic support ≠ functional support;
functional weight ≠ epistemic authority;
participation ≠ right to close.

The following also remain strengthened within their documented scopes: correct information ≠ relevant information; current state ≠ history of the state; repetition ≠ independent confirmation; content ≠ status ≠ usage permission.

STRENGTHENED ≠ UNIVERSAL LAW.

Local / bounded

What holds within a limited scope

some distinctions derived from specific textual corpora;
results from local crash tests;
hypotheses about configurational support;
hypotheses about mobility between modes of examination;
effects observed in certain emotional trajectories;
AI metrics tested in controlled scenarios;
adaptive resolution, future-space, the dependence of certain descendants, and robustness under particular perturbations.

These elements have traces or local utility; their broader scope remains open.

LOCAL ≠ FALSE.
LOCAL ≠ GENERALIZABLE.

Stability: distinguishing trajectories

The textual campaign leads us to distinguish, as avenues for observation, a maintained capacity, a recovery yet to be established and recourse to another route. A subsequent action proves neither a prior loss of capacity nor the return of a subjective state. A strategy that is actually implemented does not guarantee the intended outcome either.

SIMILAR RESULT ≠ EQUIVALENT TRAJECTORY.
TEXTUAL DYNAMICS ≠ PSYCHOLOGICAL MECHANISM.

Open

Questions that remain active

the exact nature of certain forms of integration;
the logical type of “functional availability”;
the relevance of the term “weight”;
the status of “contextual mobilizability”;
the distinct usefulness of “descendants” beyond existing notions;
the exact distinction between different types of ???;
the nature of the “core” reported in certain human observations;
the boundaries of functional permissions;
mobility between modes of examination.

This list is indicative, not an exhaustive program.

Not justified at this stage

Conclusions not promoted

a general theory of understanding, emotions, or intelligence;
established artificial consciousness;
Human / AI equivalence;
a single mechanism behind the three research areas;
the 7×9 framework as an ontology of the phenomenon, or a justified tenth instrument;
distributed support = the entire network;
functional permissions as a legal theory;
the automatic superiority of an AI methodological layer;
the demonstrated scientific novelty of every Aksoydan term;
a global score.

What we have had to revise

Important revisions

Understanding: single state → candidate profile or trajectory

Dissociations between recognition, explication, transfer, availability, and integration led us to unpack the general term into several candidate phenomena.

Availability: general term → object and situation

The term proved too compact: the availability of a trace, capability, strategy, or alternative does not automatically designate the same property.

Support: presumed carrier → distinct relations

The research now separates epistemic support, functional support, access, expression, permissive condition, and mechanism.

Emotions: stable message → possible situated function

The green/orange/red scheme is not included in the current scientific page: a single intensity axis would too easily reduce the configuration, its object, and its trajectory.

AI: internal terms → comparison with existing frameworks

Several laboratory notions strongly overlap with least privilege, provenance, knowledge editing, belief revision, and authorization. They are therefore not presented as automatically new.

Resolution: multiplication → adaptation and constraint

Adding distinctions is no longer sufficient: whether they are retained depends on their discriminating power under testing.

REVISING ≠ DISOWNING.
UPDATING ≠ REWRITING HISTORY.

A previous state may remain visible as a superseded hypothesis, earlier formulation, surpassed limitation, or revised decision. It must not be presented as though it had always been the current state.

Real exposure to failure

A hypothesis must be able to lose

A hypothesis does not enter the laboratory to be protected. It must be able to lose to a better counterexample, a more discriminating protocol, or a simpler explanation.

A HYPOTHESIS MUST BE ABLE TO LOSE.

INTERNAL COHERENCE ≠ PROTECTION FROM REVISION.

When a distinction becomes unnecessary

It may survive, be bounded, split, merge with another, lose its discriminating power, or become unnecessary.

MORE CATEGORIES ≠ MORE SCIENCE.

Visible disagreement

Contradictions

A contradiction is not automatically an error to remove. It may signal a difference in context, level, measurement, or support, or a genuine limitation of the framework.

Apparent contradiction: The formulations seem incompatible but do not concern the same condition or level.
Framework divergence: Two framings organize the question differently.
Genuinely incompatible result: The same conditions produce results that cannot be maintained together.
Unknown: The evidence is not yet sufficient to decide.

CONTRADICTION ≠ FAILURE TO CONCEAL.

External resistance

Scientific literature and prior work

External literature serves as a constraint, comparison, source of resistance and counterexamples, existing vocabulary, and boundary. It is not a bibliographic ornament.

Several laboratory questions overlap with work already conducted in cognitive science, neuroscience, psychology, philosophy, AI safety, interpretability, provenance, belief revision, and authorization. The laboratory does not claim as new what has already been extensively studied elsewhere.

CONVERGENCE ≠ VALIDATION.
DIFFERENT VOCABULARY ≠ ABSENCE OF AN EQUIVALENT.
EXTERNAL LITERATURE ≠ BIBLIOGRAPHIC ORNAMENT.

Candidate instruments

What we still need to justify

the distinct discriminating power of “descendants”;
the distinct usefulness of status inflation;
the usefulness of scope drift in our precise sense;
the value of recovery after correction as a distinct metric;
the relevance of “weight” as a shared variable;
the contribution of “functional permissions” beyond existing authorization frameworks;
the possible distinct value of adaptive resolution.

NAMING ≠ JUSTIFYING.

???

Locating what remains unknown

??? is not a shameful empty box. It may indicate missing information, an unestablished relation, an unknown mechanism, insufficient support, indeterminate scope, multiple interpretations, an undetermined right to act, or a variable that has not yet been distinguished.

??? ≠ A SINGLE TYPE OF UNKNOWN.
PRESERVING THE UNKNOWN ≠ REFUSING TO LEARN.

Propagation and correction

Re-examining descendants

When a source changes, descendants whose justification, status, or usage permission may depend on it must be re-examined in a targeted way. Some remain dependent; others may have acquired an independent justification.

CORRECTION OF A SOURCE ≠ AUTOMATIC CORRECTION OF DESCENDANTS.
NON-AUTOMATIC CORRECTION ≠ ABSENCE OF RE-EXAMINATION.

What research also produces

Research descendants

A distinction may produce a new test, page, protocol, hypothesis, tool, prototype, or question. The existence of this descendant does not retrospectively validate the source result.

The AI prototype is thus a partial technical descendant of certain research questions.

RESEARCH DESCENDANT ≠ VALIDATION OF THE SOURCE RESULT.
PROTOTYPE ≠ SCIENTIFIC VALIDATION.

Reading guide

How should the register be read?

Strengthened: “This currently withstands testing better.”
Local / bounded: “This holds here, not everywhere.”
Open: “Several possibilities remain active.”
Not justified: “We do not have sufficient evidence to go further.”
???: “The unknown is explicitly preserved.”

A distinction may change tomorrow without that revisability constituting arbitrary instability.

REVISABILITY ≠ INSTABILITY.
STABILITY ≠ IMMOBILITY.

More stable safeguards

What does not change easily

Hypotheses change more easily than certain methodological rules: observation ≠ interpretation; hypothesis ≠ conclusion; convergence ≠ validation; same output ≠ same mechanism; do not fill in the ???; a hypothesis must be able to lose.

These are methodological safeguards, not general laws of the world.

Continue without repeating everything

Connecting the register to the research

Some revisions arise when a term becomes too broad to preserve observed differences. The Container terms page explains this unpacking process.

View the three research areas Understanding Emotions AI Engineering

Last public review: September 2026. This register describes a revisable state, not a definitive closure.