# 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** — S1–S5, 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.

### 2.2 The six dimensions, established

Railiance's position on each canonical dimension, with owners where they exist.
Sub-levels are canon's, not invented here. **Status** is honest about what is
implemented versus merely named.

#### Stack — technical hosting substrate · *in use*

S1 `railiance-infra` · S2 `railiance-cluster` · S3 `railiance-platform` ·
S4 `railiance-enablement` · S5 `railiance-apps`. See §2.

#### Quality — cross-cutting architecture properties · *the priority*

| Sub-level | Railiance concern | Owner | Status |
| --- | --- | --- | --- |
| Q1 Security and Compliance | credential custody, OpenBao lanes, CCR approval, grant catalog | S3 | **implemented**, never named Q1 |
| Q2 Observability | metrics, logs, traces, alerting | `railiance-telemetry` | seeded 2026-08-11, no implementation |
| Q3 Operability and Resilience | restore drills, rollback proof, DR | — | split S1/S3/S5, **unowned** |
| Q4 Isolation and Tenancy | consumer isolation, tenant keying | `rapp-postgres` (per-consumer) | implemented, not modelled fleet-wide |
| Q5 Performance and Scalability | — | — | **unaddressed** |
| Q6 Cost and Efficiency | cost attribution, unit economics | `resource-control`, `fin-hub` | outside the family, boundary undeclared |
| Q7 Governance and Change Management | the conformance loop, policy enforcement | — | **unowned** (§5.3) |

Q1 is worth dwelling on: S3 has been building a mature Q1 practice for months —
approval gates, delegated apply, revocation semantics, front-door readiness —
without ever naming the dimension. That is evidence the dimensions describe real
structure rather than imposing vocabulary.

#### Plane — operational responsibility · *implicit, unnamed*

| Sub-level | Railiance reality |
| --- | --- |
| P1 Workload Plane | `rapp-*` packages and the workloads they run |
| P2 Control Plane | ArgoCD, CNPG operator, External Secrets, cert-manager |
| P3 Management Plane | State Hub, `railiance-master`, workplans, decisions |

Railiance already operates all three; none is declared. This is the dimension
most likely to clarify the `rapp-*` / `rail-*` relationship, since a rail is
essentially a P1 contract.

#### Capability — stable capability modeling · *partial*

C1 Capability Model / C2 Contract / C3 Realization / C4 Shared Platform
Capabilities / C5 Business Capabilities.

`railiance-fabric` models C1–C3; the per-repo `registry/capabilities/`
declarations and `reuse-surface` federation cover C4. C5 is unaddressed and is
where the boundary with `resource-control` / `fin-hub` (and Q6) belongs.
**Contradiction C1 lives here**: the hub's capability attributions do not match
repo declarations.

#### Logic — functional decomposition · *unused*

L1 Capability Domain / L2 Service Realization / L3 Composition and Integration /
L4 Solution Layer. No Railiance repo is organised on this axis. Lowest priority:
the Stack and Quality dimensions carry more weight for an operations framework,
and Logic is likely more relevant to Helix Forge's capability-first model.

#### Intelligence — degree of intelligent automation · *emerging, unnamed*

I1 Deterministic Automation / I2 Language Interaction / I3 Knowledge and
Reasoning / I4 Assisted Adaptation / I5 Agentic Delegation.

Already referenced in practice — `qonto-assistant/docs/SecurityPractice.md`
cites an "I1 Reinforced minimum" isolation profile — and the whole agent
operating model (autonomy lanes, Kaizen agents, `/ralph-workplan`) is I4–I5
territory. Declared nowhere in Railiance. Given how much of this fleet is
agent-operated, leaving Intelligence unmodelled understates what the system
actually is.

### Establishing order

Recommended sequence, for `railiance-master` to ratify:

1. **Quality first.** It holds the most unowned concerns (Q3, Q7) and the
   already-built-but-unnamed ones (Q1, Q4). Q2 has a home as of 2026-08-11.
2. **Plane second.** Cheap — it is descriptive of what already runs, and it
   sharpens the `rail-*` contract.
3. **Capability third.** Requires resolving C1 (hub attribution drift) and C4
   first, or the model is built on bad data.
4. **Intelligence fourth.** Needs a real decision about how much of the agent
   operating model belongs in Railiance versus the-custodian.
5. **Logic last**, or never, if Helix Forge owns functional decomposition.

### 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? → **REOPENED 2026-08-12 on corrected evidence**

Decided 2026-08-11 as "keep it", on the basis that S3 runs ArgoCD as its live
deployment mechanism. **That basis was wrong.** Verified against both clusters
once API access was restored:

| Fact | Evidence |
| --- | --- |
| ArgoCD does **not** run on railiance01 | namespace `argocd` NotFound |
| ArgoCD runs on **CoulombCore** | `argocd` namespace, Active 344d |
| The Applications target **CoulombCore, not railiance01** | all four declare `server: https://kubernetes.default.svc` — in-cluster |
| Their health is poor | `issue-core` Degraded, `openbao-secretstore` Degraded, `target-revenue` OutOfSync/Missing |
| The real workloads run on **railiance01** | `target-revenue` deploy 1/1 up 6d, `openbao` sts 1/1, `external-secrets` 3 deploys 1/1 |
| Nothing on railiance01 is ArgoCD-managed | no namespace carries an `argocd.argoproj.io/instance` label |

So the live S3 workloads are deployed **outside GitOps**, by Makefile and hand,
while the ArgoCD Applications describe a deployment path into the **frozen**
CoulombCore cluster — where `target-revenue` is registered but Missing, and two
others are Degraded.

The earlier claim that removing ArgoCD documentation would "describe a system
that does not exist" was backwards: **the ArgoCD documentation already describes
a deployment path that is largely not in effect.**

This does not settle the question — it restates it honestly. Three real options:

1. **Adopt properly** — stand up ArgoCD on railiance01 and bring the live
   workloads under GitOps. The Applications become true.
2. **Retire** — remove the Applications and `docs/argocd-gitops.md`, and accept
   Makefile-driven deployment as the declared method.
3. **Relocate** — GitOps becomes Helix Forge tooling, as originally suspected.

Doing nothing is the one option with a cost: a stale control plane on a frozen
cluster, showing Degraded, that no longer reflects how anything ships. Note this
is the same defect class as `RAIL-HO-WP-0009` — declared state and live state
diverged, and nothing detected it.

### 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 ways** — `railiance-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)