Milestones
List view
A fifth of genuine hits are refused, trying it means choosing an embedder first, proving it on your own traffic means writing the harness yourself, and the guards that are the whole differentiator run on the JVM alone. Each of those is measured or visible in the repository today and each of them loses an adoption decision on its own. This tier attacks them in the order a team hits them. Effort: two L, four M, one S.
Due by October 29, 2027•0/7 issues closedA standard living in one library's subdirectory is that library's appendix. The metric, the schema and the rules exist and are implementable; what has never happened is somebody outside running them. This tier gives the measurement a name, a home and a population that matches what a cache actually meets, ships the scorer where the caches are, and gets a cache this project did not write onto the same table. The library's distribution runs through this: whoever decides how semantic caches are judged decides which one looks good. Effort: three L, two M, one S.
Due by July 30, 2027•0/7 issues closed