| --- |
| license: other |
| pretty_name: Thoth OSINT archive |
| tags: |
| - osint |
| - geospatial |
| - time-series |
| configs: |
| - config_name: events |
| data_files: |
| - split: train |
| path: data/events/*.parquet |
| - config_name: endpoint_calls |
| data_files: |
| - split: train |
| path: data/endpoint_calls/*.parquet |
| - config_name: raw_events |
| data_files: |
| - split: train |
| path: data/raw_events/*.parquet |
| - config_name: source_runs |
| data_files: |
| - split: train |
| path: data/source_runs/*.parquet |
| - config_name: layer_samples |
| data_files: |
| - split: train |
| path: data/layer_samples/*.parquet |
| - config_name: collector_runs |
| data_files: |
| - split: train |
| path: data/collector_runs/*.parquet |
| - config_name: sanctions_entities |
| data_files: |
| - split: train |
| path: data/sanctions_entities/*.parquet |
| - config_name: event_dups |
| data_files: |
| - split: train |
| path: data/event_dups/*.parquet |
| - config_name: feed_health |
| data_files: |
| - split: train |
| path: data/feed_health/*.parquet |
| - config_name: collector_schedule |
| data_files: |
| - split: train |
| path: data/collector_schedule/*.parquet |
| - config_name: layer_versions |
| data_files: |
| - split: train |
| path: data/layer_versions/*.parquet |
| - config_name: schema_migrations |
| data_files: |
| - split: train |
| path: data/schema_migrations/*.parquet |
| - config_name: sanctions_meta |
| data_files: |
| - split: train |
| path: data/sanctions_meta/*.parquet |
| - config_name: worker_heartbeat |
| data_files: |
| - split: train |
| path: data/worker_heartbeat/*.parquet |
| --- |
| |
| # Thoth OSINT archive |
|
|
| Full data backup of [Thoth](https://github.com/azzindani/Thoth), a self-hosted |
| open-source intelligence terminal, taken on 2026-11-02 when the deployment was |
| shut down. It covers 2026-09-24 to 2026-11-02: about 970k normalized events from |
| 342 public sources across 44 map layers (aviation weather, weather, cyber |
| advisories, transit, flights, fires, news, markets, quakes and more), plus the |
| raw upstream payloads and the collectors' run logs. |
|
|
| Repo commit at backup time: `481fd92` (v0.2.0). |
|
|
| ## Layout |
|
|
| - `data/<table>/part-NNNN.parquet`: every table of the `public` schema, one |
| folder per table (zstd Parquet). Geometry columns are WKT text (SRID 4326), |
| `jsonb` and `numeric` columns are text, and the derived `events.search` |
| tsvector is left out (it is rebuilt from title and body). |
| - `backups/thoth-final-2026-11-02.dump`: `pg_dump -Fc` of the whole database at |
| shutdown, identical snapshot to the Parquet files. |
| - `backups/thoth-pre-v0.2.0.dump`: the earlier pre-upgrade dump from |
| 2026-09-28. It holds raw payloads from before the raw-events retention window. |
| - `manifest.json`: row counts and column types per table. |
|
|
| ## Tables |
|
|
| | Table | Rows | Contents | |
| |---|---:|---| |
| | `events` | 972,295 | Normalized map points: one row per observation (id, time, source, layer, title, body, url, severity, confidence, WKT geometry, entities + meta JSON). | |
| | `endpoint_calls` | 879,786 | Per-request upstream call log: host, path, status, latency, bytes, error. | |
| | `raw_events` | 742,697 | Raw upstream payloads as fetched (JSON text), with source, layer, HTTP status and payload hash. Retention kept 2026-10-19 onward. | |
| | `source_runs` | 534,056 | Per-source run outcomes (ok / error). | |
| | `layer_samples` | 484,317 | Per-layer, per-cell point counts over time (drives trend charts). | |
| | `collector_runs` | 132,634 | Per-collector run history: duration, item count, error, trigger. | |
| | `sanctions_entities` | 20,792 | Sanctions list entities (name, aliases, countries, dataset, meta JSON). | |
| | `event_dups` | 377 | | |
| | `feed_health` | 352 | | |
| | `collector_schedule` | 75 | | |
| | `layer_versions` | 44 | | |
| | `schema_migrations` | 11 | | |
| | `sanctions_meta` | 1 | | |
| | `worker_heartbeat` | 1 | | |
| | `collector_requests` | 0 | | |
| | `entity_links` | 0 | | |
| | `notes` | 0 | | |
| | `pins` | 0 | | |
| | `portfolios` | 0 | | |
| | `positions` | 0 | | |
| | `screens` | 0 | | |
| | `sitreps` | 0 | | |
| | `watchlists` | 0 | | |
| | `webhook_deliveries` | 0 | | |
|
|
| Empty tables are included so the schema is complete. |
|
|
| ## Restore |
|
|
| Into a fresh Thoth stack (`backend/docker-compose.yml`, PostgreSQL 16 + |
| TimescaleDB + PostGIS): |
|
|
| ```bash |
| docker compose up -d db |
| docker compose exec -T db psql -U thoth -d thoth -c "SELECT timescaledb_pre_restore();" |
| docker compose exec -T db pg_restore -U thoth -d thoth --no-owner < thoth-final-2026-11-02.dump |
| docker compose exec -T db psql -U thoth -d thoth -c "SELECT timescaledb_post_restore();" |
| docker compose up -d --build |
| ``` |
|
|
| ## Licence |
|
|
| Upstream data belongs to its publishers and keeps their terms. The per-source |
| attribution and licence notes are in Thoth's |
| [data sources reference](https://github.com/azzindani/Thoth/blob/main/docs/reference/data-sources.md). |
|
|
