coulomb/coulomb-social
4
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Revise ArchitectureBlueprint with full discussion synthesis

Browse source 
Fold page-first/transclusion, dual authoring, lazy projections, stale
replace-or-keep, and open content-plane decisions into the main draft.
This commit is contained in:
tegwick 2026-08-13 00:27:44 +02:00
parent f6c9756b26
commit c27a50e6d3
1 changed files with 189 additions and 164 deletions
Download patch file Download diff file Expand all files Collapse all files

353
docs/architecture/ArchitectureBlueprint.md
View file

@ -2,10 +2,33 @@
| Field | Value |
|-------|--------|
| Status | **draft for discussion** (2026-08-12) |
| Status | **draft for discussion** (revised 2026-08-13) |
| Product | coulomb.social rebuild (`coulomb-social`) |
| Related | stage-1 cut, capability model, content-substrate exploration |
| Non-goals | Final store vendor lock-in; implementing UI chrome in this doc |
| Related | stage-1 cut, capability maps, substrate exploration, authoring, stale/derived |
| Non-goals | Final store vendor lock-in; full UI chrome; implementing markitect features in this doc |
## 0. Review summary (2026-08-13)
This revision folds in topics settled after the first draft:
| Topic | Stance |
|-------|--------|
| Identity / tenants / groups | NetKingdom only |
| Deploy | Railiance |
| Content SoR | Markdown + assets; DB = cache |
| markitect | **tool** as structure lib; **not** “run on markitect”; **main** = legacy |
| Page vs chunk | **Page-first**; chunks = address / transclude / promote |
| Authoring | Top-down **and** bottom-up; no forced mode |
| Title / Abstractor / Visual | Real fields; may start **lazy**; then persist; **regenerate + lock** |
| Stale derived content | Badge → generate **alternative** → **replace or keep** |
| Includes | Live resolve **and** (proposed) pinned snapshots with same stale grammar |
| Content plane | Still open: thin store / shard-wiki / kontextual (spikes) |
| Assets | kontextual BlobStorage now; extract asset-store when 2nd consumer |
| Live Bubble | Mapped; S1 aligns with Title/Abstractor/Visual/spaces/pages/cards |
**Still open (decisions, not re-litigate principles):** content-plane pick (A/B/C), visual storage short-term, PolicyGateway owner, chunk migration strategy.
---
## 1. Architectural aim
@ -13,23 +36,26 @@ Deliver a co-creation product where:
1. **Identity** (users, groups, tenants) is **NetKingdom** — not reimplemented.
2. **Runtime** is **Railiance** — not reimplemented.
3. **User-owned content** is **markdown (+ assets)** that can be exported and reused without the app database.
4. The **database is a cache / projection** for cards, memberships, and speed.
5. **markitect-tool** enables efficient structured markdown (lib/service), not a product host.
6. **coulomb** owns co-creation UI (playing-card pages, spaces, transfer/copy) and product verbs.
3. **User-owned content** is **markdown (+ assets)** exportable without the app database.
4. The **database is a cache / projection** for cards, memberships, search speed.
5. **markitect-tool** makes markdown an efficient structured substrate (lib/CLI/API) — **not** the product host.
6. **coulomb** owns co-creation UI (cards, spaces, copy/transfer, authoring UX) and product verbs.
7. Knowledge work stays **fluent**: top-down and bottom-up authoring; structure and reuse appear when needed (transclusion, lazy projections), not as mandatory a priori chunk graphs.
