v0.103.0 Release Recap

v0.103.0 is a non-breaking product and reliability release that makes ranking evaluation a first-class admin dashboard experience.

What's New

Area	New in v0.103.0	Why it matters
Ranking dashboard	/admin/ now opens with a Ranking Evaluation workspace	Operators can inspect retrieval quality without leaving the live admin surface
Evaluation API	GET /admin/api/ranking-eval?top_k=5	Exposes representative corpus MRR, pass/fail, categories, query detail, and score components
Baseline compare	Browser-side baseline JSON loading	Makes release-to-release ranking drift visible without uploading artifacts to the server
Query evidence	Top-k table and selected-query detail panel	Shows expected hits, correct rank, reciprocal rank, retrieval source, and component scores
Feature metadata	admin-ranking-eval-dashboard	Lets live deployments advertise the dashboard capability through version introspection

Operator Flow

Open the observe admin UI:

open http://localhost:7702/admin/

Or fetch the ranking report directly:

curl "http://localhost:7702/admin/api/ranking-eval?top_k=5"

Use the baseline upload control with a saved contextdb eval ranking --out JSON artifact when you want a quick release comparison.

Compatibility

This release is intentionally non-breaking. It adds an admin API and dashboard surface while preserving existing REST, GraphQL, SDK, storage, and CLI behavior.

Verification

Release verification included:

npm run admin:build
go test -count=1 ./internal/admin
go test -count=1 ./...
npm run docs:build
git diff --check

The next reliability slices should focus on ranking/dashboard baseline persistence, schema catalog drift annotations, closure bundle verification, and manifest-backed evidence bundles.