Lage Bonn

Live: https://lage.bonn.machdenstaat.de/

A Lagebild for Bonn: aggregates local RSS/Atom feeds and open government APIs into one geo-tagged, deduplicated Atom stream. Deep at the local (Ortsteil) level rather than thin and broad.

Approach (a): deterministic curation — fetch, geo-tag, dedup, emit. No LLM, no rewriting of source text. LLM synthesis and a scheduled Morgenlage digest are later phases.

Pipeline

SQLite registry ─▶ fetch ─▶ geo-tag ─▶ dedup ─▶ SQLite ─▶ Atom (out/feed.atom)
   feeds (DB)       RSS+JSON   gazetteer  cluster  items      georss:point

Both the feed registry and all items live in SQLite, with pydantic as the single schema authority (LageItem + a discriminated Source union). sources.yaml is a git-tracked seed: lage feed import validates it through pydantic and upserts it into the DB (also auto-seeded on first collect when the registry is empty). After that the DB is authoritative — edit feeds with lage feed create/delete.

• fetch — RSS via feedparser; Autobahn + DWD via JSON/WFS (no auth).
• geo-tag — Bonn gazetteer (4 Stadtbezirke, ~50 Ortsteile). Ortsteil names matched in item text (Bonn-Röttgen → Röttgen/Bonn); Autobahn/DWD use native coordinates; GA district feeds tag at fetch time. Rhein-Sieg neighbor Gemeinden (Alfter, Siegburg, …) are detected so the bare word "Bonn" cannot mis-claim them → "Umland (Rhein-Sieg)". No-place items → "Bonn allgemein". Region-wide feeds can set bonn_only: true to drop items that resolve to no Bonn/Umland place.
• dedup — exact (guid/link hash, SQLite PK) + fuzzy (normalized-title similarity within 36h) so the same event from Polizei + GA + Rundschau collapses into one Atom entry that cites all sources.
• out — Atom 1.0 with georss:point, geo/category/source categories.

Sources (verified live, auth-free)

Category	Sources
Blaulicht	Polizei Bonn, Feuerwehr & Rettungsdienst Bonn, Bundespolizei Dir. Sankt Augustin (NRW-wide, bonn_only-filtered)
News	General-Anzeiger (Stadt + Bonn/Bad Godesberg/Beuel/Hardtberg), Bonner Rundschau
Amtlich	Bundesstadt Bonn Presse
Wissenschaft	Universität Bonn
Events	Bonn Veranstaltungskalender
Verkehr	Autobahn API (A565/A555/A59 roadworks + warnings)
Wetter	DWD weather warnings (Stadt Bonn)

Deferred (need work/auth): Bonn OParl RIS (HTTP 500 at build time), WDR Lokalzeit (no working feed), SWB/VRS disruptions (scrape or API key), Amtsblatt (PDF only).

Usage

uv run lage feed collect          # one poll cycle, respects per-source cadence
uv run lage feed collect --force  # ignore cadence, fetch everything now
uv run lage feed collect -vv      # DEBUG: per-item geo + dedup scores
uv run lage feed collect --window-hours 48   # Atom output time window
uv run lage                       # bare alias for `feed collect` (back-compat)

Output: out/feed.atom. State: data/lage.db (SQLite).

Status

uv run lage status                # data-store stats + per-source fetch health

Reports store size, item counts (total / last 24h / 72h), dedup effectiveness, the covered time span, breakdowns by category / Stadtbezirk / geo-source, and per source the last fetch outcome (ok/fail), consecutive failure count, last item count, and an overdue flag when a source has missed twice its cadence.

Manage sources

uv run lage feed import           # seed/refresh the DB registry from sources.yaml
uv run lage feed list             # show configured sources (from the DB)
uv run lage feed show polizei-bonn   # one source: config + live stats (DB)
uv run lage feed create --id radio-bonn --title "Radio Bonn" \
    --url https://example.org/feed.xml --category news --poll-minutes 30
uv run lage feed create --id bpol --title "Bundespolizei" \
    --url https://.../feed.rss --category blaulicht --bonn-only
uv run lage feed delete --id radio-bonn
uv run lage feed delete --all     # wipe the registry (re-seed with `feed import`)

create/delete write the SQLite registry directly (DB authority); import seeds/refreshes it from the git-tracked sources.yaml. Pydantic validates every source — a bad --category is rejected before it reaches the DB. Flags: --stadtbezirk (district scope), --bonn-only (drop non-Bonn items), --user-agent browser (for hosts that block bots).

Serve the feed

uv run lage serve                       # http://127.0.0.1:8080/
uv run lage serve --host 0.0.0.0 --port 8099

Serves a human-readable HTML view at / and /feed, and the raw out/feed.atom as application/atom+xml at /feed.atom and /atom (GET + HEAD). Dev/local convenience — it does not run the pipeline; keep fetching on the timer/cron. For production, front the file with nginx or any static server.

Google Maps traffic panel (optional)

The dashboard at / can show a live traffic map of Bonn powered by the Google Maps JavaScript API. It is hidden when GOOGLE_MAPS_API_KEY is unset.

One-time setup:

1. Open Google Cloud Console → APIs & Services → Library.
2. Search for Maps JavaScript API and click Enable.
3. Go to APIs & Services → Credentials → Create Credentials → API key.
4. Copy the key, then restrict it:
   • Application restrictions: HTTP referrers → add http://localhost:8080/* (and your production domain if applicable).
   • API restrictions: restrict to Maps JavaScript API only.
5. Add the key to .envrc:

export GOOGLE_MAPS_API_KEY=AIza...your_key_here

6. Reload the env (direnv allow if you use direnv, or source .envrc), then restart lage serve. The Verkehrslage panel appears automatically.

Deployment (continuous stream)

Deterministic, crash-safe — a timer wakes the app every 5 min; the per-source poll_minutes in the SQLite registry decides what actually gets fetched.

mkdir -p ~/.config/systemd/user
cp contrib/systemd-unit-files/lage.{service,timer} ~/.config/systemd/user/
systemctl --user daemon-reload
systemctl --user enable --now lage.timer

Or cron:

*/5 * * * * cd ~/Source/lage && uv run lage feed collect

Serve out/feed.atom with any static file server; feed readers consume it.

Roadmap

• (b) LLM synthesis: cluster → one written Lage paragraph per event.
• Scheduled Morgenlage digest (06:00) as a second Atom feed / email.
• OSM street-level gazetteer (Overpass) for sub-Ortsteil precision.
• Recover OParl (council decisions), add SWB/VRS disruptions.