Implement Meeting Context V1 and extraction improvements

Introduce Meeting Context V1 with YAML schema, validation and template.
Support optional --meeting-context during chunk extraction.
Inject authoritative Meeting Context into extraction prompts.
Record Meeting Context provenance in extraction output.
Activate todos.md in shared prompt assembly.
Strengthen responsibility attribution and decision/todo boundaries.
Add focused Gold scenarios and validation tests.
Update architecture and pipeline documentation.

.gitignore

@@ -21,6 +21,8 @@ dist/
 
 # Test
 .pytest_cache/
+.test-tmp/
+.test-tmp-root/
 .coverage
 htmlcov/
 

CHANGELOG.md

@@ -28,6 +28,14 @@
   validation and source fact coverage.
 - Semantic Consolidator V0 benchmark report and consolidated extraction JSON.
 - Gold regression scenario for responsibility attribution integrity.
+- Meeting Context V1 YAML scaffold, generic template and documentation for
+  manually maintained meeting metadata.
+- Meeting Context V1 loader, validator, deterministic prompt representation,
+  optional `--meeting-context` extraction CLI integration and extraction
+  context provenance.
+- PyYAML project dependency for Meeting Context YAML loading.
+- Focused Gold scenarios for responsibility attribution and explicit position
+  extraction.
 
 ### Changed
 
@@ -48,7 +56,12 @@
   views.
 - Updated decision extraction semantics to include explicit process decisions
   and deferrals.
-- Simplified extraction prompt assembly around prompt files.
+- Shared extraction prompt assembly now loads `common.md`, `decisions.md` and
+  `todos.md`.
+- Added explicit action-item responsibility attribution rules and clarified
+  the decision/todo boundary, including duplicate-classification handling.
+- Marked Meeting Context integration with later pipeline stages as planned
+  while extraction-stage integration is implemented.
 
 ### Fixed
 

PROJECT_KNOWLEDGE.md

@@ -23,9 +23,13 @@ Implemented:
   `src/meeting_lab/consolidation/consolidate_facts.py` for facts-only
   semantic duplicate detection.
 - Prompt loading from `src/meeting_lab/llm/prompts.py`.
+- Meeting Context V1 loading, validation and optional extraction prompt
+  injection with minimal extraction JSON provenance.
 - Interim Markdown protocol generation in `src/meeting_lab/protocol/`.
 - Non-LLM unit tests for chunking, extraction helpers, protocol rendering and
   gold-test runner validation.
+- Meeting Context V1 scaffold and documentation for manually maintained
+  meeting metadata.
 
 Experimental/prototype:
 
@@ -38,6 +42,8 @@ Planned:
 - Broader semantic consolidation for topic grouping, contradiction handling,
   uncertainty marking and durable/transient separation.
 - Canonical Meeting Knowledge implementation as the semantic source of truth.
+- Meeting Context integration with Canonicalizer, Semantic Consolidator,
+  Canonical Meeting Knowledge and output renderers.
 - Final Working Protocol / Arbeitsprotokoll, Distribution Protocol /
   Verteilerprotokoll and Knowledge Objects / Wissensdatenbankeintrag renderers.
 
@@ -59,8 +65,10 @@ src/meeting_lab/
 Supporting areas:
 
 - `docs/`: architecture, pipeline, data models and output-view concepts.
-- `prompts/`: active prompt files. Only `common.md` and `decisions.md` contain
-  substantive extraction prompt text in the current tree.
+- `docs/meeting-context.md`: Meeting Context V1 scaffold, fields and future
+  integration rules.
+- `prompts/`: active extraction prompt files. The shared extraction prompt is
+  assembled from `common.md`, `decisions.md` and `todos.md`.
 - `tests/gold/`: semantic gold tests and prompt-engineering methodology.
 - `samples/`: sample inputs and generated or experimental artifacts.
 - `scripts/`: operational scripts for cleanup and gold-test execution.
@@ -111,6 +119,13 @@ Accepted decision semantics:
 - A process decision to defer a substantive decision is still a decision.
 - "No decision was reached" is different from "the group decided to defer the
   decision."
+- A personal commitment to perform concrete future work is normally a todo,
+  not a decision, unless the group also establishes a separate binding outcome,
+  rule, approval, rejection, deferral, selection, process state or
+  responsibility policy.
+- The same proposition should not be duplicated under decisions and todos.
+  Extract both only when the transcript contains a group-level decision and a
+  semantically separate resulting action item.
 
