The New Capabilities page introduces capabilities that emerged from taking Cyber-Safety principles and operationalizing them at runtime: a category map of what each capability enables, what it owns, what it does not own, and where its evidence or reference path lives.
NovaFuse names categories only when the boundary is explicit. Each category must say what it controls, what it refuses to control, which neighboring categories it depends on, and what reference evidence supports it.
Something that can now be governed, executed, remembered, proven, protected, or adopted differently.
The ownership contract that prevents one layer from absorbing another layer's job.
A public repo, Resource Center document, implementation reference, or internal canon path that anchors the claim.
NAGS — the NovaFuse Atomic Governance System — is the system-level architecture for governing consequential digital effects. It is not a tenth category or a fifth runtime engine. It organizes the discipline, lifecycle, decision model, executable rules, release authority, composition rule, runtime mechanisms, capability taxonomy, and product portfolio into one operating model.
Cyber-Safety defines the discipline. Mechanical AI Alignment applies that discipline to AI-proposed effects. PEPRG governs before, during, and after execution. IDNA evaluates Identity, Data, Networks, and AI together.
ERI makes claims executable. ERIL is the Executable Reference Implementation Language for admissibility rules. CERI is the Containerized Executable Reference Implementation package; after declared conformance is satisfied and evidence accepted, it may carry the phase-qualified designation Certified Executable Reference Implementation. DREA is the final release authority. FICT requires every mandatory federated condition to hold without weakening.
NovaZK-ID resolves bounded authority. NovaAlign contributes an AI invariance judgment. NovaCore coordinates admissibility and routing. NovaDust enforces bounded containment and egress refusal.
The nine-category map classifies responsibilities. The wider NovaFuse portfolio contains implementations, ERIs, Proof Runs, and commercial capabilities that specialize or compose those categories.
These categories are structurally distinct. Cyber-Safety starts the stack, runtime evidence proves the motion, and each adjacent category owns a different part of governed digital capability.
Invariance-based runtime refusal, safe-execution enforcement, and Mechanical AI Alignment through NovaAlign.
Refusal, invariant enforcement, safe-execution boundaries, and the Mechanical AI Alignment doctrine.
Admissible action flow after refusal gates, authority proof, or centralized orchestration.
NovaAlign, ERI/ERIL, NovaZK-ID, and NovaCore API.
Pre-execution admissibility and bounded executable reference behavior.
Whether an action is admissible to execute and how claims become executable reference surfaces.
Identity proof, authority binding, or centralized runtime orchestration.
Cyber-Safety, NovaZK-ID, and NovaCore API.
Proof of authority without disclosure; capability and identity binding.
Who may cause action, capability binding, and identity/authority proof boundaries.
Pre-execution semantics, runtime orchestration, or visualization.
ERI/ERIL, NovaCore API, and Cyber-Safety.
Single governance locus, orchestration, evidence, routing, and policy application.
Unified control and commit locus, governance routing, evidence coordination, and policy application.
The correctness of the safety doctrine, semantic admissibility model, or authority proof itself.
Cyber-Safety, ERI/ERIL, NovaZK-ID, NovaVision, NovaDust, NovaMemX, and PDSI-LLM / NovaCortex.
Perception resolution and visualization surface: UI eliminated, perception resolved.
Resolved perception, visualization semantics, and interface-to-perception translation.
Control-plane decisions, authority proof, or confidentiality enforcement.
NovaCore API, NovaMemX, and PDSI-LLM / NovaCortex.
Bounded pre-effect containment, egress refusal, and confidentiality enforcement. Zero-egress results apply only to declared test profiles.
Declared egress checks, containment outcomes, and confidentiality enforcement inside a specified runtime boundary.
Universal non-leakage guarantees, all covert channels, admissibility decisions, authority proof, or runtime orchestration.
NovaCore API, Cyber-Safety, NovaAlign, and NovaZK-ID.
NovaMemX provides hysteresis-style persistence and long-horizon continuity.
Persistence across time, temporal continuity, and memory-state threading.
Domain reasoning, authority proof, or centralized control-plane decisions.
PDSI-LLM / NovaCortex, NovaCore API, and Cyber-Safety.
PDSI-LLM / NovaCortex owns governed domain reasoning persistence.
Persisted reasoning inside domain envelopes and governed reasoning workflow surfaces.
Long-horizon continuity storage by itself, authority proof, or unified runtime control.
NovaMemX, NovaCore API, and NovaZK-ID.
Zero-migration adoption model where customer systems conform to the control plane instead of being replaced.
Deployment posture, integration strategy, and adoption without forced platform migration.
Control-plane behavior, intelligence engines, or authority proof.
NovaCore API, Cyber-Safety, IDNA, and DREA runtime enforcement references.
The Cyber-Safe Stack keeps responsibility from drifting. Cyber-Safety refuses unsafe motion. Mechanical AI Alignment and NovaAlign bound AI-proposed effects. ERI/ERIL decides executable admissibility. NovaZK-ID proves authority. NovaCore coordinates the control locus. NovaMemX remembers across time. PDSI-LLM / NovaCortex reasons inside domain envelopes. None of those should silently absorb the others.
The responsibility the category is accountable for.
The neighbor responsibility it must not blur or absorb.
The public repo, Resource Center material, implementation reference, or canon file that anchors the category.
The Resource Center carries public documents and public repositories. The repository canon carries implementation references where a category exists in the checkout but is not yet packaged as a public standalone release.
| Category | Primary Public Reference | Implementation Evidence Path |
|---|---|---|
| Cyber-Safety | Cyber-Safety page, Paradigm Paper, Operational Reference, public repo | nova-cybersafe-framework/ |
| Semantic Execution Systems | ERI/ERIL/CERI page, ERI repo, ERIL repo | NovaFuse-ERI/spec, nova-eri-framework |
| Decision Authority Infrastructure | Resource Center coordination; public standalone release pending. | NovaZK/NovaZK-ID/Capability-Registry.md |
| Unified Deterministic Control Plane | Resource Center coordination; implementation reference in checkout. | novacore/api, novacore/cyber-safety |
| Deterministic Perception Engine | Implementation reference in checkout. | codebase-map.json, demo-novacore-tests.js |
| Deterministic Confidentiality | Implementation reference in checkout. | docker-compose.novatron-fixed.yml, alignment-invariance-battery-assessment.md |
| Continuity Memory | Featured implementation reference; public standalone release pending. | src/novamemx/coherence_audit.py, src/novamemx/temporal_threading.py |
| PDSI-LLM / NovaCortex | Implementation reference in checkout. | action.yml, docker-compose.novacortex-chat.yml |
| Non-Invasive Runtime Adoption | IDNA Reference Model, runtime refusal references. | NovaFuse/runtime/DREA-RUNTIME-ENFORCEMENT.md |
Internal canon path: NovaFuse/CATEGORY_CANON.md. Naming standard: PDSI-LLM. Avoid the transposed acronym variant.