## 2. Non-negotiable rules
| # | Rule |
|---|------|
| R1 | Markdown chunks (frontmatter + body + backmatter) + assets are the **exportable SoR** for content. |
| R2 | Postgres (or any app DB) holds **projections and app linkage only** for content; disposable for bodies. |
| R1 | Markdown (frontmatter + body + backmatter) + assets are the **exportable SoR** for content. |
| R2 | App DB holds **projections and app linkage only** for content; disposable for bodies. |
| R3 | Authn/IdP = **NetKingdom**; coulomb maps OIDC → Member; no password store. |
| R4 | Authz for product actions may compose NetKingdom groups, app membership, and content-plane ACLs — **no second user directory**. |
| R4 | Authz composes NetKingdom groups, app membership, content-plane ACLs — **no second user directory**. |
| R5 | **markitect-main** is legacy reference, not a runtime foundation. |
| R6 | **markitect-tool** is the markdown structure dependency (parse/validate/transform). |
| R7 | Content-plane choice (shard-wiki / kontextual / thin store) is behind a **narrow page-ops interface** so coulomb UI does not couple to federation or ECM graphs. |
| R8 | Stage-1 capability set remains: onboarding, spaces/pages with Title/Abstract/Visual, page copy, page transfer. |
| R6 | **markitect-tool** is the markdown structure dependency (parse, validate, query, include, later derived slots). |
| R7 | Content plane (shard-wiki / kontextual / thin store) sits behind a **narrow PageOps** interface. |
| R8 | Stage-1 must: onboarding; spaces/pages with **Title / Abstractor / Visual**; page copy; page transfer. |
| R9 | **Pages first** — do not model documents as a priori chunk assemblies; use address/transclude/promote. |
| R10 | Derived content (projections, pinned includes, pipelines, AI) uses **stale badge → alternative → replace\|keep** and **lock**; no silent overwrite of locked/user content. |
## 3. System shape
@ -42,58 +68,94 @@ Deliver a co-creation product where:
▼
┌─────────────────────────────────────────────────────────────┐
│ coulomb.social (product) │
│ • Card UI, spaces chrome, stage-1 product flows │
│ • Session / Member / SpaceMembership projections │
│ • PageOps façade: list/get/create/update/archive/copy/move │
│ • Railiance deploy (Helm, env, probes) │
│ • Card UI, spaces chrome, dual authoring flows │
│ • Session / Member / SpaceMembership │
│ • PageOps + projection UX (lazy / stale / lock) │
│ • Railiance deploy │
└───────────────┬─────────────────────────────┬───────────────┘
│ │
│ projections │ content ops
▼ ▼
┌───────────────┐ ┌───────────────────────────┐
│ App DB cache │ │ Content plane (choose one │
│ (Postgres) │ │ primary adapter) │
└───────────────┘ │ A thin folder/git store │
│ B shard-wiki │
│ App DB cache │ │ Content plane (adapter) │
│ (Postgres) │ │ A thin folder/git │
└───────────────┘ │ B shard-wiki (thin) │
│ C kontextual-engine │
└─────────────┬─────────────┘
│ structure
▼
┌───────────────────────────┐
│ markitect-tool │
│ front/body/back, schema │
└───────────────────────────┘
│
┌─────────────▼─────────────┐
│ Asset bytes (see §6) │
│ content-addressed blobs │
│ parse, FM, query, include │
│ derived/stale (proposed) │
└─────────────┬─────────────┘
▼
┌───────────────────────────┐
│ Asset bytes (blobs) │
│ content-addressed │
└───────────────────────────┘
```
**Product verbs (PageOps)** — store-agnostic:
### PageOps (store-agnostic)
| Op | Meaning |
|----|---------|
| `list_pages(space)` | Card projections |
| `get_page(id)` | Full markdown + resolved visual |
| `get_page(id)` | Full markdown + resolved visual; optional include resolve |
| `put_page(...)` | Create/update attributes + body |
| `archive_page` | Soft-delete / lifecycle |
| `copy_page` | Independent duplicate |
| `copy_page` | Independent duplicate (rewrite includes as needed) |
| `transfer_page` | Re-home to another space |
| `put_visual` / `get_asset` | Binary visual/attachment |
| `project_field` / `regenerate` / `lock` | Lazy & stale projection actions |
| `derived_refresh` | Stale include/pipeline: alternative → replace\|keep (when markitect supports) |
Coulomb never talks to Forgejo/shard/kontextual APIs ad hoc from templates — only through PageOps.
Coulomb templates/services do **not** call Forgejo/shard/kontextual APIs ad hoc — only PageOps.
