hackathon-spark

Use when working with official Bundestag data in graph-protokoll: parsing plenary-protocol XML (dbtplenarprotokoll DTD), Saalreaktionen from <kommentar>, fetching/resolving sessions (Open Data, DIP-API, Mediathek/YouTube), the dashboard/accessible outputs, or the protocol↔video gap analysis (WER). Trigger on ingest_bundestag.py, bundestag_xml.py, dashboard.py, accessible.py, gap_analysis.py, scripts/fetch-session.sh, *.xml tasks.

Use when working on the audio→text front of graph-protokoll: speech recognition (ASR), speaker diarization, word/speaker alignment, sound-event detection (Beifall/Buhrufe/Lachen), or subtitle/VTT ingestion. Trigger on tasks touching pipeline/asr.py, diarize.py, align.py, sound_events.py, subtitles.py, faster-whisper, pyannote, WhisperX, PANNs, or `--audio` runs.

Use when working on claim verification in graph-protokoll: detecting checkable statements, assigning verdicts (bestätigt/teilweise/irreführend/falsch/unbelegt), attaching sources, the evidence corpus, or the "every fact-check has a source" invariant. Trigger on pipeline/factcheck.py, data/evidence/evidenz.json, Aussage/Verdikt/Quelle/Evidenz tasks.

Use when answering natural-language questions over the protocol graph in graph-protokoll: the offline intent-router (pipeline/graphrag.py), the Neo4j Text2Cypher path (pipeline/neo4j_graphrag.py), the NEO4J_SCHEMA / few-shot examples, multi-hop traversals, or query-with-provenance work. Trigger on GraphRAG/Cypher/Text2Cypher/"ask the graph" tasks.

Use when extracting structure from a meeting or building/exporting the protocol knowledge graph in graph-protokoll: turning Utterances into a Protocol, building the SPARK-format graph, the 5-layer ontology (normativ/zeitlich/prozedural/fallbezug/provenienz), provenance edges, and JSON/CSV/Cypher/Neo4j export. Trigger on pipeline/extract.py, graph_build.py, export.py, neo4j_loader.py, node/relationship/ontology/Cypher work.