Publish CSOC-WP-0005 resource demand and cost evidence

Add low/base/high demand forecasts, service objectives with timestamped
observations, and workload operations labor for resource:tenant:coulomb:coulomb-social.

docs/resource-evidence/demand-forecast-2026-08.md

@@ -0,0 +1,95 @@
+# Workload demand forecast — coulomb-social (2026-08)
+
+| Field | Value |
+|-------|--------|
+| Resource | `resource:tenant:coulomb:coulomb-social` |
+| Workplan | CSOC-WP-0005-T01 |
+| Forecast version | `csoc-demand-2026-08-v1` |
+| Created | 2026-08-12T09:03:47Z |
+| Machine record | `demand-forecast-2026-08.json` |
+
+## Scope
+
+Demand for **this production instance** (Coulomb reference tenant on
+Railiance). Deployment shape is **instance-per-client** (DR-1 C): future
+external customers are **separate instances** and separate resource ids —
+they are **not** folded into the high scenario.
+
+| Case | What it models |
+|------|----------------|
+| **low** | Current parallel-host posture: few operators, Bubble still full product on apex |
+| **base** | Coulomb community day-to-day on the rebuild (still one tenant) |
+| **high** | Full Bubble-parity load on this Coulomb instance (HA-ish replicas/resources) |
+
+## Proxies (stable units)
+
+| Proxy | Unit | Notes |
+|-------|------|--------|
+| active_tenants | count | Expected 1 for this instance |
+| active_members | MAU-ish Member count | App-owned |
+| http_requests_per_month | count | App Service ingress |
+| oidc_logins_per_month | count | Successful app logins |
+| background_jobs_per_month | count | 0 until workers exist |
+| app_database_gb | GB logical | `coulomb_social_db` on apps-pg |
+| stored_media_gb | GB | App-owned binaries only |
+| forgejo_content_gb_bound | GB | Attribution for space markdown on shared Forgejo |
+| ingress_gb / egress_gb | GB/month | App path only |
+| cpu/memory request·limit | mCPU / Mi | Declared K8s requirements |
+| operator_hours_per_month | hours | Workload labor (T03) |
+
+## Scenario table (monthly)
+
+| Proxy | low | base | high |
+|-------|----:|-----:|-----:|
+| active_tenants | 1 | 1 | 1 |
+| active_members | 5 | 50 | 500 |
+| http_requests | 5 000 | 150 000 | 2 000 000 |
+| oidc_logins | 80 | 1 500 | 20 000 |
+| background_jobs | 0 | 0 | 5 000 |
+| app_database_gb | 0.1 | 1 | 10 |
+| DB growth %/mo | 5 | 10 | 15 |
+| stored_media_gb | 0 | 2 | 50 |
+| forgejo_content_gb_bound | 0.01 | 0.5 | 20 |
+| ingress_gb | 0.5 | 5 | 40 |
+| egress_gb | 1 | 15 | 120 |
+| cpu request (m) | 100 | 200 | 500 |
+| mem request (Mi) | 256 | 512 | 1024 |
+| replicas | 1 | 1 | 2 |
+| operator h/mo | 2 | 4 | 8 |
+| labor €/mo @ €60/h | 120 | 240 | 480 |
+
+Infrastructure EUR is **not invented** here. Cost joins should treat
+infrastructure as an allocated share of cluster, apps-pg, ingress, and
+identity platform once those owners and fin-hub supply figures.
+
+## Capacity floor (current deploy declaration)
+
+From `railiance-apps` chart defaults (production values do not override
+resources as of 2026-08):
+
+| Dimension | Value |
+|-----------|--------|
+| replicas | 1 |
+| cpu request / limit | 100m / 1000m |
+| memory request / limit | 256Mi / 1Gi |
+| media PV | disabled |
+| Image tag (values) | `7fcd0cf` (railiance-apps helm values; may lag git HEAD) |
+
+## How resource-control should use this
+
+1. Pick scenario band from observed MAU / requests (or default **low** until
+   Bubble migration starts).
+2. Feed proxies into monthly forecast/actual control (null infrastructure
+   until platform prices exist).
+3. Attribute Forgejo-bound content via `forgejo_content_gb_bound` without
+   double-counting storage in the app resource.
+4. Recalibrate after three comparable months (MAPE on members, requests, DB
+   GB, labor hours).
+
+## Assumptions and falsifiability
+
+See JSON `assumptions` array. Primary falsifiers after each calendar month:
+
+- measured `active_members` outside the active scenario band by >2×
+- `app_database_gb` growth vs declared growth %
+- operator hours vs T03 labor log