tickvault is a market data warehouse that proves its data on four dimensions: idempotent reprocessing, corrections, lineage, and observability. Built on a medallion architecture, every downstream dataset is derived, never fabricated.
Each of these is backed by an automated test or a dbt check that runs on every build, not just claims in a README.
Reprocessing an already-landed partition never changes its logical result, even though every reprocess writes a physically new, immutable file.
When source data for an already-landed partition genuinely changes, the new value wins everywhere downstream, and the old value is never deleted, just superseded.
Every row carries its run ID, source system, fetch time, and a content checksum. Full version history is queryable, not reconstructed by hand from raw files.
Missing or suspiciously incomplete data is flagged automatically, as a dbt monitoring model and as a per-partition Dagster asset check, not discovered downstream by a confused analyst.
Medallion architecture. Only bronze is written by ingestion. Everything downstream is derived and safe to rebuild.
Raw facts exactly as ingested. Never updated or deleted.
Cleaned, deduped, versioned views for analysis.
Curated, stable dataset for downstream consumption.
Gaps and anomalies surfaced as first-class data, not a side channel.
A read-only API and dashboard for exploration and integration.
Run real DuckDB queries against exported data, read-only, entirely in your browser. No server involved.