Data quality
Bad data stops at the door, with the evidence.
Stale prices, spikes, zeros and missing curve points arrive every week. Masseef checks values as they land, holds what fails, and puts each problem in a queue for your team with the evidence attached. A curve is never built from 9 of its 14 tenors without someone knowing.
The queue
What your operations team works from.
Every exception arrives with what was checked, what failed and what the sources said. Where several sources supply a value, a source rule picks one, and the published value records which and why: the question auditors and PMs both ask.
- StaleCorp bond 4.25% 2031price unchanged for 3 business daysEscalated
- CoverageEUR 6M curve9 of 14 tenors present, build heldHeld
- SpikeEquity close+38% against its 20-day bandEscalated
- TriangleEURUSD · GBPUSD · EURGBPtriangle out by 41 bpEscalated
- ZeroOption vol point0.0 received from feedEscalated
Illustrative
Why is this the value?
USD 10Y swap · 14 Jun close3.888%
- SourcesVendor A 3.888 · Vendor B 3.887
- Checksstale ✓ · spike ✓ · range ✓
- Chosen byrule v12 · A before B, 0.5 bp tolerance
- Rule approvedmaker JC · checker RP · 2 May
Works on a single feed. Most of what goes wrong arrives on the one vendor you already have. Checks for staleness, spikes, zeros, coverage and cross-rates need no second source.
Adoption
Adopt it without swapping anything.
Each step is taken only when the previous one has earned it.
Keep your process
Your own consolidation keeps writing directly, exactly as today. That is fully supported, not a halfway state.
Monitor mode
The checks run alongside on the feeds you have and fill the queue. Nothing downstream changes.
Firewall mode
When the evidence has convinced you, flagged data stops publishing and everything downstream reads checked values.
Rules where it hurts
After a wrong ex-date costs money, or an auditor asks you to prove a price, add source rules for that data. Only there.
You can leave
Leaving is never a data recovery project.
Raw data kept as received
Vendor values and every restatement are stored untouched. Published values are derived from them and can be dropped.
Everything exports with SQL
No proprietary format to decode. The evidence behind each value leaves with you.
No vendor rule language
Checks and source rules are written in a mainstream programming language that any developer can read and test.
Opt in per data type
Trying checks or source rules on one population never puts the rest at risk.
See what entered your system this week.
Point monitor mode at one feed you already have. We'll show you the queue it builds, with the evidence, on a running system.