ETF Security Master
The first reference implementation. Proof, not product — it exists to show the full loop end to end on synthetic and public-source data.
The decision it answers
Which source records represent the same ETF, and which attributes are authoritative? The outcome is a governed golden ETF record for operations, analytics and downstream systems.
Sources
- A licensed vendor reference-data export
- OpenFIGI
- SEC EDGAR
The loop
Vendor export lands in cloud storage and is ingested incrementally through bronze, silver and gold layers. A data-quality anomaly surfaces with its lineage blast radius. Probabilistic matching runs against the identifier sources, and the uncertain middle band is ranked by governed match evidence rather than a black box. A steward decides. The decision commits to the operational store with an append-only audit record and an outbox entry, survivorship publishes the golden dimension, and the governed analytics and API surfaces pick it up.
The temporal pair in practice
as_of
Known — known_as_of
A correction can arrive late. When it does, the two questions separate, and the platform answers both rather than overwriting one with the other.
| Question | Parameter | Answer for 1 March |
|---|---|---|
| What was effective on that business date? | as_of |
The corrected value — today's reconstructed view of 1 March |
| What did the system know at that point in time? | known_as_of |
The pre-correction value — what was actually known before 15 March |
Reproducing both after a late-arriving source correction is the acceptance test for temporal reconstruction, not a description of intent.
How the record is kept honest
Bitemporal versioning
Golden-record history is bitemporal throughout: business validity and system knowledge are tracked as separate intervals rather than collapsed into one effective-dated chain.
Append-only audit
Decision audit records who, what, when and why, with immutability enforced at the database permission level rather than by convention. Corrections are compensating records, never mutation in place.
One authority for operational state
The operational store is authoritative for MDM state; the published gold tables are a downstream publication projection, never a co-equal truth. If publication lags, the decision still stands and reconciliation reports the gap.
What is measured
The scenario's success measures are auto-match rate, false-match rate, unresolved exceptions, stewardship service level, and source-arrival to publication time.
Placeholder — no measured values are published here. For master data management a false positive is worse than a false negative, so precision and recall are reported per match band rather than as a single headline number.