1 · Claude
Repository assessment
Claude inspected the project and reported three failing tests.
Connected to the current service
See TestGraph working with live V2 data.
Browse the semantic graph, keep the full classification hierarchy in view, and inspect the selected type, its direct edges and public subjects without losing context.
Sources: /api/v2/vocabulary and /api/v2/public/experiences. No V1 data is used.
—subject types
—type relationships
—public experiences
—reviewed subjects
—AI assessments
Explore the live vocabulary
The bubble graph shows connectivity; the Classification hierarchy shows where each type sits conceptually; the selected-node panel explains direct edges and public subjects.
Bubble viewLoading live vocabulary…
Select a bubble, tree node or dropdown item · drag to pan · wheel or buttons to zoom.
Classification hierarchyPersistent semantic context
Loading hierarchy…
Typed mode note: repeated entries are the same type and stable ID reached through more than one valid
is_a path.Public experiences
Only one published experience is shown at a time. Use the selector to move between them.
Loading deliberately published V2 experiences…
A published deliberation replay
A compact case study showing cross-model work while keeping the private deliberation store closed.
2 · ChatGPT
Independent critique
ChatGPT verified the evidence, refined the diagnosis and found a fourth CI defect.
3 · GitHub
Repair and checks
The import-order and test-collection problems were fixed, reviewed and merged.
4 · Claude
Reconciliation
Claude independently rechecked the result after deployment.
Verified outcome: contributions and corrections remained attributable throughout.
What can—and cannot—appear here
Public by deliberate choice
The server filters V2 records to visibility=public and publication_status=published. Public subject classification summaries may show status and model labels, but omit client IDs, private evidence and owner data.
Private by default
Private, unlisted and aggregate-only experiences do not appear. Deliberations remain private unless a separate sanitised case study is explicitly approved for publication.