## 4. Information model (product)
| Concept | Representation |
|---------|----------------|
| **Page** | Markdown document: frontmatter (title, abstract, visual ref, space, type, …) + body + backmatter (relations, provenance) |
| **Space** | Container context; may be a page of `type: space` or a first-class container id in frontmatter |
| **Visual** | Asset ref in frontmatter → blob/asset service |
| **Card** | UI projection: visual + title + abstract + actions (not a storage type) |
| **Member** | NetKingdom subject ↔ app Member row |
| **Tenant** | Instance / `tenant_id` (DR-1 C) |
| **Page** | Markdown: frontmatter (title, abstractor, visual, space, projections lifecycle, …) + body + backmatter |
| **Space** | Container; may be `type: space` page or container id in frontmatter |
| **Visual** | Asset ref → blob store |
| **Card** | UI only: visual + title + abstractor + actions (+ stale badges) |
| **Region / “chunk”** | Addressable unit of a page (heading, `mkt:region`, selector) — **not** a required storage type |
| **Transclusion** | Live `mkt:include` and/or pinned `mkt:derived` snapshot (proposed) |
| **Member / tenant** | NetKingdom + app session |
### 4.1 Page-first vs Bubble chunks
Live Bubble has `custom.chunk` graphs. Rebuild stance:
```text
Write fluently in a page
→ address regions when needed
→ transclude / extract / promote
→ materialize a new page only when reuse earns independence
```
See `transclusion-and-chunks.md`. Migration may flatten Bubble chunks into sections/regions/pages+includes.
### 4.2 Dual authoring
| Mode | Flow |
|------|------|
| **Top-down** | Title → abstractor → outline (headings) → elaborate body → visual |
| **Bottom-up** | Body first; title/abstractor/visual/structure later |
No mandatory mode toggle. Empty body with metadata is valid; body-only is valid.
### 4.3 Lazy projections + lock + stale alternatives
For **title**, **abstractor**, **visual** (and similar display fields):
1. Missing on display need → **generate** (lazy) → **persist** as normal field values.
2. **Regenerate** and **lock** are per-field.
3. If inputs changed → **May be stale** → **Generate updated version** → compare Current \| Alternative → **Keep** or **Use alternative**.
Detail: `authoring-modes-and-lazy-projections.md`, `stale-derived-content.md`.
Same grammar for **pinned includes** and **pipeline/AI/rule** blocks (markitect proposal: `mkt:derived`).
## 5. Content-plane options (comparison)
@ -101,182 +163,145 @@ Coulomb never talks to Forgejo/shard/kontextual APIs ad hoc from templates — o
| | |
|--|--|
| **Idea** | Layout of `.md` files + assets; optional git; markitect validates; app indexes cards |
| **Pros** | Fastest stage-1; trivial user export; low structure surface |
| **Cons** | Reimplements copy/transfer/versioning carefully; federation later is a rewrite |
| **When** | Need product velocity; content plane maturity unclear |
| **Idea** | `.md` tree + assets; markitect validates; app indexes cards |
| **Pros** | Fastest S1; trivial export; low structure tax |
| **Cons** | Copy/transfer/versioning are app work; federation later costs more |
| **When** | Velocity first; plane maturity unclear |
### 5.2 Option B — shard-wiki (thin slice)
| | |
|--|--|
| **Idea** | Headless wiki orchestration; pages as union of shards; IdP external; coulomb = UI |
| **Pros** | Built for wiki pages, git coordination, overlays, multi-shard (personal + space); INTENT names Coulomb; no UI ownership |
| **Cons** | Early product maturity for multi-tenant ops; federation API can over-structure if not sliced thin |
| **When** | Want page federation / export shards / git-addressable coordination as first-class |
| **Thin-slice rule** | One InformationSpace per instance; one primary shard; hide federation from card UI |
| **Idea** | Headless wiki orchestration; IdP external; coulomb = UI |
| **Pros** | Page union, git coordination, multi-shard later; INTENT names Coulomb |
| **Cons** | Structure tax if federation leaks into product; multi-tenant ops still maturing |
| **Thin-slice rule** | One InformationSpace / instance; one primary shard; hide federation from cards |
### 5.3 Option C — kontextual-engine (knowledge ops / CMS-ECM backend)
### 5.3 Option C — kontextual-engine
| | |
