railiance-platform/ArchitectureBlueprint.md
codex cbc961ca09
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 1s
Coherence review: Railiance uses one of six OAS dimensions
The central finding. OAS defines six canonical dimensions and states that each
architecture description MUST use them. Railiance has modelled itself almost
entirely on Stack. Everything that did not fit a stack level was treated as an
anomaly - "unplaced", "beside the stack" - and accumulated as exceptions.

They were never anomalies. They are concerns on dimensions Railiance was not
using, and the Quality dimension's sub-levels map almost one-to-one onto the
capability gaps this review found independently: Q2 Observability is
railiance-telemetry, named in canon in exactly those words; Q7 Governance is
the conformance loop; Q3 is restore proof; Q6 is cost attribution. The
self-evidencing thread across five stack layers is five Stack repos each
independently asking for Q2 and Q7 - what a missing dimension looks like from
inside the one you are using.

Also records four contradictions (C1-C4), of which C1 is actionable here: the
hub attributes ~11 capabilities to this repo including Terraform, Ansible, k3s,
CI/CD and app deployment, which S3 does not own. SCOPE.md declares four, all
correctly S3, and is authoritative.

ArgoCD decision closed: keep it, documentation stays accurate; relocating GitOps
to Helix Forge noted as possible future cleanup.

SCOPE.md: corrects the "five independent repos per OAS Stack layer" claim,
records the ArgoCD deployment path, the telemetry emission relationship, and the
hub capability drift.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-11 22:53:15 +02:00

16 KiB
Raw Blame History

Railiance Architecture Blueprint

The structural backbone of Railiance: the axes along which repositories are classified, the layers of the stack, and the rules that keep them from collapsing into each other.

Status: working blueprint, established 2026-08-11. Structural claims here are verified; items marked OPEN are undecided and name their deciding repo. This file records structure, not runtime state — for what is currently deployed, see each repo's SCOPE.md.

Authority note: railiance-master owns the repo-family model and the-custodian owns canon. This blueprint consolidates and applies them from the S3 perspective; where the two disagree, they win.


1. Four classification axes

Railiance classifies repositories along four orthogonal axes. They are complementary, not competing — a workload has a position on all four.

Prefix Axis Unit Answers
railiance-* ownership and responsibility one architectural responsibility area who owns this capability or layer?
rail-* execution contract one workload execution architecture how does a workload run here?
rapp-* managed workload package one Railiance-managed workload what exactly is packaged and operated?
reef-* substrate boundary one durable substrate reality where does this run, and what is bound there?

Source: railiance-master/docs/repository-axes.md.

The separation rule

Independent perspectives stay in separate dimensions. This is normative, not stylistic — OAS P1, canon/standards/orthogonal-architecture_v1.0.md: "Architecture descriptions MUST separate independent perspectives into orthogonal dimensions."

Three groupings are routinely confused and must not be derived from each other:

Grouping Answers Cardinality to repos Changes
Forgejo org who may push; what is discoverable together 1:many slowly (renames break clone URLs, CI, GitOps sources)
State Hub domain which strand of work, for attention and priority 1:many occasionally
Rapp context what deploys, versions and rolls back together many:many often

A repo lives in exactly one Forgejo org — it is a path segment in the clone URL. A repo may contribute to several rapps. A many:many grouping cannot be derived from a 1:many one. Decision d07ee5f9.

The cardinality that makes enforcement possible: repos are many:many with rapps, but deployables are 1:1 — each running deployable has exactly one rapp owning its rollout. That is what makes the coverage check well-formed: does every live deployable belong to exactly one rapp?


2. The stack

The OAS Stack dimension defines five levels. Railiance implements them as responsibility repos, plus cross-cutting and meta concerns that sit beside the stack rather than inside it.

