Taxonomy
Every event defined before it is trusted.
Quantiv keeps a registry of every event your apps send: its properties, their types, whether they are required, and how sensitive they are. Ingestion enforces it, so the data you analyse is the data you meant to collect.
Event definition
v4 · Active
purchase
Norvelle Web
| Property | Type | Required | Class |
|---|---|---|---|
| order_id | string | yes | none |
| order_value | number | yes | none |
| currency | string | yes | none |
| string | no | PII | |
| condition_code | string | no | PHI |
Events that do not match an Active definition are rejected in prod and flagged in test and stage.
Illustrative data
Lifecycle
Draft, Active, Deprecated.
Definitions move through a simple lifecycle so changes are deliberate and visible.
- 01
Draft
A definition is written and reviewed. Events that match it are not yet accepted in prod.
- 02
Active
Ingestion validates every incoming event against the Active version of its definition.
- 03
Deprecated
Retired events stop being accepted, and their history stays intact for analysis.
Sensitivity
Every property is classified as none, PII, or PHI.
Classification is part of the definition, not a separate spreadsheet. It travels with the data and decides what is allowed to leave.
The egress filter reads it on every outbound payload. PII is hashed and forwarded only where allowlisted. PHI is never forwarded.
- Classification set per property, in the definition
- Enforced on every destination, not reviewed after the fact
- Changes to classification are recorded in the audit log
Egress filter · purchase
→ Meta CAPI
- order_idforwarded
- order_valueforwarded
- emailPIIhashed
- gclidforwarded
- condition_codePHIblocked
A property classified PHI is not forwardable. Not by policy, not by a reviewer remembering - the filter has no path that lets it leave.
Illustrative data
Why it matters
A taxonomy is the difference between data and noise.
Errors surface early
A misspelled event or a wrong type is rejected at the door, while the developer is still looking at it.
Analysis stays comparable
Versioned definitions mean last quarter's purchase and this quarter's purchase are the same thing.
Agents work from a clean base
Proposals from Quantiv's agents rest on data that has been validated, typed, and classified.
Coverage
Gaps are found for you, and approved by you.
The instrumentation agent watches incoming traffic and proposes definitions for events that nothing yet describes, with inferred types and a suggested classification.
Proposals arrive as drafts. A person confirms the classification and activates them. Nothing reaches production unreviewed.
Proposal · instrumentation agent
needs classification
subscription_paused
Norvelle Web · prod · 1,284 events / 7d
- plan_codestring
- pause_reasonstringPII?
- resume_datedate
Inferred from 1,284 unmatched events. Classification on pause_reason needs a human decision before this can go Active.
the agent proposes · a human activates · nothing reaches prod unreviewed
Illustrative data
Questions
Common questions.
Who can change a definition?
Owners and administrators. Every change is versioned and recorded with who made it and when.
What happens to events that do not match?
In prod they are rejected and reported. In test and stage they are accepted with a warning, so teams can iterate quickly before promoting a definition.
See it on your own numbers.
A two-week pilot runs your conversions through Quantiv alongside what you report today. The difference between the two is the business case.