The producer's view. For each source: how many cards were distilled, how many passed your distillation review onto the shelves, and how many still await it. Freshly distilled zips land on the Intake tab, waiting for you to Import. ("Review" here always means the distillation review — your check on each freshly distilled card before it goes live.)
Add a source to distil
A PDF or epub — it uploads to the bank and the Distillery on your Mac pulls it and distils it at the depth you pick. Under 200MB. Freshly distilled cards land on the Intake tab.
Click to add a file or drag it here
PDF, epub, txt, md · up to 200MB
Recent uploads
Loading…
The import gate. The Distillery drops each finished distil here as a .zip; nothing enters the bank until you review the run and Import. Each item carries the story of the distillation that made it. Imported cards arrive as awaiting distillation review → Vet → Keep.
Retrieval metadata — geography · sector · effect direction — backfilled onto each card from its own text. These power filtering & ranking in search (the embedding is never touched). Rejected cards are skipped.
Candidate completed RCTs & peer-reviewed papers from top development/economics journals (via OpenAlex). None are in the bank — review, then Queue to distill: open the PDF and send it to your Distillery (Telegram bot or the app). Fetchable PDF ones distil directly; the rest need you to find the open-access copy. Refresh pulls the latest.
Every action on every card — the append-only audit trail. Curators and the AI models both write here (vet, dedup, reanchor by AI; keep / set aside / merge by you). The health strip below reads the whole log for a rough pulse on pipeline quality.
ActionBy
Pipeline monitor
→
Every distill's journey — who sent it on Telegram, when it was distilled (+flags), and when it reached Intake. Match the ID to the bot's /runs table.