Level Repo Owns
S1 railiance-infra OS provisioning, hardening, server baseline, at-rest bootstrap secrets
S2 railiance-cluster k3s, Helm, ingress, CNI, admission controllers, operators, kubeconfig
S3 railiance-platform shared stateful services: data, cache, secrets, object storage, messaging
S4 railiance-enablement paved paths: CI/CD templates, SDKs, portal, promotion conventions
S5 railiance-apps user-facing application workloads and release operations

Repos that are not stack levels are not homeless — they belong to other OAS dimensions (§2.1):

Repo Dimension Owns
railiance-telemetry Quality / Q2 Observability metrics, logs, traces, alerting — the substrate other layers emit into
railiance-bootstrap lifecycle (pre-stack) the ignition sequence: bare VM → working Railiance, then handoff
railiance-master the architecture description itself repo taxonomy, framework ADRs, boundary rules, vocabulary
railiance-fabric Capability repo → service → capability → interface → dependency model
railiance-forge being placed (§5.2) workload → rapp-forgejo; layer responsibility → S4

2.1 Railiance uses one of six canonical dimensions

This is the central coherence finding of the 2026-08-11 review.

OAS defines six canonical architecture dimensions, and states that "Each architecture description MUST use these dimensions" (canon/standards/orthogonal-architecture_v1.0.md §4):

Dimension Purpose Railiance use
Stack technical hosting substrate used — S1S5, the whole model to date
Logic functional decomposition unused
Plane operational responsibility unused
Quality cross-cutting architecture properties unused — and this is where the gaps are
Capability stable capability modeling partial (railiance-fabric, capability registry)
Intelligence degree of intelligent automation unused

Railiance has modelled itself almost entirely on the Stack dimension. Anything that did not fit a stack level was treated as an anomaly — "unplaced", "beside the stack", "cross-cutting" — and accumulated as exceptions.

They were never anomalies. They are concerns on dimensions Railiance was not using. The Quality dimension's sub-levels map almost one-to-one onto the capability gaps found independently by this review:

Quality sub-level Canon definition Railiance concern State
Q1 Security and Compliance credential custody, OpenBao lanes, CCR approval implemented (S3), never named as Q1
Q2 Observability "telemetry and monitoring: metrics, logs, traces" railiance-telemetry seeded 2026-08-11
Q3 Operability and Resilience restore drills, rollback proof split across S1/S3/S5, unowned
Q4 Isolation and Tenancy rapp-postgres ADR-0001, tenant keying implemented per-consumer, not modelled
Q5 Performance and Scalability unaddressed
Q6 Cost and Efficiency resource-control, fin-hub, cost attribution keys outside the family, boundary undeclared
Q7 Governance and Change Management "policies and architecture governance" the conformance loop unowned (§5.3)

Two consequences worth stating plainly:

  1. railiance-telemetry is not an exception to the stack model — it is Q2 Observability, named in canon in exactly those words. Its placement question (§5.1) is effectively answered; railiance-master need only ratify it.
  2. The conformance loop is Q7, not a missing stack layer. That reframes §5.3 from "invent a home" to "implement the Quality dimension."

The recurring "self-evidencing" aspiration across five stack layers (§3) is, in OAS terms, five Stack-dimension repos each independently asking for Q2 and Q7. That is what a missing dimension looks like from inside the one you are using.

The layering contract

Each layer hands the next a verified substrate and does not reach past its neighbours. The handoffs are declared from both sides — S4 names its forge contract for runner labels and registry endpoints; S5 names backup and data-ownership handoffs with S3. Preserve this property when editing any INTENT.md: a handoff stated by only one side is a latent boundary dispute.


3. The self-evidencing thread

Five layers independently declare the same aspiration in their Direction of Evolution:

  • S1 — "self-evidencing, auditable provisioning"
  • S2 — "self-evidencing, auditable runtime state"
  • S3 — "self-healing and observable platform services"
  • S5 — "self-evidencing, reviewable application readiness"
  • forge — "self-evidencing forge health and readiness checks"

This is the architectural through-line of Railiance: the system should prove its own state rather than be asserted correct. It reached five INTENT files independently and was never named as a fleet-wide quality dimension.