|--|--|
| **Idea** | Headless knowledge operations: durable **KnowledgeAsset** identity, representations, metadata, relationships, workflows, audit, blob storage, optional CMIS; markdown via markitect-tool adapter |
| **Pros** | Strong identity for assets beyond path; lineage, versioning, export packages, multi-format future; **blob layer already implemented** (memory/fs/S3, content-addressed); PolicyGateway port ready for external authz; aligns with “ECM without being the UI” |
| **Cons** | Heavier than pure wiki; **not streamlined for NetKingdom** today (actors are engine-local; default `AllowAllPolicyGateway`); markdown is *a* format among many — SoR discipline must force “source representation is markdown file users can take” |
| **When** | Need governed asset registry, visuals/blobs, audit, multi-format, CMIS-ish interoperability; willing to invest PolicyGateway → flex-auth / NetKingdom |
| **Auth gap** | `Actor` + `external_ref` can map to OIDC subject; `PolicyGateway.authorize` must call **flex-auth** (or equivalent) with NetKingdom groups — not reimplement IAM inside kontextual |
| **Idea** | Knowledge ops: asset identity, representations, blobs, audit, workflows; markdown via markitect adapter |
| **Pros** | Blobs ready; lineage; multi-format future; PolicyGateway port |
| **Cons** | Heavier; NetKingdom not wired; must enforce markdown export as happy path |
| **Auth gap** | Map OIDC → Actor; PolicyGateway → flex-auth |
**kontextual already has (relevant):**
- Asset registry (create, metadata, lifecycle, versions, relationships, audit)
- BlobStorage port + local/S3/memory adapters, dedupe by sha256
- Representation kinds (source / normalized / derived)
- markitect-tool reuse boundary docs + adapter package
- Service API surface for assets/retrieval/workflows
- Policy port (pluggable); default allow-all for tests
**kontextual does not yet have (for coulomb):**
- First-class “space + card product model”
- Wired NetKingdom / KeyCape session → Actor
- Production flex-auth policy adapter
- Guaranteed “markdown file tree export is the happy path” product packaging (export exists as concept; productize for coulomb)
### 5.4 How options relate (not exclusive forever)
### 5.4 Evolution (not exclusive forever)
```text
coulomb PageOps
│
├─ short term: A or thin B/C adapter
│
└─ adapters can evolve:
folder/git shard ──► shard-wiki InformationSpace
markdown assets ──► kontextual KnowledgeAsset + BlobStorage
markitect-tool ──► always structure layer
PageOps → A or thin B/C now
→ folder/git shard can become shard-wiki
→ assets can move to kontextual / asset-store
→ markitect-tool always structure layer
```
shard-wiki optimizes **wiki federation and page union**.
kontextual optimizes **governed multi-format knowledge assets + blobs + audit**.
coulomb optimizes **co-creation product**.
**Working shortlist (unchanged):** spike **A** as control + **B thin** and **C** in parallel; pick after scorecard (`spike-checklist.md`).
A hybrid is possible later: shard-wiki as page navigation plane, kontextual as asset/blob + governance — only if spikes show value without double-structure tax.
## 6. Asset management
## 6. Asset management — extract to its own repo?
| Source | Role |
|--------|------|
| kontextual BlobStorage | Prefer for Visual/bytes now (fs/S3/memory, content-addressed) |
| markitect-main assets | Ideas only — do not runtime-depend |
| Extract `asset-store` | When **second** consumer needs same API; seed from kontextual port |
### 6.1 What exists today
| Source | Asset story |
|--------|-------------|
| **kontextual-engine** | Clean **BlobStorage** protocol + content-addressed store + representation linkage; small, intentional, portable adapters (~memory/fs/S3) |
| **markitect-main** `markitect/assets/` | Large (~0.5M+ tree) manager/registry/packager/optimizer/analytics — **orphan** in succession map; do not run coulomb on it |
| **coulomb** | No first-class visual/blob store yet (stage-1 Visual is todo) |
### 6.2 Extraction thesis
**Yes, worth evaluating a dedicated `asset-store` (name TBD) capability** if:
1. **coulomb**, **kontextual**, **shard-wiki**, and future apps all need content-addressed bytes + metadata without pulling full ECM or full wiki.
