The Archive mirrors ~1,000 of 45,919 stations, and the dataset has no DOI

## Why this matters The pitch is that the Archive is the citable artifact: "the open, continuously updated, citable record of the world's public water gauges". The catalog delivers that, 45,919 stations across six agencies including ones nobody else publishes. The observation mirror is a smaller thing, and the gap is worth naming out loud before the data paper is written against it. From `obs/manifest.json` (harvest of 2026-08-18): | bundle | stations | rows | first | | --- | --- | --- | --- | | uk_ea / discharge | 373 | 4,656,602 | 1986-08-17 | | usgs / discharge | 322 | 2,219,145 | 1986-08-17 | | hubeau / discharge | 306 | 2,285,708 | 1986-08-17 | Plus water level for `uk_ea` and `usgs`, precipitation for `uk_ea` and `taiwan_cwa`, and groundwater level for `uk_ea`. That is **about 1,000 stations with mirrored daily discharge out of 45,919 catalogued**, and every bundle starts at the same 1986 boundary as the app (see the record-window issue). Two consequences. **The mirrored coverage sits inside Caravan's footprint.** US, UK and France only. Taiwan appears once, as 241 KB of precipitation, with no discharge and no water level. Asian coverage is the wedge Caravan does not have, and right now it is the one thing the mirror does not carry. On the axis Caravan is judged by, basins with mirrored records, Caravan has 6,830 and this has about 1,000, all in regions Caravan already covers. **The archive cannot be cited as a dataset.** `Cite this` offers the software concept DOI `10.5281/zenodo.21903143` (`explorer/src/methods.js:82`). There is no DOI for `Rekin226/aquascope-gauges` itself, so anyone who uses the archive cites the library instead, which is the outcome the whole archive-first thesis was meant to avoid. ## What to change Three tracks, independent, smallest first. 1. **Say what it is.** Until the mirror is wide, describe it accurately in `ROADMAP.md`, the README, the dataset card and the Space description: a global catalog of 45,919 stations, plus mirrored daily records for N of them. The catalog number is the impressive one and it is not the same number. 2. **Give the dataset a DOI.** Deposit a versioned snapshot on Zenodo, get its own concept DOI, and add it to `Cite this` beside the software one, to `CITATION.cff`, and to the dataset card. This is the step that makes the citable artifact actually citable, and it is a weekend. 3. **Widen the mirror, Taiwan first.** Whatever quota governs which stations get mirrored, Taiwan discharge is the highest-value addition and depends on #211 (the WRA daily discharge collector). After that, raising the per-source cap matters more than adding sources. ## Steps 1. Audit the wording in ROADMAP, README, dataset card and Space description against the manifest numbers. 2. Zenodo deposit for the archive, DOI into `methods.js`, `CITATION.cff` and the dataset card. 3. Decide and document the mirroring policy (which stations, why, what the cap is) in the dataset card, so the number is a stated choice rather than a surprise. 4. Taiwan discharge once #211 lands. ## Acceptance criteria - [ ] No public text implies daily observations exist for all 45,919 stations - [ ] The archive has its own DOI, offered in `Cite this` alongside the software DOI - [ ] The dataset card states the mirroring policy and the current station count per variable - [ ] Taiwan discharge in the mirror (tracked by #211) Part of #188. The record-window issue matters here too: every bundle currently starts in 1986 because of the same 40-year default. Found while QA-ing the live Explorer on 2026-08-23 (build `bf7606a`).