# Business Platform Decision Records **Created:** 2026-07-10 **Workplan:** `CUST-WP-0058` **Status:** All three decisions **pending** — Bernd's to resolve; this document frames options and recommendations only. ## Context Binky Hedgehog GmbH intends to build commercial business applications on the coulomb/HelixForge platform (Forgejo → railiance-forge → railiance-apps → Core Hub ecosystem): 1. **vergabe-teilnahme** — tender-participation web app as the first commercial offering for business clients. 2. **coulomb.social rebuild** — replace the current bubble.io implementation as the second business project. The hub-ecosystem review (2026-07-09, see `hub-ecosystem-architecture.md`) concluded the architecture is sound but three decisions gate the business lane. Each is recorded here with options and a recommendation; resolution happens via State Hub decision records (`resolve_decision`). --- ## DR-1 — Tenancy model for vergabe-teilnahme **Status:** pending ### Question How are business clients isolated: one deployed instance per client, or one multi-tenant instance? ### Options | Option | Description | Consequences | | --- | --- | --- | | **A. Instance-per-client** | Each client gets a dedicated deployment (own DB, own release values in `railiance-apps`) | Strong isolation; simpler compliance story; per-client backup/restore; matches vergabe-teilnahme SCOPE ("no multi-tenancy"); ops-hub must model *client instances*; fin-hub needs per-instance cost attribution; N instances = N upgrade operations | | **B. Multi-tenant single instance** | One deployment, tenant column / row-level isolation | Cheaper to operate at scale; contradicts current SCOPE; requires tenancy rework in the app; single blast radius for incidents and data-protection failures | | **C. Hybrid (A now, B later)** | Start instance-per-client; revisit at a client-count threshold | Defers rework until demand proves it; risk of never consolidating | ### Recommendation **Option A (instance-per-client) for v1**, with an explicit revisit trigger (e.g. >5 paying clients or upgrade toil exceeding one operator-day/month). It matches the app's current SCOPE, the railiance-apps per-app values pattern, and gives the cleanest early compliance/data-protection story for German business clients. ### Downstream effects once resolved - ops-hub: add client-instance entity (environment × client mapping) - fin-hub: per-instance cost attribution keys - Business-app service contract (CUST-WP-0058-T04): codify instance provisioning/backup/upgrade obligations --- ## DR-2 — coulomb.social rebuild architecture **Status:** pending ### Question Is the coulomb.social rebuild a **Core Hub extension** (hubs/widgets/ manifests, ops-hub pattern) or a **standalone application** (vergabe- teilnahme pattern)? ### Options | Option | Description | Consequences | | --- | --- | --- | | **A. Core Hub extension** | Built on `/api/v2` framework: hubs, widgets, manifests, interaction events | Reuses production framework already at `hub.coulomb.social`; interaction-platform semantics fit; couples product evolution to core-hub framework maturity; framework gaps become product blockers | | **B. Standalone app** | Own service repo, own DB, deployed via forge→apps lane like vergabe-teilnahme | Full product freedom; no framework coupling; duplicates interaction/identity plumbing core-hub already has; two social-ish surfaces to keep coherent | | **C. Defer until after the vergabe-teilnahme pilot** | Run the delivery-lane pilot first; decide with evidence | Costs calendar time; decision made with real platform evidence instead of speculation | ### Recommendation **Option C, with a leaning toward A.** The delivery-lane pilot (CUST-WP-0058-T07) will expose whether the forge→apps path and Core Hub framework are mature enough to host a client-facing interaction product. Deciding now would be speculation; the bubble.io exit additionally needs a separately-scoped **data migration** assessment regardless of which shape wins. ### Downstream effects once resolved - A → core-hub gains its first external-product extension; widget/manifest contract hardening becomes product-critical - B → second entry in the business-app service contract; shared identity becomes the only coupling point - Either → bubble.io data-export/migration workplan in the owning repo --- ## DR-3 — External identity path for business clients **Status:** pending ### Question What identity stack serves *external* users (client employees, later coulomb.social members): the current lightweight internal stack (KeyCape/Authelia/LLDAP) or the proposed expanded-mode Keycloak lane (`NK-WP-0011`)? ### Options | Option | Description | Consequences | | --- | --- | --- | | **A. Lightweight stack, app-local accounts for v1** | vergabe-teilnahme manages its own user accounts per instance; platform identity stays internal | Fastest to ship; consistent with instance-per-client isolation; no SSO story; identity debt accrues per app | | **B. Activate NK-WP-0011 (Keycloak expanded mode) now** | Platform-level OIDC for external users before first client onboarding | One identity story from day one; enterprise federation ready; heavyweight — delays first client; NK-WP-0011 is only *proposed* | | **C. Trigger-based: A for vergabe-teilnahme v1, B gated on defined triggers** | Define explicit conditions that activate the Keycloak lane (e.g. coulomb.social greenlight, first client requesting SSO/AD federation, >2 apps with local accounts) | Ships v1 fast without pretending identity debt doesn't exist; triggers make the debt visible and bounded | ### Recommendation **Option C.** Vergabe-teilnahme v1 explicitly excludes external partner accounts and is instance-per-client, so app-local auth is defensible *for that app*. coulomb.social is the forcing function: any resolution of DR-2 that greenlights the rebuild should automatically activate the NK-WP-0011 lane, because a social platform on app-local accounts is not viable. ### Downstream effects once resolved - NetKingdom: NK-WP-0011 gets concrete activation triggers instead of indefinite "proposed" status - Business-app service contract: identity section states which mode each app class must use --- ## Resolution protocol 1. Each DR is registered as a pending State Hub decision (custodian topic). 2. Bernd resolves via `resolve_decision()` or by stating the choice in a session; the resolving agent updates this document's status lines and the hub record together. 3. `CUST-WP-0058` tasks T01–T03 track the facilitation; downstream tasks stay `wait` until their gating DR is resolved.