The Science-to-Decision Ecosystem Map
Who is building what between a research finding and a person deciding what to do — published with its own audit attached.
146 organizations across two lifecycles, research and consumer decision, with 193 dated sources. Every source carries an evidence class, every entry carries its verification status, and the review that says what is still wrong is published beside the map rather than kept internal.
Most landscape maps tell you who exists. This one follows the chain all the way to a person deciding what to do, and publishes the report card on itself — including the part that says this release does not yet pass it.
Open the map full-screen ↗ · The essay this came out of ↗ · How entries get on it ↗ · Release notes and certification ↗
What it is
146 organizations placed across two lifecycles. The research lifecycle runs from literature grounding through synthesis, peer review, and clinical or regulatory translation. The consumer decision lifecycle picks up where that stops, and is the reason this release exists: encounter, explore, ground, evaluate, translate, contextualize, decide, act, observe, revise.
193 sources, each dated and tagged with what kind of evidence it is: a peer-reviewed paper, an independent party’s announcement, a news report, or an organization describing itself. That last column is the reason to build it. Most market maps cite company websites and present the result as market intelligence. Here the source class travels with the claim.
The finding worth the trouble
Follow the chain to its end and the last stage empties out. Almost nothing is built to revise: to find out the thing is not working, and stop.
That is structural rather than lazy. Stopping needs a rule set in advance. A rule needs to know whether the thing is working. That needs a benefit signal, and we built surveillance for harm and almost none for benefit. So nothing tells you it is working, no stopping rule can be written, and nothing flows back to steer what gets studied next. Everything clustered near the decision is built to start and to sustain, because stopping requires a measure nobody has.
What is still wrong with it
This release is published uncertified, and the reason is worth stating on the first screen rather than in a footnote: the certification harness reads the previous release’s schema, so it parses zero entities from this one and every entity-level check has silently not run.
The checks that still work found that 87% of sources carry no date and 88% are organizations describing themselves — defensible for what a product claims to do, not for whether it works.
Three findings stand beyond that: the delta cannot represent a removal, “reviewed baseline” covers 95 entries nobody reviewed, and one acquisition is blocked rather than verified. Stage counts remain our coding rather than the companies’, so they describe a vocabulary and should not be quoted as findings.
Full release notes, with the certification detail and what was corrected after publishing →
Keeping the detail in the release record rather than on this page is deliberate. A map that accumulates every past defect on its front screen becomes a catalogue of its own mistakes, and stops being usable for the thing it is for. The record should be complete; the map should be current.
How entries get on it
Every stage now has a written inclusion test, and each is written so that it excludes something. A stage whose test excludes nothing is a vocabulary rather than a classification, and its counts cannot be quoted.
They were adopted in August and apply from the September release, so the map above does not yet reflect them. Where the two disagree, the method page is the standard and this map is behind it. Applying them cuts the decision layer from seven organizations to one — a smaller, less impressive, more defensible number, and the first one on this page that could be quoted.
What changes each release
Each refresh produces a changelog: who was added, who was removed, who quietly repositioned, and the one worth watching, whose claim moved from a press release to an independently verified result.
The map is the artifact. The changelog is the product.
There is no changelog yet. Both builds so far carry the same date — r2 superseded the August baseline the day it shipped — so there is no month-over-month pair to diff and none can exist before September. That edition will produce the first one, and it will be dominated by the taxonomy change rather than by the market: a method diff, not a market one. October is the first honest like-for-like comparison. Saying so now seems better than letting a monthly promise accumulate quietly against an empty directory.
What would make us stop
The essay this map came out of closes by asking what a product would have to observe to tell you to stop using it. It would be a poor advertisement for that standard if the map had no answer.
So: if the November release has not drawn corrections from people who build in this space, the map drops to quarterly or moves to the archive. Correction-as- contribution is the whole premise — that a landscape published with its holes marked invites people to fill them. If nobody does, the premise is wrong, and the right response is to stop rather than to keep publishing a monthly artifact nobody corrects.
That is a real threshold with a date attached, and it is the same test we ask of everyone else on the map.
Fix something
Every entry carries a suggest a correction link that opens a message naming that specific organization. If you build in this space and we have described you wrongly, that is the fastest way to fix it, and corrections get credited in the following release.
Every build, with its notes and its certification — the release archive. Releases are immutable once published; corrections appear as new entries rather than edits.
Comments