Connect · item 15
Data quality gates at intake
Bad data does not become less expensive further down the line. Sixteen checks sit at the first station of every line, covering completeness, validity, duplication, referential integrity, timeliness and consistency. A unit that fails is blocked outright or held in a quarantine queue with a named owner — never quietly waved through with a flag nobody reads.
Health
What the rules are catching
Active and monitoring rules, and everything they have held, released or rejected.
What the gates test for
Six dimensions. A check that has never fired is shown as never having fired, not as evidence of a clean estate.
What happens on failure
The action is stated on the gate, not decided later.
Gates that have never fired
Reported as a zero, not as a pass.
The gates
Each names the dimension it tests, the expression it evaluates, what happens when it fails, and who owns the failures.
| Gate | Dimension | Applies to | Expression | On failure | Run | Failed | Pass | Held | Owner |
|---|---|---|---|---|---|---|---|---|---|
| DQ-01 Mandatory header fields present High | Completeness | Every unit at the first station of its line all-lines | requester, value, materiality and jurisdiction are all non-empty at intake | Quarantine threshold 99.5% | 16,420 | 67 | 99.6% | 4 | N. Ferreira |
| DQ-12 Configuration item has a named owner Medium | Completeness | Technology change and incident intake function Configuration Management Database | the configuration item referenced has an owning team recorded in the configuration database | Warn threshold 95% | 12,436 | 65 | 99.5% | 7 | J. Petrakis |
| DQ-15 Personal data fields carry a classification Highmonitoring only | Completeness | Any unit whose payload may contain personal data all-lines | each field flagged as person-linked carries a classification label before the unit enters the line | Quarantine threshold 99% | 14,392 | 43 | 99.7% | 2 | S. Bergqvist |
| DQ-16 Research payload provenance recorded Mediummonitoring only | Completeness | Research and development intake function Research Notebook and Experiment Data | experiment payloads name their notebook entry and protocol version | Warn threshold 95% | 9,928 | 0 | 100% | 0 | P. Iyer |
Quarantine queue
6 items match the current filter. Every held item has an owner and an age.
| Item | Function · line | Gate | Field | Observed | Expected | Source | Age | Status | Owner | Outcome |
|---|---|---|---|---|---|---|---|---|---|---|
| IT1B-1018 Ticket at request DQ-15-Q-002 | Information Technology IT1B | DQ-15 | each field flagged as | empty | each field flagged as person-linked carries a classification label before the unit enters the line | Enterprise Data Warehouse | 7.0 days | Held | S. Bergqvist | — |
| FI6A-1000 Cash action at propose position DQ-15-Q-005 | Finance FI6A | DQ-15 | each field flagged as | empty | each field flagged as person-linked carries a classification label before the unit enters the line | Enterprise Data Warehouse | 5.8 days | Rejected at intake | S. Bergqvist | Rejected at intake. The unit never entered the line and the requester was told why. |
| MK5C-1018 Campaign send at detect defect DQ-15-Q-001 | Marketing MK5C | DQ-15 | each field flagged as | empty | each field flagged as person-linked carries a classification label before the unit enters the line | Enterprise Data Warehouse | 5.7 days | Released | S. Bergqvist | Corrected at source and released back onto the line with the correction recorded. |
| CS9C-1024 Quality review at score independently DQ-15-Q-004 | Customer Service CS9C | DQ-15 | each field flagged as | empty | each field flagged as person-linked carries a classification label before the unit enters the line | Enterprise Data Warehouse | 3.0 days | Held | S. Bergqvist | — |
| SL2A-1012 Account plan at rep review DQ-15-Q-003 | Sales SL2A | DQ-15 | each field flagged as | empty | each field flagged as person-linked carries a classification label before the unit enters the line | Enterprise Data Warehouse | 29 hr | Rejected at intake | S. Bergqvist | Rejected at intake. The unit never entered the line and the requester was told why. |
| GB9D-1006 Service review at challenge DQ-15-Q-006 | AI GBS & Global Business Services GB9D | DQ-15 | each field flagged as | empty | each field flagged as person-linked carries a classification label before the unit enters the line | Enterprise Data Warehouse | 11 hr | Released | S. Bergqvist | Corrected at source and released back onto the line with the correction recorded. |
Which lines carry the intake burden
Quarantine volume by function, all time.
Actions
What is waiting on a person
Quarantine is a holding pen, not a decision. Every item below is work that stopped and is waiting for somebody to say what happens to it.
Judge a quarantined item
Held by a rule. Nobody has looked at it yet.
Fix the source behind a rejection
Rejected items keep arriving until the upstream cause changes.
Finish a review already open
Picked up, not concluded.
Promote a monitoring rule to active
The rule watches and reports but does not hold anything.
Operations
What this desk is allowed to start
A surface that only reports is not operable. This is the work this page can set in motion, and the bound it runs into.
Trigger and bound
This desk can start a rule run across a named batch, and it can move an item into quarantine. It cannot release an item, reject one, or change a threshold — releasing bad data is the whole risk this layer exists to manage, so it stays with a person.
Live observability
What the record shows right now
What happened to everything the rules have held.
Current distribution
171 items
Is policy and strategy coming to fruition
Whether the written intent is holding here
14 of 16 rules are enforcing. 95 items are held and undecided.
Not holding on the record
The written intent is that no unchecked record reaches a downstream figure. 14 rules enforce that today and 2 only watch. The queue is the real test: 67 items are held with nobody assigned and 28 are open with somebody assigned, against 47 released and 29 rejected historically. A held item is not a failure of the rule — it is the rule working — but a held item nobody judges is a figure quietly missing its input.
The gates, the expressions, the thresholds and the quarantine queue are real mechanisms. They run at the first station of every line and they hold work that fails. What they run against is the modeled feed described on the connector registry — connect a customer instance and the same gates run unchanged against real intake.
2 of the 16 gates are set to monitor rather than enforce. They are counted in the check totals but they stop nothing. That is a deliberate state for a gate whose false-positive rate has not yet been characterized, and it is shown rather than hidden behind an overall pass rate.
62 items have been held for more than seventy-two hours, the oldest for 9.1 days. A quarantine queue that ages without clearing is a queue that people have learned to route around, so the age is shown on every row and never averaged away.