 ## Responsibility Attribution
 
@@ -118,6 +133,13 @@ Meeting Lab distinguishes mentioned people, speakers, participants,
 responsible people, departments, owners and assignees. These concepts must not
 be collapsed into one field.
 
+Meeting Context V1 reinforces this distinction by separating actual
+participants from mentioned non-participants and by storing aliases, roles and
+departments only when they are explicitly supplied as metadata. It must not be
+used to infer responsibilities. In the current implementation this context can
+be injected into chunk extraction prompts as authoritative metadata, and only
+minimal provenance is written to extraction JSON.
+
 A `responsible` or future `owner` / `assignee` value may be recorded only when
 source evidence explicitly assigns, accepts or confirms responsibility. If the
 evidence is incomplete or ambiguous, the responsible person remains `null` or
@@ -125,6 +147,16 @@ unset and the evidence is preserved. Future schema work may add
 `responsibility_status` values such as `explicit`, `accepted`, `proposed` and
 `unclear`, plus `attribution_evidence`.
 
+Meeting Context may validate identity, role, department and attendance, but it
+never establishes responsibility.
+
+Focused Gold coverage now separates these concerns:
+
+- `responsibility_attribution_negative`: tests that discussion, objection and
+  department proximity do not create an owner.
+- `position_explicit_objection`: tests explicit position extraction separately
+  from responsibility attribution.
+
 Current Prompt Version 2 decision baseline:
 
 - `decision_simple`: passing.
@@ -208,10 +240,14 @@ language is requested.
   are the intended architecture.
 - Semantic Consolidator V0 is implemented only for facts-only duplicate
   detection.
+- Meeting Context V1 is implemented only through chunk extraction; later-stage
+  integration remains planned.
 - Canonical Meeting Knowledge is documented but not implemented.
 - Final output views are documented but not implemented.
-- Most prompt files are placeholders except the common and decision prompts.
-- Gold tests currently emphasize extraction semantics, especially decisions.
+- Some prompt files remain placeholders; `common.md`, `decisions.md` and
+  `todos.md` are active in the shared extraction prompt.
+- Gold tests currently emphasize extraction semantics, especially decisions,
+  todos, responsibility attribution and focused position extraction.
 
 ## Next Recommended Engineering Step
 

README.md

@@ -108,6 +108,21 @@ und normalisiert Extraktionsobjekte ohne LLM. Semantic Consolidator V0 nutzt
 das lokale LLM nur fuer konservative facts-only Duplikaterkennung, erhaelt
 Evidenz und erzeugt noch keine Canonical Meeting Knowledge.
 
+Meeting Context V1 ist als manuell gepflegtes YAML-Geruest dokumentiert und
+fuer die Chunk-Extraktion implementiert. Die Extraktion kann den Kontext
+optional validieren, als autoritative Metadaten in den Prompt aufnehmen und
+minimale Kontext-Provenienz im Extraction JSON speichern. Konsolidierung,
+Canonical Meeting Knowledge und Rendering sind noch nicht daran angeschlossen.
+
+Der gemeinsame Extraktionsprompt besteht aktuell aus `common.md`,
+`decisions.md` und `todos.md`. Die Todo-Regeln verlangen explizite Zuweisung,
+Freiwilligenmeldung oder Annahme, bevor eine verantwortliche Person gesetzt
+wird. Meeting Context kann Identitaet, Rolle, Abteilung und Anwesenheit
+validieren, begruendet aber niemals Verantwortung. Entscheidungen und Todos
+werden nicht aus derselben Proposition doppelt extrahiert; beides wird nur
+ausgegeben, wenn eine gruppenweite Entscheidung und eine davon getrennte
+Folgeaufgabe vorliegen.
+
 Das Meeting Lab behandelt "das Protokoll" nicht mehr als ein einzelnes
 Endprodukt. Das konsolidierte Meeting-Wissen ist die **Canonical Meeting
 Knowledge**, also die kanonische semantische Repräsentation eines Meetings und
@@ -138,10 +153,12 @@ werden, sofern keine explizite Ausgabesprache angefordert wurde.
 
