| source_release | period_start | period_end | value | |
|---|---|---|---|---|
| 0 | 2024 | 2022 | 2022 | 6.5 |
| 1 | 2025 | 2023 | 2023 | 6.6 |
Three time axes
Every measure carries two notions of “when” in its columns, and the release you attached is a third. Conflating them is the characteristic error in this domain.
| Axis | Column(s) | Meaning |
|---|---|---|
| Period | period_start, period_end |
the time the estimate describes |
| Source release | source_release |
the upstream edition that published it |
| Cancer Basemap release | the folder you attach (v2026-11-02/, or latest/) |
which monthly bundle you are reading |
A real example: PLACES 2024 vs. 2025
Two rows, same measure, same county (Denver County, CO). The source release is '2024' or '2025' — the PLACES edition. The period each release actually describes is one year earlier: the 2024 release models 2022 BRFSS data, the 2025 release models 2023 data. A query that filters on source_release = '2025' expecting “2025 data” is filtering on the release label, not the period — the value it gets describes 2023.
Both source releases are in every Cancer Basemap release, because source_release is a column, not a folder. The Cancer Basemap release is the third axis: a bundle holds the current rows only, so if a source corrects a value within one source_release, the corrected row replaces the old one in the next bundle. The earlier value stays in the earlier release, which is why a dated folder, or its Hugging Face tag, is how you reproduce an analysis (see Versions and citation).
A second example: an edition’s own period moves
SVI’s source_release is the edition, and each edition’s own period is a five-year ACS window that moves with it — never the edition year alone:
| edition | period_start | period_end | |
|---|---|---|---|
| 0 | 2000 | 2000 | 2000 |
| 1 | 2010 | 2006 | 2010 |
| 2 | 2014 | 2010 | 2014 |
| 3 | 2016 | 2012 | 2016 |
| 4 | 2018 | 2014 | 2018 |
| 5 | 2020 | 2016 | 2020 |
| 6 | 2022 | 2018 | 2022 |
The 2020 edition’s period is 2016–2020, not '2020' alone — treating an edition’s release year as its period would be off by up to four years.
Why this matters for a catchment report
The sentence a cancer center has to defend — “in our catchment, as reported by source S in release R, the measure was X” — names the release, not the period, on purpose: the release is what you can point back to when someone asks “where did this number come from.” Reproducing a past analysis means filtering on source_release, not on period_start/period_end (a later release can restate an earlier period with revised numbers), and pinning the Cancer Basemap release you attached, since a bundle’s rows can change between releases.