Not hidden — overwritten. Agencies restate figures in place. Registers show only what is true today. Rankings and leaderboards keep no history at all. By the time a question is worth asking, the evidence needed to answer it is gone, and no one had to decide to remove it.
These repositories capture it first, on a schedule, in public. Analysis built on them is published at getqed.app.
86 sources across 8 archives, as of September 2026. Every fetch is recorded whether the bytes changed or not, so the archives answer when did this change and not merely what does it say now.
| repository | what it keeps | cadence |
|---|---|---|
| wss-gho | WHO Global Health Observatory indicators. WHO restates history in place; this keeps what it said before. | monthly |
| wss-drug-scarcity | Which medicines the US cannot get and for how long — shortages, import bans, recalls, inspections. | weekly · monthly |
| wss-hugging-face | Model and dataset adoption from first release to abandonment. | weekly |
| wss-cloud-footprint | Where the internet's physical infrastructure is actually being built. | weekly |
| wss-openrouter | What inference costs and which labs are gaining share. | weekly |
| wss-food-trace | What is in the food supply, and what was withdrawn from it. | monthly |
| wss-forest-harvest | Forest harvest permissions and what follows them. | weekly · monthly |
| wss-mining-pipeline | Mineral tenements and exploration permits — registers show today, and what lapsed is not recorded. | monthly |
Each archive is a plain git repository. The captured bytes, the record of every fetch, and the code that produced them are all in the same place, and you can clone the whole thing.
wss-engine runs all of them and holds no data itself. An archive declares its sources in a registry; the engine fetches on schedule via GitHub Actions and commits what it saw. Adding a source is one YAML file — there is never a workflow per source.
What the engine guarantees, because an archive is worth nothing if you cannot trust how it was made:
- Raw bytes, stored verbatim. No parsing, cleaning or normalising between the publisher and the archive. Derived tables are rebuilt from the bytes and are always reproducible from them.
- A record of every fetch, including fetches where nothing changed. Absence of change is itself an observation, and an archive that only records diffs cannot tell you the difference between stable and not looked at.
- Two clocks, kept separate. When a thing was true, and when we saw it. Conflating them is how archives quietly invent history.
- Failures are loud. A response that fails its gate is quarantined rather than stored. A run that captures nothing goes red. Nothing is silently skipped.
- Polite by construction. robots.txt is honoured, requests are delayed and identified with a contact address. We are guests on other people's servers.
Write-ups at getqed.app cite the archive and commit they were computed from, so a claim can always be traced back to the bytes it came from — and the archives point forward to the analyses that use them. A finding you cannot re-derive from a specific commit is a finding we would not publish.
Archives carry a CITATION.cff. wss-forest-harvest
is deposited with Zenodo at 10.5281/zenodo.22647878;
others follow as they stabilise. Everything here is open — code and data are
licensed separately in each repository, and captured material remains under
whatever terms its original publisher set.
- open-research — methodology and documentation for reproducing analyses on public datasets
- parawl — Southeast Asia parliamentary document crawler and parser