The Perishable Record: The Deer Data That Disappears

There is a quiet assumption underneath most data work: that a source you can read today you can read again tomorrow. For deer data, that assumption is often false. The most granular records (a row for every registered deer, a per-unit table, a county's back-history) are frequently the least durable. Minnesota's registration feed is rewritten each season, overwriting last year's per-deer detail. North Dakota's prior-year files quietly return errors. Wisconsin, as a companion paper documents, simply removed eleven counties from its published record. None of this is malicious; it is how agencies manage storage and schemas. But the consequence is stark: much of the richest deer data is perishable, and the only defense is to capture it the moment you find it.

A durable record grows: each season is appended, the old seasons stay put, and the series lengthens. A perishable record is different. It holds a window, and as a new period enters, an old one leaves. The distinction is invisible until you need the part that left. Deer data lives on both sides of this line. Long statewide harvest series are usually durable; you can pull decades at once. But the granular layers (per-deer registration rows, per-unit tables, fine geographic detail) are frequently published as a live view of the current season, backed by a system that has no obligation to preserve what it showed last year. Read it late and you do not get an error that says “deleted.” You get a shorter, thinner record and no indication that it was ever fuller.

Minnesota is the cleanest example. The state's registration feed is granular to the individual animal (conceptually one row per registered deer, on the order of 186,000 rows in the most recent season) and it reflects the current season. It is not versioned into an annual archive of per-deer files; the live feed is the record, and the live feed moves on. Capture it after the season closes and you hold that year's per-deer detail; wait, and the next season overwrites the view. FDR's reproduced aggregate for the latest season (the registered total summed across all deer-permit areas) is 186,333 deer, which is exactly why the granular feed matters: the state-level number survives in summary tables, but the per-deer and per-area texture behind it lives only in the snapshot. And the aggregate is not even complete: pull the summed series and the 2014 season is missing entirely, a gap between 2013 and 2015 that no amount of later effort can fill.

The state-level number survives. The row-per-deer detail behind it lives only in this season's snapshot. Pull it now, or it is gone for good.

Minnesota's snapshot is one shape of the problem; there are others. In North Dakota, prior-year data releases return access errors once superseded: the state effectively publishes a rolling window in which last season's file can become unreachable the moment the new one posts. There is no deletion notice; the URL simply stops answering. Wisconsin shows the third shape. In a 2025 rebuild it dropped eleven Northern-Forest counties from its published county-harvest series, contracting the public record from 72 counties to 61: the same perishability expressed spatially, as a map that got smaller rather than a window that rolled forward. (FDR recovered that Wisconsin series from archived tables; a companion paper documents it.) Three states, three mechanisms (overwrite, expire, delete) and one lesson.

If the richest data can vanish without warning, the analytical posture has to change. You cannot treat a granular public feed as a library you will visit later; you have to treat it as a harvest with a season of its own. FDR's practice follows directly: pull perishable feeds on a fixed annual cadence tied to each state's season close, store every vintage with its retrieval date, and never assume a source read once can be re-read. Minnesota's per-deer feed is re-pulled every year after the season ends, precisely because next year it will describe next year and nothing else. North Dakota's window is captured before it rolls. Wisconsin's deleted counties were reconstructed from an archive that itself might not survive the next rebuild. The through-line is not pessimism about agencies (they are doing ordinary systems work); it is realism about what “public” guarantees. It guarantees the data is available now. It promises nothing about later. The record is perishable; the discipline is to capture it while it is fresh.