Changelog

  1. Draft stories get an annotation review loop, and the story checks catch broken charts

    Draft data stories are now published to a private-link Flowershow site with annotations on, so a reviewer can select any passage and leave a note, and an agent pulls the notes with fl annotations pull, revises the story and republishes; drafts stay off this site until approved. The story checks gained S7, which fails any chart SVG containing NaN, undefined or leaked function source, and the story skill now asks writers to state the difference or ratio rather than leave two numbers for the reader to compare.

  2. /validate now checks the data, not just the metadata

    /validate now reads every CSV value against the dataset's declared schema (types, real calendar dates, row width, primary key uniqueness, foreign keys and the house CSV format) and says which line is wrong; all seven DataPressr datasets pass clean.

  3. A test bench for better data stories

    We now run the story skill blind on real questions, have a model from another vendor critique the results, and turn the lessons into skill edits that we re-test on new questions, with the owner's blind preference as the measure that counts. The story lab page explains what we optimise, how the loop works and what the first rounds have shown, including the first open-data story run (WWII) and a France A/B now awaiting the owner's blind judgement.

  4. All four data stories, and seven datasets, now on DataHub

    The other three data stories are now live on DataHub beside the oil story: The Keeling Curve, Planetary Boundaries and Where France's public money goes. Like the oil story they are drafts pending the author's voice pass, and the site pages stay as the working copies.

    Six more datasets are published with them, so every story's data link now lands on a DataHub dataset page, and those pages link back to the stories: co2-ppm, france-public-finances, airports, population-growth, tesla-quarterly-deliveries and us-natural-hazard-statistics. The datasets page links each one. Everything is listed at datahub.io/datapressr.

    The Keeling Curve story on DataHub: title, description and the annual-mean CO₂ chart
  5. Installed skills now work outside the DataPressr repo

    After npx skills add datasets/datapressr, init now scaffolds a dataset with its conventions file and validator in an empty folder, and validate runs the real checks instead of improvising them. The skills carry what they need and link to everything else by full URL, and capture files ideas in your own project (Beads, its issues or a local INBOX.md) rather than in DataPressr's inbox.

  6. First data story published on DataHub

    "WTI Went Negative. Brent Didn't." is now live on DataHub at datahub.io/datapressr/wti-went-negative, beside the oil-prices dataset it is built from, and each now links to the other. It is published as a draft pending the author's voice pass; the site page stays as the working copy.

    The oil story on DataHub: title, description and the Brent vs WTI chart
  7. The datasets page now lists every dataset

    The datasets page is now generated from each dataset's own metadata, so it shows all eight datasets with their current status, licence, source, story and, for oil prices, its DataHub page.

  8. A first look at DataPressr inside BB

    A working proof of concept previews data tables, dataset READMEs, and stories with charts inside BB, including beside an AI conversation, and updates the preview as files change. We also proved that the AI can build, reload, and visually test the plugin itself—an essential step toward a usable data workspace.

    This is a local prototype using sample files, not a released plugin yet.

    DataPressr's BB prototype showing a story and chart, with Data and README previews available
  9. Parsed datasets now get an adversarial review before they count as structured

    A fresh reviewer re-derives values from the source, reads every label and tries to break the build.

    The structure skill now requires an independent review before a dataset built with a custom parser (PDF, HTML, messy spreadsheets) or from many source documents is marked structured, and recommends one for everything else: the reviewer re-derives values from the source, reads every category label, deliberately breaks the build to see whether any check notices, and records an APPROVED verdict. In both of our source-discovery runs, the most valuable finding came from this kind of review.

  10. A freshly scaffolded dataset now passes validation

    The validator no longer fails a new stub for having no data files yet.

    A dataset fresh from /init now passes /validate with no errors or warnings, since a stub with no data files yet is publishable; missing resources are still an error from archived onward.

Built with LogoFlowershow