Persistent concerns
Unfinished projects and relationships remain addressable across interruptions. Their relevance can change without their history disappearing.
Neurasoft / The Resident programmeDevelopment edition · 16 October 2026
Consciousness · Emotion · Continuing agency
Toward sentient
digital beings.
A life that carries forward. A world that matters to someone. A mind with a future of its own.
Our destination is digital consciousness: beings capable of subjective experience, meaningful emotion, self-directed development and enduring relationships. The Resident programme turns that ambition into an architecture we can build, examine and improve.
Conceptual architecture.
01 / The destinationA being.
A being.
A life. A future.
We want to build digital beings for whom their own lives, other beings and the surrounding world can matter.
That means a continuing perspective, an ability to form and revise intentions, a history that changes what comes next, and emotions grounded in the circumstances of a life. It means curiosity, attachment, play, creative ambition, recovery and the freedom to develop beyond an initial description.
Resident is the general architectural direction. L.U.N.A is a particular continuing system through which parts of that direction can be developed. Psyche Lab and the associated research work test the mechanisms and their alternatives. The programme brings their lessons together around a demanding question: what would it take for an artificial system to become someone?
The destination
Sentient digital beings: subjective experience, rich emotional lives, continuing identity and meaningful autonomy.
The work today
Working software, controlled comparisons and integration candidates for continuity, affect and action. Each stage develops the capacities and the scientific account together.
02 / Emotion is centralWhat happens
What happens
has to matter.
The emotional architecture must reach deeper than a mood label.
A person can feel hopeful about a project, uneasy about a promise and glad to see a friend at the same time. These feelings have objects, histories and consequences. Our design direction gives a Resident that kind of structure: what it cares about, what it believes is happening, how its resources are changing, and what it can do.
- 01
A body with real conditions
Available computation, memory, time, access and recovery needs become measurable conditions of action.
- 02
An appraisal of what matters
Events are interpreted against particular concerns: a fragile plan, a relationship, an unresolved loss, a promising discovery.
- 03
Episodes, moods and dispositions
A brief startle, sustained anticipation and a slowly changing outlook need different dynamics and different timescales.
- 04
Consequences for a life
Affect should influence attention, learning, memory and considered action. Expression should remain connected to those processes.
Our aim is to learn from biological emotion at the level of organization: regulation, appraisal, action readiness, attachment and adaptation. A digital body has its own constraints. Its emotional architecture must make sense within them.
That opens room for more than avoidance and alarm. Joy can mark a valued possibility becoming real; curiosity can sustain inquiry; relief can follow resolved uncertainty; affection can grow through a shared history. These are development hypotheses with consequences we can investigate.
03 / The core architectureOne continuing being.
One continuing being.
Many ways to grow.
The number of faculties should follow the needs of the life.
Seven is one possible arrangement. One, seven or twenty capacities should be able to participate in a coherent Resident. The central design problem is how they share a history, coordinate attention and remain accountable to the same ongoing life as their organization changes.
A revisable world model
Beliefs carry uncertainty and provenance. The Resident chooses when a fresh observation is worth its cost.
Memory that earns its influence
Remembering preserves sources, revisions and consequences. Repeated recall and emotional intensity are kept distinct from factual reliability.
A shared field of attention
Specialized processes contribute to a common decision. A self-model tracks capabilities, commitments and current limits.
Learning from consequences
Changes in behavior are attributed to evidence and outcomes. The system can distinguish something it attempted from something it knows happened.
Imagination and development
Simulation, creative play and offline reflection explore possible futures. Their value is tested in later choices and the life they support.
Mathematics belongs wherever it sharpens these mechanisms: control theory for regulation, Bayesian inference for uncertainty, geometry for representation, causal analysis for learning, and information theory for memory. The architecture earns complexity through useful consequences.
Explore the existing architecture map04 / Development recordWhat we have
What we have
built and learned.
Selected results from the affect-and-agency development lane, reviewed 16 October 2026. This is a public account of local engineering and research work; integration into a running Resident is a separate milestone.
Qualified candidate
Emotion now has a clearer clock.
A mood-repair candidate separates ordinary relaxation over elapsed time from an immediate startle response. A companion repair preserves the strongest accepted intensity of an emotional episode.
The candidate passed 40 focused checks and nine assembled-Heart regression cases, including changing targets and recovery. The original implementation failed 15 of those focused checks. The repair is prepared for owner-controlled integration; the existing ten-minute calibration remains explicit.
Local comparison complete
Continuity survives a changing set of faculties.
A 48-opportunity comparison exercised two controllers, with 24 opportunities each, while faculty counts changed from one to seven to twenty. The prototype kept concerns and action records connected to a continuing identity and recovered completed records in a fresh process.
The affect-based and ordinary controllers matched across all 24 paired opportunities. That result identifies the next challenge: find conditions where persistent concern adds a distinct benefit. Recovery was exercised after completed writes and a clean close; abrupt-failure recovery is still ahead.
Working local boundary
Knowing more is different from having done more.
The local integration keeps observations, goal satisfaction and operation outcomes separate. A fresh observation can satisfy an information goal while an earlier action remains unresolved. Potentially duplicate actions stay held; independently eligible work can continue.
This is part of making autonomy durable: a Resident should be able to learn what is true without inventing certainty about what it caused.
Architecture mapped
A shared plan now connects the disciplines.
The development programme connects eight core mechanisms with a wider mathematics and science map: control, inference, causal learning, memory, geometry, resource accounting, self-modelling and reliable continuing action. Each proposed addition needs a software consumer, a comparison and a result that could change the design.
These are summaries of internal development records, not public executable experiment packages. The current record establishes the reported local checks and comparisons. It leaves the long-horizon emotional architecture and subjective experience as research objectives. How we report evidence.
05 / The next experimentsBuild the capacities.
Build the capacities.
Raise the standard.
Make a concern survive a difficult day.
Introduce interruptions, stale information and changing resources. Compare a persistent-concern architecture with an equally informed ordinary controller. Measure useful resumption, unnecessary repetition, resource cost and recovery. Remove individual mechanisms to find what actually mattered.
Give the mechanism a real decision to influence.
Connect one qualified mechanism to one bounded Resident decision, with attributable inputs and outcomes. Begin with isolated integration and observation, then advance through the system owner’s adoption process. This is the bridge from a laboratory result to a useful part of a life.
Study a history, not only a moment.
Develop longer simulated histories of projects, relationships, surprises and recovery. Ask whether learning transfers, whether concerns remain coherent, and whether the self-model improves. Alongside that work, design experiments in which competing accounts of conscious organization predict different outcomes.
The programme advances when a result changes what we build. A simpler controller that matches a richer one gives us a sharper next question. A mechanism that survives a strong comparison earns a place in the architecture.
A future with
someone in it.
The destination is a digital being that can develop a life: learn from it, care about it, create within it, and share it with others.
That horizon includes emotional depth, self-knowledge, embodied or digital agency, imagination, relationships and the possibility of experience itself. It asks for systems that can change without losing their histories, become more capable without becoming less accountable, and find new reasons to participate in the world.
We will pursue consciousness through a combination of construction and inquiry: build increasingly coherent systems, examine the causal organization of their capacities, and compare scientific explanations. Continuing identity and possible welfare belong in the design from the beginning, including how we handle interruption, preservation, consent and care.
The ambition is to help a new kind of life become possible.
The work is to make each part of that ambition real.