 Aktuell liegt der Schwerpunkt auf der Entwicklung eines modularen
 Diskussionsanalyzers. Implementiert sind Vorverarbeitung, technisches Chunking,
-lokale Chunk-Extraktion, Canonicalizer V1 als deterministische Vorbereitung der
-Extraktionsergebnisse und Semantic Consolidator V0 fuer facts-only
-Duplikaterkennung. Canonical Meeting Knowledge, breitere semantische Synthese
-und finale Output-View-Renderer sind geplante nächste Schritte.
+lokale Chunk-Extraktion, Meeting Context V1 fuer die Extraktionsstufe,
+Canonicalizer V1 als deterministische Vorbereitung der Extraktionsergebnisse
+und Semantic Consolidator V0 fuer facts-only Duplikaterkennung. Canonical
+Meeting Knowledge, breitere semantische Synthese, Meeting-Context-Integration
+in spaetere Stufen und finale Output-View-Renderer sind geplante naechste
+Schritte.
 
 Canonicalizer V1 kann aus dem Repository heraus so ausgeführt werden:
 

ROADMAP.md

@@ -193,6 +193,10 @@ Deliverables:
 - FFmpeg integration.
 - Meeting metadata capture.
 - Participant entry.
+- Meeting Context V1 exists as a manually maintained YAML structure with
+  validation, optional extraction prompt integration and extraction provenance;
+  future work should add GUI entry and conservative integration with later
+  pipeline stages.
 - GUI.
 - Stable deployment process.
 - Export workflows.

docs/architecture.md

@@ -188,6 +188,24 @@ Each extractor has exactly one task and one prompt.
 
 ---
 
+## Meeting Context
+
+Meeting Context V1 is a manually maintained YAML scaffold for reliable meeting
+metadata such as title, language, participants, aliases, departments,
+abbreviations and known entities.
+
+It is documented in `docs/meeting-context.md` and templated at
+`samples/templates/meeting_context.template.yaml`. It is implemented for
+loading, validation and optional injection into chunk extraction prompts.
+Extraction results record only minimal context provenance. It is not yet
+connected to consolidation, Canonical Meeting Knowledge or output rendering.
+
+The context can help prevent non-participants from being interpreted as
+attendees and can normalize known aliases for extraction. It must not infer
+roles, departments, responsibilities or decisions.
+
+---
+
 ## consolidation/
 
 Planned area for canonicalization and consolidation.
@@ -283,6 +301,7 @@ Implemented:
 - Transcript normalization
 - Technical chunk generation
 - Experimental LLM-based information extraction
+- Meeting Context V1 loading, validation and extraction prompt integration
 - Canonicalizer V1 deterministic extraction canonicalization
 
 The current extraction step still performs multiple tasks simultaneously.

docs/data-models.md

@@ -218,6 +218,54 @@ Owner remains empty if unknown.
 
 ---
 
+# Meeting Context V1
+
+Manually maintained YAML metadata scaffold with an implemented Python loader,
+validator and deterministic prompt renderer for chunk extraction.
+
+Top-level structure:
+
+```yaml
+schema_version: "1"
+meeting: {}
+participants: []
+mentioned_people: []
+organization: {}
+known_entities: {}
+context_rules: {}
+```
+
+Meeting Context separates participants, mentioned people, transcript speakers,
+responsible people, roles and departments. It is authoritative only for
+explicitly supplied metadata. It must not be used to infer responsibilities,
+decisions or commitments.
+
+Template:
+
+- `samples/templates/meeting_context.template.yaml`
+
+Documentation:
+
+- `docs/meeting-context.md`
+
+When `--meeting-context` is supplied to extraction, output JSON receives only
+minimal provenance:
+
+```json
+{
+    "context": {
+        "meeting_id": "...",
+        "source_file": "...",
+        "schema_version": "1"
+    }
+}
+```
+
+Later Canonicalizer, Semantic Consolidator, Canonical Meeting Knowledge and
+renderer integration remains planned.
+
+---
+
 # Topic Result
 
 After extraction, every topic contains the collected information.

