Release 2026-08-01-r2 — Science-to-Decision

146 entities, 193 sources. Adds the consumer decision lifecycle. Uncertified — the harness cannot read this release’s schema, so no entity-level check has run against it.

Release notes for the 2026-08-01-r2 build of the Science-to-Decision Ecosystem Map. Records what changed, what is still wrong, and why this release carries no valid certification.

Published

August 1, 2026

Status: published, certified 2026-08-02 — DO NOT PUBLISH, 1 fail, 3 warn. This release shipped uncertified: the harness read the previous release’s schema and parsed zero entities, so every entity-level check silently did not run. The harness has since been fixed and this is its verdict.

Open the map ↗ · How entries get on it · The essay it came out of

What changed

This versioned refresh preserves the reviewed August baseline and extends coverage into consumer decision capacity.

25 entities added across primary care, supplements, consumer measurement, behaviour change and ecosystem convening.
The consumer decision lifecycle, which is the reason this release exists: encounter → explore → ground → evaluate → translate → contextualize → decide → act → observe → revise.
Function Health’s acquisition of SuppCo (2026-05-12) represented as one historical event carrying both parties’ announcements, rather than as two.
The Stanford Consumer Health Conference represented as a discovery network. A speaking slot is not evidence that a product does what it claims, so the speaker list feeds the candidate queue and nothing else.
Structural additions: stable entity and source IDs, explicit verification status, source dates or a stated date basis, a decision ledger, a candidate queue, and separate tables for events and classification changes.
Zero classification changes across 121 carried entities. The drift discipline held.

Certification

Verdict: DO NOT PUBLISH — 1 fail, 3 warn. Certified 2026-08-02, after publication.

Why it shipped uncertified

The r1 build used organizations and stages. This release renamed those to entities and split stage assignment across research_stages, decision_stage_primary and decision_stage_adjacent. The harness still read the old keys, so it parsed zero organizations from a 146-entity file and reported a structure failure.

That was worse than a failed check. Every entity-level test — stable IDs, stage-tuple duplication, tag elasticity, stage-hogging, justification coverage — did not run. They did not pass. They were skipped, silently, while the harness printed a verdict and exited non-zero. As a gate it would have blocked this release for the wrong reason without checking anything it claimed to check.

The harness now normalises both schemas, and a schema it cannot parse reports SCHEMA MISMATCH naming the unrecognised keys, stating that the entity checks were skipped rather than passed.

What the certification found

FAIL — 168 of 193 sources (87%) carry no date. Staleness is undetectable by construction and any maturity claim resting on them is unauditable. Worse than r1’s 53%.

WARN — 111 of 146 entities share 28 identical stage tuples. Nine evidence-synthesis tools carry the same triple; so do the five clinical-evidence tools and the seven provenance-rail entries. Identical tag sets across functionally different products mean a template was applied rather than each entry assessed. This is the measured version of the argument the inclusion tests make, and the strongest single reason not to quote any stage count from this release.

WARN — 132 of 146 entities (90%) have no corroborating source beyond their own domain. Worse than r1’s 81%.

WARN — 14 developments carry no evidence caveat. The caveats are the map’s distinguishing feature; an uncaveated event reads as an endorsement.

Notably, the breadth check did not fire. Mean stages per entity is well under threshold now that there are 31 stages, so this taxonomy discriminates better than r1’s on breadth while being less disciplined on repetition — a distinction the harness could not see until it could read the file.

The certification linked from the map page describes r1, not this release. That mismatch stands until the release bundle is restructured.

What is still wrong

Beyond the certification gap:

The delta cannot represent a removal. Insilico Medicine, Clarivate and Digital Science are in r1 and absent here with no recorded reason. The schema has no removals field, so a reader comparing releases sees 25 additions and no losses, which is false. They are recorded on the map as unexplained removals — a disclosure, not a fix.
“Reviewed baseline” overstates what was reviewed. 95 of 121 carried entries are marked baseline but were never in the reviewed 31. Baseline status is sticky, so entries nobody has checked are shielded by a mechanism meant to protect checked judgements.
One acquisition could not be verified. The Function Health / SuppCo transaction returns 403 to us. By our own rule that is blocked, not disproven, and it carries that note.
Stage counts remain our coding, not the companies’. They describe a vocabulary rather than a market and should not be quoted as findings. The inclusion tests exist to fix this and apply from September.

Corrections applied after publication

2026-08-02 — business-development flag removed. A collaboration_priority field shipped in this release’s embedded JSON on all 146 entities, and the release notes named two companies as collaboration priorities. That is an internal commercial flag and it should never have reached a public artifact. It has been stripped from the map and the line removed from these notes.

The publisher carries a guard against exactly this. It did not fire, because this release was published by copying the generator’s export into the site rather than by running the publisher — a chokepoint only works when everything passes through it. The guard now also needs to cover every file in a release bundle, not just the map.

The original claim is described here rather than deleted, because a correction that erases its own subject is not a correction.