The methods
File exports are universal and often the only option for legacy systems. APIs give structured, repeatable access but rarely expose everything. Direct database connections reach the most but require trust and care. xAPI streams capture granular activity from content. SFTP and cloud storage suit scheduled batch transfers. Each method has a place; most real integrations use several.
What gets lost in translation
A completion date stored as text in one system and a timestamp in another. A score on a 0 to 100 scale beside one on 0 to 10. An enrolment status that means "completed" in one platform and "passed" in another. Normalization makes these comparable, but the original value must survive alongside the clean one or the transformation cannot be verified.
Lineage defined
For each record: the source system, connector and extraction run; the original identifier and field values; every mapping and transformation applied, with its version; identity match decisions that touched the record; and who approved its promotion into the governed layer.
Why it earns its keep
Auditors ask where a number came from. Migration teams must prove nothing changed in transit. BI teams need to know which source a field reflects. Lineage answers all three from the same place. It is not overhead; it is the thing that makes the rest trustworthy.
A note on xAPI
xAPI statements are rich but not self-governing. A learning record store full of statements still needs identity reconciliation, normalization and lineage to be useful for reporting or audits. Treat it as another source, not as the answer.
More from the blog
Institutional memory should not depend on a vendor contract
Organizations remember through their records. When learning records live only inside platforms that get replaced every few years, the organization forgets.
Learn more →Neutral by design: why we chose not to build an LMS
The easiest product to sell is another platform. We built a layer instead, and the reason is structural.
Learn more →Why AI pilots in learning stall, and the inventory that unblocks them
It is rarely the model. Pilots stall because nobody can say what data exists, who owns it, whether it may be used and whether it is accurate.
Learn more →Ready to see your learning data in one place?
Tell us which systems you run and we will show how the platform connects, normalizes and preserves their records.