Why this exists
`docker-compose.dogfood.yml:50` mounts `./backend:/app` into the backend container. Live code reload during dogfood iteration is useful, but it means the dogfood backend is running source files directly off the host, not the same baked image as prod. Findings discovered in dogfood may reflect environmental drift (Python path resolution, read-time race, etc.) rather than real prod-reachable bugs. Lowers the signal-to-noise ratio of the dogfood loop, which strategy locked belief #4 names as the primary validation signal.
Source: static audit 2026-05-13. Logged as dogfood finding F-016.
What ships
- `docker-compose.dogfood-fidelity.yml` override file: removes `./backend:/app` volume mount, forces image rebuild
- `make dogfood-up-fidelity` and `make dogfood-down-fidelity` Makefile targets
- Section in `docs/dogfood.md` titled "Iteration mode vs fidelity mode" — explains the trade-off and when to use each
- Section in `oci/dogfood-runbook.md` (DevVault, vault-only): triage protocol — before triaging a dogfood finding to a GH issue, re-verify under fidelity mode
Acceptance criteria
Wave
pre-thesis
Type
chore
Stack scope
Priority
low
ADR required
no - infra tweak, no architectural change
Dogfooding signal
yes - audit finding F-016 from 2026-05-13. Reviewing the dogfood compose file for isolation guarantees surfaced this. Self-referential — improves the dogfood loop itself.
Related
- Wave: pre-thesis
- Prior ADRs: `oci/decisions/2026-05-13-dogfood-pipeline-v0.1.md`
- Linked: dogfood-findings F-016
- External: n/a
Filed from OCI audit 2026-05-13. Full context: `oci/dogfood-findings.md` F-016.
Why this exists
`docker-compose.dogfood.yml:50` mounts `./backend:/app` into the backend container. Live code reload during dogfood iteration is useful, but it means the dogfood backend is running source files directly off the host, not the same baked image as prod. Findings discovered in dogfood may reflect environmental drift (Python path resolution, read-time race, etc.) rather than real prod-reachable bugs. Lowers the signal-to-noise ratio of the dogfood loop, which strategy locked belief #4 names as the primary validation signal.
Source: static audit 2026-05-13. Logged as dogfood finding F-016.
What ships
Acceptance criteria
Wave
pre-thesis
Type
chore
Stack scope
Priority
low
ADR required
no - infra tweak, no architectural change
Dogfooding signal
yes - audit finding F-016 from 2026-05-13. Reviewing the dogfood compose file for isolation guarantees surfaced this. Self-referential — improves the dogfood loop itself.
Related
Filed from OCI audit 2026-05-13. Full context: `oci/dogfood-findings.md` F-016.