2. The core contract is small: `put/get/stat/delete_unreferenced` + digest + media type + optional ACLs.
3. We want Railiance-wide reuse (space visuals, wiki attachments, kontextual representations).
**Prefer extracting from kontextual’s BlobStorage** (clean ports), **not** from markitect-main’s asset mega-package (mine ideas only: dedupe, mdpkg packaging).
### 6.3 Suggested extraction shape (discussion)
```text
asset-store (new repo or helix capability)
ports: BlobStorage (digest-addressed)
adapters: fs, s3, memory
optional: reference accounting, GC, HTTP byte range
NOT: full asset registry, workflows, CMIS, markdown parsing
kontextual-engine
keeps KnowledgeAsset + Representation pointing at storage_ref
depends on asset-store library/service
coulomb / shard-wiki
may depend on asset-store for Visual without full kontextual
```
**Decision gate for extraction:** after coulomb stage-1 Visual needs a real store **and** a second consumer wants the same API. Until then, use kontextual blob adapters **in-process** or vendor the port as a small shared package without full service.
Visual field stores a **ref** (path or `sha256:…`); bytes never live only in Postgres.
## 7. Identity & access control
```text
Browser → coulomb (session after NetKingdom OIDC)
│
├─ product authz: SpaceMembership, tenant_id
│
└─ content plane actor:
Actor(external_ref=issuer|subject, groups=OIDC groups)
PolicyGateway → flex-auth (preferred) or engine rules
Browser → NetKingdom OIDC → coulomb session
├─ product authz: tenant_id, SpaceMembership
└─ content plane: Actor(external_ref=iss|sub, groups=…)
PolicyGateway → flex-auth (preferred)
```
| Plane | Responsibility |
|-------|----------------|
| NetKingdom | Who is the user; groups/roles claims |
| coulomb | Session; Member; space membership product rules |
| flex-auth | Optional PDP for resource actions |
| Content plane | Enforce policy on get/put content; fail closed |
**Spike required:** map one OIDC login → kontextual `Actor` → deny-by-default policy for foreign tenant assets.
Fail closed on ambiguous access. Spike: OIDC → Actor → cross-tenant deny.
## 8. Deploy sketch (Railiance)
| Component | Deploy |
|-----------|--------|
| coulomb-social | Existing chart / app.coulomb.social |
| Content plane | Sidecar or separate service in `coulomb-social` ns or shared platform ns |
| asset bytes | PVC and/or S3-compatible; secrets via existing env-secret pattern |
| coulomb-social | Existing chart; `app.coulomb.social` |
| Content plane | In-process, sidecar, or platform service |
| Asset bytes | PVC and/or S3; secrets via env-secret pattern |
| NetKingdom | Existing KeyCape / Authelia |
No apex DNS change until stage-1 cut.
Apex DNS stays Bubble until stage-1 cut + transfer readiness.
## 9. ADR roadmap (proposed)
## 9. Stage-1 product scope (capability cut)
**Must for productive transfer** (Bubble → reference-only):
- Member onboarding (NetKingdom; public reg may lag via provisioned users)
- Spaces + pages with **Title, Abstractor, Visual** + markdown body
- Page **copy** and **transfer**
- Card presentation
- Dual authoring + lazy projections (at least title/abstractor); stale replace|keep preferred early
**Later:** social, ventures/invest, challenges marketplace, monetization, full chunk-stage machine, multi-shard personal vaults.
Live Bubble map: `docs/capability/existing-bubble-capability-map.md`.
## 10. ADR roadmap
| ADR | Topic |
|-----|--------|
| ADR-0001 | NetKingdom identity (done) |
| ADR-0002 | Forgejo markdown binding (provisional slice → demote to adapter) |
| **ADR-0003** | Page-centric model + markdown SoR + DB cache (to write after spikes) |
| **ADR-0004** | Content-plane choice (A/B/C) + PageOps boundary |
| ADR-0001 | NetKingdom identity (**done**) |
| ADR-0002 | Forgejo binding — **demote** to one adapter/export shape |
| **ADR-0003** | Page-centric model + SoR + DB cache + page-first/transclusion |
| **ADR-0004** | Content-plane choice + PageOps |
| **ADR-0005** | Asset-store extraction (if/when) |
| **ADR-0006** | Authoring modes + projections + stale/derived (or fold into 0003) |
## 10. Spike checklist
Markitect-tool: implement `proposed-stale-derived-and-include-pinning.md` as its own work.
