scythe410/vectorscape-reducer

fix(reducer): make Redis-backed progress optional; /status survives without Redis

feat(reducer): add /embed primitive endpoint for demo search

fix(reducer): drop bogus reducer/ COPY breaking HF Space build

fix: replace arq/Redis queue with in-process background task to fix Vercel FUNCTION_INVOCATION_TIMEOUT

fix(reducer): snippet-fallback labels + in-place /relabel endpoint

fix(reducer): keyword fallback when IDF zeros every term in a cluster

fix(reducer): always queue embed-reduce + warm-load embedder on boot

chore(reducer): drop orphan reducer/ shim package

chore(reducer): refresh uv.lock

fix(reducer): phase 6a — mypy clean (F-1, F-2, F-3)

chore(qa): phase 5 — static audit baseline + findings recorded

test(qa): phase 4 — security tests

test(qa): phase 3 — integration tests against fakes + RLS SQL files

test(qa): phase 2 — unit tests for pure logic

feat(adjacency): faint links between semantically-near clusters

feat(search): aggregate matches by cluster, surface named regions

feat(reducer): real cluster labels via c-TF-IDF, opt-in LLM upgrade

feat(search): natural-language search over project galaxy

feat(bridge): gate Gemini free-tier path behind user consent

feat(reducer): add Gemini fallback for /bridge LLM summary

chore(reducer): dockerize for hugging face spaces deploy

chore: reduce CSV preview to 20 rows & add multi-topic sample dataset

fix(security): apply qa_fixes — auth, prompt injection, upload cap, race lock, truncation, sanitization

feat(bridge): explain shared theme + contrast between two clusters

feat(web): SKM lens, landing page, and pre-baked demo galaxy

feat(reducer): arq worker for >10k jobs + /status endpoint with error capture

feat(reducer): /embed-reduce pipeline (MiniLM + conditional PCA + PaCMAP + HDBSCAN) and CLI harness

chore: scaffold bun-workspaces monorepo