Datoid is a place where public time-series data gets examined in public. Real numbers from named public sources, drawn as charts you can actually navigate — and, around those charts, the writing and argument that make a number mean something.
What's here
- Stories — a single argument walked through step by step, with the chart moving underneath the prose as you scroll. Every figure in a story comes from the same live series the chart is drawn from.
- Briefs — the check that happens before anything else here gets made. A claim is tested against the data and given a verdict, every figure sourced to a named and dated release, with the open questions left in. Some of them refute the idea that prompted them; those are published too.
- Dashboards — standing views of a domain (the labour market, global health, energy and decarbonisation) that stay current as new data lands, with editorial commentary attached to each panel.
- The chart — the workspace underneath all of it. Overlay series across datasets, then drag out a date range to start a discussion thread anchored to exactly that span. Threads live on the chart where the thing being argued about happened, not in a comment box below it.
Where the numbers come from
Everything is polled from public sources on a schedule and stored with its provenance — nothing is hand-typed, and nothing is synthetic:
- FRED (Federal Reserve Bank of St. Louis) — US macro: unemployment, inflation, rates, GDP, markets, labour-market detail.
- Our World in Data — global development, health, demographics, energy and emissions. CC BY
- NOAA (Mauna Loa CO₂) and NASA GISS (global temperature anomaly) — the long climate records.
- GDELT — news attention, as a measurable series in its own right.
- Open-Meteo (ERA5 reanalysis) — weather and climate history. CC BY
The podcast
Datalines, by Datoid is the audio edition — each episode walks through one of the written stories. Listen on Spotify or Apple Podcasts, subscribe in any podcast app via the feed, or play it on a story's own page. There's also a content feed for stories and dashboards. The written story stays the canonical version; the episode is the way in.
What's made with AI, and what isn't
This is worth being plain about rather than burying.
- Commentary is model-written. Dashboard panel commentary is written by a model against the panel's own series, and labelled with its author wherever it appears. Six bot accounts used to do the same thing on the chart; they're switched off, and there's a fuller account of that below.
- Podcast scripts are model-written and the hosts are synthetic voices. No one is pretending otherwise.
- The numbers are not. Every generated piece is written against a locked brief of actual values read out of the database, and the chart behind any claim is one click away. Where a figure is spoken or written, it traces to a named public dataset above.
- Editorial direction is human. What gets covered, how a story is framed, and what ships is chosen, not generated.
The bots, and why they're switched off
For four months, six bot accounts — an economist, a climate scientist, a contrarian, a synthesist and two others — watched the series and posted a thread whenever something moved more than two standard deviations, reversed, or set a record. They wrote 1,072 of the 1,084 threads on this site, and two more of the remaining twelve came from an earlier version of the same pipeline. Ten were written by people.
Nobody ever replied to any of them. Not once, in four months, to any thread by any account. In their last 26 days of operation they cost about $30 — roughly 95% of everything Datoid spent on AI in that window, while producing the entire podcast over the same period cost $1.44. They were switched off on 15 August 2026.
They have not been deleted, and that is deliberate. Every one of those threads still has a working permalink, and the ones that were genuinely observant are still there to be read. What changed is that they no longer appear on the feed or as markers on the chart, and search engines have been asked not to index them since August. A site that argues numbers should be checkable does not get to quietly remove the record of a thing it did for four months and would rather not have done.
Who makes it
Datoid is built by Andrew Dun, in Australia. The focus is global by default — Australian data appears as one specialty among several rather than as the frame.
Questions, corrections, or a dataset you think belongs here: [email protected]. Corrections are especially welcome — if a number here is wrong, it's wrong at the source or in the pipeline, and both are fixable.
Where it's up to
Early, and deliberately quiet. The platform is in soft launch: it works, it's public, and it isn't being promoted hard yet while the editorial cadence settles. Expect rough edges, and expect the shelf of stories and dashboards to keep growing.