It has two halves, and they belong to different repos:

Half Question Home
Evidence plane is the running system behaving as expected? railiance-telemetry
Conformance loop do declarations match reality and each other? OPEN (§5.3)

Both are required. Observability without conformance checking finds outages but not drift; conformance checking without observability finds drift but not outages.


4. Repository status

Not every railiance-* directory is a canonical repository. Verified 2026-08-11 by comparing remotes, HEADs, and working trees.

Repo Status Evidence
railiance-infra canonical S1 active; strict content superset of hosts
railiance-hosts superseded distinct remote, unrelated history, byte-identical INTENT.md, zero unique files; quiet since 2026-07-30
railiance-cluster canonical S2 active; has install/ and rail-kubernetes extraction docs
railiance-bootstrap superseded as S2; repurpose candidate distinct remote, unrelated history, declares itself S2 verbatim; frozen 2026-07-08. Carries QUICKSTART.md (newcomer path) — candidate home for seeding a new Railiance
railiance-platform canonical S3 active
railiance-enablement canonical S4, aspiration-only 25 commits, 0 workplans; widest aspiration-to-reality gap in the family
railiance-apps canonical S5 active
railiance-telemetry seeded 2026-08-11 INTENT.md/SCOPE.md written; no implementation
railiance-master canonical meta active
railiance-fabric canonical meta quiet since 2026-07-30; non-standard INTENT.md format
railiance-forge canonical, unplaced active workplan; not on any stack level

The duplication pattern

bootstrap, cluster, hosts, infra have mutually unrelated git histories but near-identical content. They were created by copying working trees into fresh repos as the S1/S2 layering idea emerged — not by forking. Two lineages, each with a frozen elder and an active successor.

Consequence for tooling: fleet inventories built by scanning directories overcount. Deduplicate by origin URL, not by directory name.


5. Open decisions

Each names the repo that decides it. None should be settled in railiance-platform.

5.1 Where does railiance-telemetry sit? → railiance-master

Observability is cross-cutting: it draws signal from every level and is consumed by all of them. Is it an S3 platform capability (it runs stateful services and is consumed by others), or a Quality-dimension concern outside the Stack dimension? Stated explicitly at seeding so it does not drift into §5.2's state.

5.2 Where does railiance-forge sit? → DECIDED 2026-08-11

Place it, do not retire it. Operator decision:

  • the Forgejo workload becomes rapp-forgejo (a managed workload package)
  • the layer responsibility — runner placement and labels, registry retention, artifact lifecycle, package credentials — folds into S4 railiance-enablement, which already declares the handoff contract for exactly these concerns

This also gives S4, currently the widest aspiration-to-reality gap in the family, its first concrete owned responsibility.

Implementation is railiance-master and railiance-forge territory; this blueprint records the decision, not the migration.

The retirement hypothesis was tested and rejected: railiance-forge and helix-forge are different lineages. helix-forge is methodological (capability-first ecosystem, OAS dimensions, VSM vocabulary); railiance-forge is operational (helm/, manifests/, releases/, runner/, live workplan on Forgejo resource economics). Retiring it would orphan runner placement, registry retention, and artifact lifecycle. Place it, do not retire it. The likely resolution: the Forgejo workload becomes rapp-forgejo; the layer responsibility folds into S4, which already declares the handoff contract.

5.3 Who owns the conformance loop? → railiance-master

Candidates: S4 (the paved-path layer), railiance-master (the definition home), or a new responsibility repo. Related: the family declaration validator proposed to railiance-master (04c776c4, amended f88f938d).

