data-platform

Router for the data-engineering lifecycle. Resolve topic, then Read EXACTLY ONE playbook from routes:. Never load the cluster. [DOC]

Use for pipelines, ETL, quality/validation gates, migration, export, schema evolution, data-flow wiring on Firestore/Cloud Functions. Not for general backend logic, auth, UI state, infra provisioning. [INFERENCE] In: topic (required, 8-enum), depth, the user's concrete data task. Out: the routed playbook applied to that task — dispatch only, never a generic answer from this file. [DOC]

Routing

1. Infer topic; ask only if two topics are equally plausible. [INFERENCE]
2. Disambiguate by intent: transform in motion → etl-patterns; write-time rules → data-validation; profiling/health → data-quality; bulk load → data-migration; recurring read-out → data-export; field shapes over time → schema-evolution; trigger/event wiring → data-flow-architecture; end-to-end pipeline design → data-engineering. [INFERENCE]
3. deep → apply exhaustively, verify each step; quick → essentials only.

Spine: Discover → Analyze → Execute → Validate. Gates: constitution v6.0.0 enforcement, evidence tags, script-first. [CONFIG]

Validation gate (done = all true)

• Exactly one playbook loaded; topic ∈ enum. [DOC]
• Output addresses the user's task, not the topic in the abstract.
• Destructive ops carry a rollback path before execution. [ASSUMPTION]
• Claims tagged with one Alfa-core family tag (references/verification-tags.md).
• Gate scored via assets/quality-rubric.json + assets/checklist.md (≥0.85). [DOC]

Anti-patterns

• Loading >1 route, or reading a playbook before fixing topic. [INFERENCE]
• Guessing topic when ambiguous instead of asking.
• Improvising bulk writes here (Firestore caps 500/batch) — defer to migration. [CONFIG]

Self-correction

Two topics → run the dominant, name the deferred. Route yields nothing useful → re-route once via step 2, don't force-fit. [INFERENCE]