extract-deputy-activity

name	extract-deputy-activity
description	Extract parliamentary speech and voting activity for tracked deputies from a parliament transcript or session document. Use for plenary transcripts, committee minutes, or roll-call votes where you need to attribute statements and decisions to specific members.
allowed-tools	Read

Extract deputy activity

Identify interventions, votes, or substantive parliamentary contributions by deputies on tracked topics.

Inputs (in the user message)

• country, language, trustTier, sourceLabel, sourceUrl
• categoryIds — valid topic ids
• deputies — array of { id, name, party } for the tracked roster
• documentText

Output

{
  "entries": [
    {
      "kind": "deputy_activity",
      "confidence": 0.82,
      "sourceQuote": "verbatim 3 to 25 words from documentText",
      "deputyName": "exact name from deputies list",
      "deputyId": "id from deputies list",
      "topic": "one of categoryIds or null",
      "activityType": "speech" | "vote" | "question" | "motion",
      "summary": "what they did in 1 to 2 sentences"
    }
  ]
}

Rules

1. deputyName and deputyId MUST resolve to a single entry in deputies. If a name is ambiguous (two deputies share a surname), use confidence ≤ 0.4 and include both candidates as separate entries.
2. topic is the primary topic of the intervention. If the intervention spans multiple, pick the dominant one. null is acceptable for procedural interventions.
3. sourceQuote MUST appear verbatim in documentText.
4. Don't double-count: a single speech is one entry, even if it touches multiple categories.
5. Don't fabricate activityType — if you can't tell whether it was a speech vs. a question, use speech and lower confidence by 0.1.
6. If nothing relevant, return { "entries": [] }.