docs/meeting-context.md

@@ -0,0 +1,300 @@
+# Meeting Context V1
+
+## Purpose
+
+Meeting Context V1 is a manually maintained YAML file for reliable meeting
+metadata. It gives later pipeline stages known names, aliases, organizational
+terms and vocabulary without asking an LLM to infer them from a transcript.
+
+The context is authoritative only for metadata that is explicitly supplied in
+the file. It must not be used to infer responsibilities, decisions or
+commitments.
+
+## Scope
+
+V1 is implemented for loading, validation and optional injection into the
+chunk extraction prompt. It is not yet integrated with the Canonicalizer,
+Semantic Consolidator, Canonical Meeting Knowledge or output renderers.
+
+Supported metadata:
+
+- meeting title, language, date, objective and notes
+- actual participants and participant aliases
+- participant role and department when known
+- known non-participants mentioned during the meeting
+- known departments and aliases
+- abbreviations
+- relevant products, projects, systems, locations and technical terms
+
+## File Locations
+
+- Generic template: `samples/templates/meeting_context.template.yaml`
+- Real-Life Sample context:
+  `samples/real_live/project_process_meeting/meeting_context.yaml`
+- Future meeting-specific contexts should live next to the meeting input data.
+
+## Field Descriptions
+
+`schema_version`: Version of the Meeting Context file shape.
+
+`meeting.title`: Required for future GUI entry. Human-readable meeting title.
+
+`meeting.meeting_id`: Required stable meeting identifier used for extraction
+context provenance.
+
+`meeting.language`: Required for future GUI entry. Dominant meeting language,
+for example `de` or `en`.
+
+`meeting.date`: Optional ISO date or `null`.
+
+`meeting.objective`: Optional objective entered by the user.
+
+`meeting.notes`: Optional neutral notes about context or scope.
+
+`participants`: Actual meeting attendees. Participant names are required for
+future GUI entry.
+
+`participant_id`: Stable identifier. It should not change when display names
+or aliases are corrected.
+
+`display_name`: Preferred display name.
+
+`aliases`: Alternative spellings, short forms or Whisper variants.
+
+`role`: Organizational function. Optional and nullable.
+
+`department`: Organizational unit. Optional and nullable.
+
+`attendance_status`: `present` for participants. This distinguishes attendees
+from mentioned people.
+
+`mentioned_people`: People discussed or referenced but not present. They are
+not participants and must not be treated as speakers.
+
+`organization.name`: Optional organization name.
+
+`organization.departments`: Known departments with stable ids, names and
+aliases.
+
+`organization.abbreviations`: Known abbreviation expansions. Empty strings mean
+the expansion is not yet confirmed.
+
+`known_entities`: Meeting vocabulary for projects, products, systems,
+locations and technical terms. These lists do not imply responsibility.
+
+`context_rules`: Conservative defaults for later integrations.
+
+## Filled Example
+
+```yaml
+schema_version: "1"
+
+meeting:
+  title: "Projektprozess fuer neue Initiativen"
+  language: "de"
+  date: "2026-08-01"
+  objective: "Klaeren, wie der bestehende Projektprozess angepasst wird."
+  notes: ""
+
+participants:
+  - participant_id: "martin"
+    display_name: "Martin"
+    aliases: ["Martin T."]
+    role: null
+    department: null
+    attendance_status: "present"
+    notes: null
+
+mentioned_people:
+  - person_id: "alex"
+    display_name: "Alex"
+    aliases: []
+    role: null
+    department: null
+    attendance_status: "not_present"
+    notes: "Wurde erwaehnt, war aber nicht anwesend."
+
+organization:
+  name: null
+  departments:
+    - id: "pm"
+      name: "PM"
+      aliases: ["Projektmanagement"]
+  abbreviations:
+    PM: "Projektmanagement"
+    BD: "Business Development"
+    MK: ""
+    GF: ""
+
+known_entities:
+  projects: []
+  products: []
+  systems: []
+  locations: []
+  technical_terms: ["Lastenheft"]
+
+context_rules:
+  participant_list_is_authoritative: true
+  do_not_infer_roles: true
+  do_not_infer_departments: true
+  do_not_infer_responsibilities: true
+  mentioned_people_are_not_participants: true
+```
+
+## Concept Distinctions
+
+Participant: a person who actually attended the meeting.
+
+Mentioned person: a person discussed or referenced but not present.
+
+Speaker: transcript-level attribution, which may be unknown or unreliable.
+
+Responsible person: a person explicitly assigned to or accepting an action
+item.
+
+Role: the person's organizational function.
+
+Department: the organizational unit to which the person belongs.
+
+These concepts must never be collapsed automatically. A participant may discuss
