The history you had before the system did

Implementations convert what the new system needs to operate: who works here now, at what pay, in which organization, with service dates as reference fields. Loading twenty years of transactions is expensive and risky, so almost nobody does. Prior careers land as a hire date, free-text notes, maybe a document scan.

The consequence surfaces the first time someone asks for a trend longer than the tenant’s age. Turnover over ten years, growth by division across the merger, pay progression for the veterans: the delivered stores hold three years of month ends on a good day, and nothing before go-live at all. The question is not slow. It is structurally unanswerable inside the tenant, and no future release changes what was never loaded.

The raw material still exists: final extracts from the retired system, payroll registers, the conversion workbooks themselves. Reconstruction is warehouse work in three moves. Land the legacy data as delivered. Conform it to the same dimensions the live model uses, so a department in 2015 rolls up like a department today. Then generate the same monthly snapshot series the owned snapshot produces going forward, back through the legacy years.

Rebuilt history is a model of the past, not an observation of it. Definitions differed, populations shifted, and some legacy fields simply have no counterpart. The honest reconstruction carries that on its face: a lineage column on every historical row, seam markers on every chart that crosses go-live, and a written register of what could not be rebuilt and why.

-- one trend, the seam explicit
SELECT snapshot_month,
       headcount,
       history_source        -- 'legacy_rebuilt' or 'workday_observed'
FROM fct_headcount_monthly
ORDER BY snapshot_month

Done this way, the twenty-year chart exists and defends itself. The left side says rebuilt, the right side says observed, the seam says where, and the register says what the rebuild could not know. That is more honesty than the tenant offers about its own window, and it is the difference between a long trend and a long guess. This page ships at the designed maturity stage; the worked example arrives with the synthetic harness.