Research library
StaleStrong freshnessArtifacts and learned candidates produced by AutoResearch experiments.
- Documents
- 36
- Chunks
- 488
- Tokens
- 288,100
- Storage
- 11.6 MB
- Embedded
- 100%
- Updated
- 4d ago
strong
Retrieval precision
Artifacts retrieve precisely.
0
Stale documents
No stale documents.
0
Unembedded docs
Fully embedded.
0
Orphan chunks
No orphaned chunks.
healthy
Source coverage
Experiment and learned sources present.
Inventory
Documents and their chunk metadata. Open any document for source type, embedding status, token counts, and chunk ids, never raw content or vectors.
Retrieval
Canned query examples show which chunks were returned, the coarse similarity score, and why each chunk matched.
“What retrieval threshold did the experiment settle on?”
Run by AutoResearch
- 0.93chk_rt_1 · doc_exp_retrieval
Final decision recorded in the artifact.
These are curated, canned queries. Arbitrary querying over private documents is intentionally disabled.
Learning
Constrained autolearn: candidates from feedback pass eval and reviewer gates before promotion, with demotion and rollback when they regress.
Autolearn is constrained: feedback creates candidates and promotion gates, it does not silently mutate production. Reviewer decisions and rollbacks are shown below.
- cand_stale_detectorAutoResearchPromotedEval: StrongReviewer: approvedSignals:reviewer 3decided 5d ago
Freshness timeline
Recent writes, re-embeds, retrieval regressions, and candidate promotions and rollbacks.
Candidate promoted×1
5d ago
KB write×6
4d ago