Milestones
List view
Corpus export (CSV/Parquet) with codebook + provenance, diffusion/network analysis, geo, sampling + bias-map, and cost tracking.
No due date•0/5 issues closedOK, Dzen, and Rutube connectors; near-duplicate + cross-network linking; period summaries; chat-to-data over a SQL allow-list; a custom front end with filtering and export.
No due date•0/7 issues closedLemmatization (Natasha), TF-IDF/c-TF-IDF + NER word clouds, BERTopic with a Russian embedder + topics-over-time, spike detection, and a provider-agnostic LLM layer (topic labels, config assistant, output audit, aspect/actor sentiment).
No due date•0/7 issues closedTelethon-based Telegram connector, full per-network pools with sticky proxies + health, flood-wait/captcha handling, Celery/Redis idempotent queues, metric-snapshot history.
No due date•0/8 issues closedOne connector end-to-end. Acceptance: a single VK job runs on schedule, fetches only new material via hash/cursor increment, never duplicates, and renders a single-job dashboard.
No due date•0/10 issues closed