From source to usable evidence.
Each stage answers a different question: may this data enter, what does it mean, what may analysis use and who may see the result? A completed import is only one part of that chain.
Admit
A read-only connection admits a defined set of observations from its source. The company, source and field scope stay bound together. CSV imports enter through named templates with their own membership and validation checks.
Company · Source · Approved fields
Normalize
Map source fields to defined measures, units and reporting periods. Preserve the source position and correction history. Similar field names do not make different measures comparable, and a missing period must not become a zero.
Definitions · Periods · Comparable cells
Minimize
The generated-Case evidence path aggregates and applies source-scoped privacy gates. It carries source coverage and gaps instead of substituting raw rows for missing context.
Aggregates · Privacy gates · Coverage
Project
The result is read through its permitted audience. Import success does not automatically grant readers access to the source or permission to write to the connected system.
Current audience · Permitted record
A completed import does not guarantee a new Case. Evidence admission, privacy thresholds and Case generation have their own checks.
Define the interface before connecting.
| Integration concern | Technical boundary | Evaluation question |
|---|---|---|
| Schema & units | Metric definitions, value types and supported dimensions. | Do the source and comparison describe the same measure? |
| Time & coverage | Period boundaries, available observations and data gaps. | Is the evidence current and comparable across periods? |
| Failure & retry | Connected reads use durable checkpoints; CSV imports use resumable batches. | What was accepted, what remains incomplete and which result is still current? |
| Read & write | Source-read scope and governed action scope are separate. | Which fields may be read, and which effects are explicitly approved? |
A connection is not
a blanket data grant.
Review the source, purpose, data class, permitted audience and recovery path together. Technical provider use does not establish a processing region, retention period or approved contract.
Sources, refresh and recovery.
Two sources. Distinct questions.
Zendesk Support supplies ticket status, timing and mapped customer-satisfaction observations. The launch mapping measures full resolution in calendar minutes. It does not substitute that measure for first reply or claim first-contact resolution. Ticket text, comments and attachments stay outside the retained analysis data.
HiBob supplies workforce observations for headcount, tenure and turnover, with broad department and employment cohorts. Absence and time-off data are outside the launch mapping. Support and workforce measures retain their own definitions before any comparison is considered.
Know where a refresh stopped.
Scheduled reads follow a recorded position in the source. A checkpoint advances only after the data has been durably accepted. Corrections and deletions must be reflected in the admitted records, while duplicate reads must not create new evidence simply by arriving twice.
A completed read is separate from evidence admission and Case generation. Stale data, an incomplete period or a withdrawn permission can stop the chain later. The status needs to identify that boundary, so a fresh connection indicator cannot hide an outdated analytical result.
CSV is another way in.
For a file-based source, an editor, administrator or owner selects Support/CSAT, HR/People or declared capacity and checks the field mapping. Define the unit and reporting period: a column called “duration” does not tell you whether it means first reply or full resolution.
Files are sent in sequential batches. If a batch is interrupted, rerun the same import: stable fingerprints recognize completed batches and continue the unfinished work. The mapping preview inspects a sample; the import receipt establishes what was actually accepted.
Source access has a purpose.
The connection is tied to a company, a source and an approved field scope. Reading a source does not authorize writing back to it, and a credential does not grant every downstream reader access. Current membership, source permission and privacy state still govern the result.
A source can contain instructions as well as data. Connected content is treated as untrusted input: it cannot grant a model new tools, widen permissions or change the review’s objective. Processing providers and their roles are described in the privacy notice.
