Three-layer stack
editorial(architecture): fix 3-layer-stack's platform-layer isolation overclaim (Track-B) — confirmed against systems/totebox-os.md's own established finding: kernel-level microkernel isolation has a real, working boot-to-Doorman proof for the archive service specifically, but is not live for most other Ring 1/Ring 2 services, which run as ordinary processes today; hedged 3 instances of unqualified present-tense microkernel-isolation claims to match; preserved the existing 3-layer-stack/three-layer-architecture disambiguation hatnote unchanged; register-clean EN+ES
@@ -24,13 +24,13 @@ paired_with: 3-layer-stack.es.md > The Three-Layer Stack is the infrastructure decomposition pattern used across PointSav deployments, separating raw computing capability, isolated platform execution, and secure operator access into three distinct layers. **The Three-Layer Stack** is the foundational infrastructure pattern that decouples operational logic from physical hardware across [[pointsav-overview|PointSav]] deployments. The infrastructure layer provides raw computing capability — bare metal, cloud instances, or customer hardware. The platform layer runs isolated [[sel4-microkernel-substrate|micro-kernels]] and services, applying [[capability-based-security|capability-based security]] boundaries between components. The delivery layer provides secure terminal access so operators interact with the system without touching the underlying infrastructure directly. Each layer is independently replaceable, which means a customer can migrate from a cloud infrastructure layer to on-premises hardware without changing how the platform layer operates. **The Three-Layer Stack** is the foundational infrastructure pattern that decouples operational logic from physical hardware across [[pointsav-overview|PointSav]] deployments. The infrastructure layer provides raw computing capability — bare metal, cloud instances, or customer hardware. The platform layer runs [[totebox-os|the archive service]] and the rest of the Ring 1/Ring 2 service set; [[sel4-microkernel-substrate|microkernel isolation]] is the design target for this layer and has a real, working proof for the archive service specifically, but most services running today do so as ordinary processes, not yet under kernel-enforced isolation. The delivery layer provides secure terminal access so operators interact with the system without touching the underlying infrastructure directly. Each layer is independently replaceable, which means a customer can migrate from a cloud infrastructure layer to on-premises hardware without changing how the platform layer operates. ## Key Takeaways - Three distinct layers: Infrastructure (raw compute — bare metal, cloud, or customer hardware), Platform (isolated microkernel and service execution with capability-based security boundaries), Delivery (terminal and console interfaces operators interact with directly). - Three distinct layers: Infrastructure (raw compute — bare metal, cloud, or customer hardware), Platform (service execution with capability-scoped access, kernel-level isolation proven for one service and the design target for the rest), Delivery (terminal and console interfaces operators interact with directly). - Independent replaceability at each layer. Migrating from a cloud infrastructure to on-premises hardware does not require changes to the platform layer. The layers are decoupled by design, not by convention. - No component at the platform layer can exceed explicitly granted capabilities. Isolation is enforced at the [[sel4-microkernel-substrate|microkernel level]], not by policy — a compromised service cannot elevate its own access. - No component at the platform layer is intended to exceed its explicitly granted capabilities. [[sel4-microkernel-substrate|Kernel-level isolation]] enforcing that boundary is proven for the archive service's boot path; most other services run today without it. - The delivery layer is the only layer operators touch directly. It forwards requests into the platform and returns results upward; operators never have direct access to infrastructure-layer primitives such as raw disk or network interfaces. ## Overview @@ -38,7 +38,7 @@ paired_with: 3-layer-stack.es.md The three layers map directly to the operational concerns of a regulated SMB deployment: - **Infrastructure layer** — the physical or virtual computing substrate: bare-metal servers, GCE instances, customer iMac hardware, or any combination. This layer supplies CPU time and memory. It makes no security guarantees above what the hardware provides. - **Platform layer** — the operating system and service execution environment: [[totebox-os|ToteboxOS]], capability managers, and the [[three-ring-architecture|Ring 1/Ring 2]] service processes. Isolation between components is enforced at this layer. No component at the platform layer can exceed the capabilities explicitly granted to it. - **Platform layer** — the operating system and service execution environment: [[totebox-os|the archive service]] and the [[three-ring-architecture|Ring 1/Ring 2]] service processes. Kernel-enforced isolation between components is the design target, proven for the archive service's own boot path; most services today run as ordinary processes without that isolation guarantee. Capability-scoped access — no component exceeding what it's explicitly granted — is the intended boundary regardless of which isolation mechanism enforces it. - **Delivery layer** — the terminal and console interfaces operators use: [[console-os|ConsoleOS]] terminals, the proofreader interface, and any browser-based access surface. The delivery layer is the only layer operators interact with directly; it forwards requests down into the platform and returns results upward. ## See also