Releases

Every build of the map, what changed in it, and whether it passed its own certification — including the ones that did not.

The release archive for the Science-to-Decision Ecosystem Map. Each entry records what changed, what the certification harness said, and what was still wrong at the time of publishing. Releases are immutable once published; corrections appear as new entries rather than edits.

The map is the artifact. This is the record. Each release keeps its own notes, its own certification result, and its own list of what was still wrong — so a claim made in August can be checked against the build that supported it.

Current

2026-08-01-r2 — Science-to-Decision 146 entities · 193 sources · 14 events Certified 2026-08-02: DO NOT PUBLISH — 1 fail, 3 warn. Adds the consumer decision lifecycle and renames the map from AI-for-Science to Science-to-Decision.

Superseded

2026-08-01 — AI-for-Science baseline 31 organizations · 60 sources · 27 developments Certified: DO NOT PUBLISH — 2 fail, 8 warn. The first build, kept live because r2 carries its entities forward and the reasoning that produced them should stay checkable.

How to read these

Immutable. A published release is not edited. If something in it was wrong, that appears as a correction in a later release, with the original left standing. An archive that quietly improves itself is not an archive.

Certification is published whether or not it passes. Both builds so far have failed their own harness. Publishing anyway is a deliberate choice, stated each time, on the grounds that a map with its holes marked is more useful than no map — but the verdict is never softened to make that easier.

Method changes and market changes are different things. From the September release onward, the changelog separates reclassification caused by a change in our own rules from movement in the world. A count that moved because we changed the definition is not a finding about the field.

Cadence

Monthly, from September 2026. September applies the stage inclusion tests and is therefore a method migration — its diff will be dominated by our own rule changes rather than by the market. October is the first month-over-month comparison that means anything, and only if no further method changes land in it.

There is no changelog yet. Both builds carry the same date, so no month-over-month pair exists to diff.