Cost & Value Tracker
Living ledger, not a one-time estimate — Service-Based-Cost-Analysis.md priced out a different (Kubernetes/AWS) architecture and is now stale against the Coolify/Hetzner/Postgres-only stack actually decided in Stack-Finalization-Draft.md; this doc replaces it as the thing to actually update monthly. Two things this tracks that a plain invoice list doesn't: what did this spend actually buy, and is it still worth what it costs at current scale — a service that made sense at 50 users can be waste at 5,000, and vice versa.
How to use this
Update monthly (or whenever a new service is added/dropped). One row per paid service. "Value delivered" should be a real answer, not a restatement of what the tool does — e.g. not "Grafana: monitoring" but "Grafana: caught the Redis broker filling disk before it caused an outage, week of Sept 3."
Budget ceiling per phase
| Phase | Monthly ceiling | Note |
|---|---|---|
| Phase 1 (10s–100s users) | (set a number here — nothing is free except the Free/solo-testing window itself; even the single 8 vCPU/16GB Hetzner CCX box plus Supabase/Sentry/Resend free tiers has a real floor cost) | Fill in once actual Hetzner CCX US pricing + Supabase/Sentry free-tier limits are checked — don't leave this blank past first real spend |
| Phase 2 (1,000s–10,000s users) | (TBD — tied to the "monetized enough to afford a hire or managed-HA upgrades" condition in the Phase definitions table) | Should be set as a fraction of actual/projected revenue at that point, not a fixed number decided today |
| Phase 3 (scaled) | (TBD) | — |
Monthly ledger
| Month | Service | Cost | What it bought / value delivered | Still worth it? |
|---|---|---|---|---|
| (fill in first real month) |
Cost-per-value questions to ask each month, not just once
- Did anything in this month's spend sit unused (a service paid for but not actually load-bearing yet)? That's the first thing to cut or defer.
- Did anything this month's spend prevented — an outage avoided, a data-loss scenario closed, a legal/compliance gap closed — actually happen, or is it still speculative? (Insurance-shaped spend, like Redis Sentinel or backups, is legitimately valuable even with a "nothing happened" month — don't cut it just because the month was quiet.)
- Is any managed service still passing the "Managed-service acceptance criteria" test in
Stack-Finalization-Draft.md(cheaper than self-hosting, full control retained, feeds central observability)? Re-check periodically — a service can stop passing as usage grows (e.g. Upstash pricing at higher Redis throughput).
Changelog
- 2026-08-28: Created — replaces reliance on the now-architecturally-stale
Service-Based-Cost-Analysis.mdfor ongoing spend tracking. No real spend data yet (Phase 1 not yet launched) — first real entries land once the Hetzner VPS and managed services (Supabase, Sentry, Resend) are actually provisioned and billing. - 2026-08-30: Recreated after the original file was found missing from disk — an earlier write never persisted (confirmed via git history and file mtimes showing no trace of the original creation). Content restored verbatim from conversation history; committing immediately this time to avoid a repeat.