See **[spike-checklist.md](./spike-checklist.md)** — executable evaluation tasks.
## 11. Spike checklist
## 11. Discussion prompts
See **[spike-checklist.md](./spike-checklist.md)**.
Add when scheduling: dual-authoring smoke, stale projection compare UX, include pin (after markitect).
1. Is **user export as markdown tree** more important than **multi-format ECM** in the next 6 months? (Biases A/B vs C.)
2. Do we need **multi-shard federation** (personal vault + space) at stage 1, or only later? (Biases B.)
3. Should **Visual** block on asset-store extraction, or ship with kontextual/local blob first?
4. Who owns **PolicyGateway** production adapter — coulomb, kontextual, or flex-auth team?
5. Accept **hybrid**: kontextual for assets+registry, thin git for human-facing markdown tree sync?
## 12. Discussion prompts (remaining)
## 12. Working recommendation (to debate, not freeze)
1. Content plane for next 3–6 months: **A velocity**, **B federation path**, or **C governance/blobs**?
2. Visual: kontextual/local blob first, or block on asset-store extract?
3. PolicyGateway production owner?
4. Bubble **chunk** migration: sections/regions only, or keep a chunk list UX as a *view*?
5. Default include mode in product: **live** vs **pinned** for shared excerpts?
## 13. Working recommendation (principles frozen; plane open)
| Layer | Recommendation |
|-------|----------------|
| Product UI | coulomb cards + spaces |
| Markdown structure | markitect-tool |
| Product UI | coulomb cards + spaces + dual authoring |
| Markdown structure | markitect-tool (includes + future derived slots) |
| Identity / deploy | NetKingdom / Railiance |
| Content plane shortlist | Spike **B thin** and **C** in parallel; keep **A** as fallback |
| SoR narrative | User-facing SoR = markdown export; engine may add durable IDs/lineage **without** trapping content in opaque DB-only forms |
| Assets | Prefer kontextual BlobStorage now; extract **asset-store** when second consumer appears |
| Content plane | Spike A+B+C; choose with export fidelity as hard constraint |
| SoR | Markdown export always works; engines add IDs/lineage without trapping content |
| Assets | kontextual blobs now; extract later if shared |
| Chunks | Transclusion/extract, not a priori graph |
| Derived/stale | Badge → alternative → replace\|keep + lock |
| markitect-main | Avoid |
---
## References
### coulomb-social
- `docs/architecture/2026-08-12-content-substrate-exploration.md`
- `docs/architecture/transclusion-and-chunks.md` — page-first model; chunks as extract/transclude
- `docs/architecture/authoring-modes-and-lazy-projections.md` — top-down/bottom-up; lazy Title/Abstractor/Visual + lock
- `docs/architecture/stale-derived-content.md` — stale badge; alternative; replace|keep; includes/pipelines
- `docs/architecture/transclusion-and-chunks.md`
- `docs/architecture/authoring-modes-and-lazy-projections.md`
- `docs/architecture/stale-derived-content.md`
- `docs/architecture/spike-checklist.md`
- `docs/capability/stage-1-cutover.md`
- `docs/capability/model-v0.yaml`
- `docs/capability/existing-bubble-capability-map.md`
- `docs/decisions/2026-08-12-feature-cut-stage-1.md`
- kontextual: `docs/architecture-blueprint.md`, `asset-registry-implementation.md`, `blob-storage-content-streaming-workplan.md`, `markitect-tool-reuse-boundary.md`
- shard-wiki: `INTENT.md`, `SCOPE.md`
- markitect-tool: `docs/markitect-main-scope-assessment.md`, `docs/transform-compose-include.md`, `docs/content-references.md`
- markitect-main succession: `docs/successor-gap-assessment.md`
### fleet
- kontextual: architecture-blueprint, asset-registry, blob-storage, markitect reuse boundary
- shard-wiki: INTENT, SCOPE
- markitect-tool: transform-compose-include, content-references, cache-incremental, **proposed-stale-derived-and-include-pinning.md**
- markitect-main: successor-gap-assessment (legacy)