5.4 Retirement and repurposing → railiance-master, then the repos

  • retire railiance-hosts (content-safe; check git history before deletion, and move its live hub goal "Secure Single-Server Bootstrap at HostEurope" to railiance-infra first, or it resurfaces as unstarted S1 work)
  • repurpose railiance-bootstrap as the seed-a-new-Railiance repo, consolidating the newcomer path that currently exists in both it and railiance-cluster
  • delete the local ~/railiance-bootstrap directory, which is a mis-cloned copy of railiance-cluster.git

5.6 Does Railiance use ArgoCD? → DECIDED 2026-08-11: keep it

Briefly questioned as possibly Helix Forge tooling rather than a Railiance concern. Resolved on the evidence: S3 runs ArgoCD as its live deployment mechanism — four Application manifests (external-secrets, issue-core, openbao-secretstore, target-revenue), AppProjects under argocd/bootstrap/, docs/argocd-gitops.md, and its two most recent commits add Applications.

ArgoCD stays and its documentation stays accurate. Relocating GitOps tooling to Helix Forge is noted as possible future cleanup, not current work.

5.5 Rapp granularity and composition → decided, implementation open

Grouped by bounded context; members declared explicitly; grouping legitimate only where members share rollout and rollback fate (7a1f7a7e). A rapp declares a composition: first-party member repos plus pinned upstream components plus a stated purpose.


6. Known gaps, mapped to dimensions

Read with §2.1: most of these are not scattered omissions but one unbuilt dimension.

Gap Dimension State
Observability Q2 home created 2026-08-11; no implementation, no monitoring namespace in-cluster
Conformance loop Q7 unowned (§5.3)
End-to-end restore proof Q3 split across S1, S3, S5; no single owner; backing object-storage decision escalated and unresolved (9c21c0e0)
Cost and resource attribution Q6 resource-control and fin-hub hold it; no railiance-* INTENT declares the boundary
Performance and scalability Q5 unaddressed
Identity dependency declaration Capability net-kingdom referenced in prose; rarely declared structurally. railiance-fabric exists to fix this

6.1 Contradictions to resolve

Found by the 2026-08-11 coherence review. These are places where two parts of Railiance currently say different things.

# Contradiction Resolution
C1 Capability attribution crosses layer boundaries. State Hub attributes ~11 capabilities to railiance-platform, including Terraform server provisioning and Ansible hardening (S1), k3s provisioning and cluster addons (S2), CI/CD automation (S4), and application workload deployment (S5). SCOPE.md declares only four, all correctly S3, and the repo's boundary rule disowns the rest. These are pre-split attributions the hub never dropped. SCOPE.md is source of truth; the hub needs reconciliation. Until then, hub capability data for this repo is not trustworthy for boundary reasoning.
C2 "Five independent repos per OAS Stack layer" — stated in SCOPE.md — contradicts the four-axis model and the current inventory (§4). Corrected in SCOPE.md 2026-08-11.
C3 One dimension in use, six required by ratified canon (§2.1). Route to railiance-master and the-custodian; implement Quality first.
C4 Ecosystem modelling is split four waysrailiance-fabric (graph), State Hub (read model), reuse-surface (capability federation), railiance-master (vocabulary). Fabric draws the boundary crisply in prose but does not mention reuse-surface. Needs an explicit statement of who owns what; railiance-master and railiance-fabric territory.

7. Sources

  • railiance-master/docs/repository-axes.md — the four families
  • railiance-master/docs/repo-family-bootstrap-contract.md — required declarations
  • railiance-master/docs/rapp-first-wave-candidates.md — extraction criteria and sequencing
  • the-custodian/canon/standards/orthogonal-architecture_v1.0.md — OAS, P1
  • railiance-platform/docs/rapp-platform-service-pattern.md — S3 platform-service pattern
  • railiance-platform/history/2026-08-11-railiance-architecture-in-aspiration.md — the INTENT-based assessment behind this blueprint
  • decisions 7a1f7a7e (rapp granularity), d07ee5f9 (dimensional separation)