The applications share a cognitive substrate, but their implementation depth, evidence and next proof obligation differ. Maturity is recorded per domain rather than averaged into a platform-wide label.
These states describe current integration depth. They are not a ranking of domain importance or a promise of production readiness.
Application field atlas
The adapter changes. The accountability requirement does not.
Each record names the domain problem, mechanisms, adapter, verifier, implemented scope, evidence, limitation and next milestone. The records intentionally remain asymmetric.
I
Formal and computational truth conditions
Autonomous research
Persistent hypotheses, evidence, gaps, verifiers and campaign state for long-running mathematical and scientific exploration.
Maturity
Research runtime
Evidence status
Operational observation
Truth condition
Formal and computational truth conditions
Domain problem
Long investigations fragment their hypotheses, failed approaches, unresolved gaps and verification state across many model calls and artifacts.
NSP mechanisms used
01Persistent typed research state
02Knowledge gaps and pursuit state
03Evidence provenance and recall
04Producer–auditor–verifier separation
05Pre-action gating for consequential steps
Domain adapter
A research-domain encoding adapter maps problems, hypotheses, experiments, constraints and verifier outcomes into the shared structural substrate.
Verification regime
Formal proof when available
Deterministic computation and experiment sentinels
Explicit verifier state and evidence trails
Human review where formal closure is unavailable
Implemented scope
A running research environment for structured pursuits, knowledge gaps, anchors, recall, campaign state and bounded outcome evaluation.
Evidence available
Operational campaign records plus bounded experiments reported in public preprints. This is one-system evidence, not independent replication.
Verifier coverage is finite and extracted constraints may still require independent checking.
No second backend has reproduced the reported research behaviour.
II
Tests, tools and runtime behaviour
Reliable coding agents
Understanding gates, architecture context and action-aware safeguards around coding work.
Maturity
Operational integration
Evidence status
Engineering record
Truth condition
Tests, tools and runtime behaviour
Domain problem
A coding agent can modify a system before its interpretation, assumptions and affected architecture are explicit enough to inspect.
NSP mechanisms used
01Understanding and preview gates
02Confidence and assumption state
03Architecture and convention context
04Persistent memory and error history
05Action-aware rules before consequential tools
Domain adapter
A coding-domain adapter maps files, tool calls, failures, tests and architecture context into addressable state around the agent workflow.
Verification regime
Build and test outcomes
Type and lint checks
Runtime behaviour and tool failures
Human code review for uncovered semantics
Implemented scope
A hook-based operational integration with understanding state, architecture context, memory, history, rules, failure capture and an inspectable viewer.
Evidence available
Internal engineering records and continuous operational use. They demonstrate deployment behaviour, not a controlled quality-improvement study.
Known limitations
Process safeguards do not prove generated code correct.
Verifier quality is limited by the project tests and checks that exist.
The full axiom-application and cross-project transfer loop is not yet integrated.
III
Subjective, noisy and perspective-dependent feedback
Creative and interactive systems
Perspective-aware cognition, evolving beliefs, persistent traces and domain-specific expression.
Maturity
Experimental
Evidence status
Working hypothesis
Truth condition
Subjective, noisy and perspective-dependent feedback
Domain problem
Creative quality and interactive continuity depend on perspective, accumulated judgement and expression rules that rarely have a single formal oracle.
NSP mechanisms used
01Perspective-aware state
02Belief and evidence updates
03Persistent experience traces
04Domain axiom accumulation
05Expression adapters with feedback
Domain adapter
Creative encoding and expression adapters map domain structure, version history and human verdicts into persistent state without treating taste as objective truth.
Verification regime
Structural and rule-based checks
Disclosed human-listener or evaluator verdicts
Version-to-version comparison
Negative, rejected and null outcomes retained
Implemented scope
Experimental state and expression adapters for selected creative domains; music composition is the current documented internal example.
Evidence available
A private working-paper record and bounded internal composition trials. The present status remains a working hypothesis.
Known limitations
Current music verdicts come from one listener.
The rendering pipeline sets a production-quality ceiling.
Cross-domain comparisons use asymmetric evidence and remain partly qualitative.
Shared core / different truth regimes
Domain adaptation begins where truth conditions diverge.
The common substrate keeps entities, relations, evidence, constraints and persistent cognitive state inspectable. An encoding adapter decides what a domain signal means; its verifier decides what can count as correction.
Problems, hypotheses, experiments and formal objects
Primary correction signal
Proof, computation and explicit verifier state
II
Reliable coding agents
Encoding / expression adapter
Files, tools, failures, tests and architecture context
Primary correction signal
Builds, tests, types, runtime behaviour and review
III
Creative and interactive systems
Encoding / expression adapter
Perspective, versions, structural rules and human verdicts
Primary correction signal
Structural checks plus disclosed subjective evaluation
Scope boundaries
What these application records do not claim.
A maturity label is useful only when the excluded claim remains visible beside it.
01
Not three production products
The three fields do not share one release state, customer promise or validation regime.
02
Not one universal verifier
Formal proof, software tests and human aesthetic judgement are not interchangeable evidence.
03
Not guaranteed correctness
Persistent state and action gates improve inspectability; they cannot close gaps in models, tests or human review.
04
Not cross-implementation proof
The current evidence comes from one cognitive-engine family and does not establish backend-independent generality.
Next proof obligations
Each field advances by closing a different gap.
The next milestone is stated as an evidence obligation, not a feature promise.
I
Autonomous research
Run dense campaigns in additional domains and reproduce the architecture on an independent backend.
II
Reliable coding agents
Integrate the full cognitive-engine learning loop and measure bounded cross-project quality improvement.
III
Creative and interactive systems
Run blind multi-listener evaluation, improve the rendering pipeline and repeat structured campaigns on additional creative surfaces.
Collaboration
Bring a domain problem and its correction signal.
Useful collaboration starts with a bounded operating condition: what state must persist, what action matters, what evidence is available and who or what can verify it.