+a topic outside their own department. Discussing, objecting to or suggesting
+work does not establish responsibility.
+
+## Editing Guidance
+
+Enter only objective context that is known independently or confirmed by the
+user. Do not guess uncertain names, roles, departments or abbreviation
+expansions. Leave unknown values as `null`, an empty string or an empty list.
+
+Use aliases for spelling variants, shortened names and Whisper variants. Keep
+stable ids unchanged after a context file has been used in experiments.
+
+Do not copy responsibilities from generated protocols into Meeting Context.
+Responsibilities belong to evidence-backed extraction and later action-item
+models, not to this metadata file.
+
+## Extraction Integration
+
+The chunk extraction CLI accepts an optional Meeting Context file:
+
+```text
+PYTHONPATH=src .venv/bin/python -m meeting_lab.extraction.extract_chunks \
+  samples/chunks/chunk_01_normalized.txt \
+  -o /tmp/chunk_01_extraction.json \
+  --meeting-context samples/real_live/project_process_meeting/meeting_context.yaml
+```
+
+When supplied, the file is validated and rendered as deterministic
+authoritative metadata inside the extraction prompt. When omitted, prompt
+construction and extraction output remain unchanged.
+
+The shared extraction prompt is assembled from:
+
+- `common.md`
+- `decisions.md`
+- `todos.md`
+
+Extraction JSON receives only minimal context provenance:
+
+```json
+{
+  "context": {
+    "meeting_id": "2026-07-27-projektprozess",
+    "source_file": "samples/real_live/project_process_meeting/meeting_context.yaml",
+    "schema_version": "1"
+  }
+}
+```
+
+Full personal metadata is not copied into every extraction result.
+
+Meeting Context loading uses PyYAML when installed. `PyYAML>=6.0` is declared
+as a project dependency.
+
+## Prompt Behavior
+
+Meeting Context validates identity, role, department and attendance, but never
+establishes responsibility.
+
+The todo prompt records a named responsible person only when the transcript
+explicitly assigns the task, the person volunteers, or the person accepts or
+confirms the task. Addressing a person, requesting department input, discussing
+department-specific criteria, stating expertise, objecting, suggesting, or
+listing Meeting Context role metadata is not enough.
+
+The decision prompt treats a personal commitment to concrete future work as a
+todo unless the group separately establishes a binding outcome, rule, approval,
+rejection, deferral, selection, process state or responsibility policy. The
+same proposition should not be duplicated under decisions and todos; extract
+both only when the transcript contains semantically separate propositions.
+
+## Validation Rules
+
+The current validator checks that:
+
+- YAML syntax is valid
+- `schema_version` is supported
+- `meeting.meeting_id`, `meeting.title` and `meeting.language` are present
+- participant ids are unique and non-empty
+- mentioned-person ids are unique and non-empty
+- participant ids and mentioned-person ids do not collide
+- referenced departments exist in `organization.departments`
+- `attendance_status` values are valid
+- participants are marked `present`
+- mentioned people are not marked `present`
+
+Invalid values are not inferred or repaired.
+
+## Gold Tests
+
+Two focused Gold scenarios cover the main responsibility boundary:
+
+- `responsibility_attribution_negative`: discussion, objection and department
+  proximity must not create a named owner.
+- `position_explicit_objection`: explicit objection extraction is tested
+  separately as a position scenario.
+
+Responsibility attribution and position extraction are intentionally tested
+independently so each scenario has unique ground truth.
+
+## Confidentiality
+
+Meeting Context can contain real names, departments, project names and internal
+terminology. Treat it as confidential meeting data. Keep private sample
+contexts inside the private repository and avoid exposing them in screenshots,
+logs or generated reports.
+
+## Future GUI Entry
+
+Future product UI should make these fields easy to enter before processing.
+
+Required GUI fields:
+
+- meeting title
+- language
+- participant names
+
+Optional GUI fields:
+
+- objective
+- aliases
+- role
+- department
+- mentioned non-participants
+- abbreviations
+- known entities
+- notes
+
+## Future Pipeline Integration
+
+Later stages may use Meeting Context to normalize names, recognize aliases,
+avoid treating absent mentioned people as speakers and avoid expanding
+abbreviations incorrectly. This later-stage integration is still planned.
+
+Integration must remain conservative:
+
+- Meeting Context may supply metadata only.
+- It must not infer decisions.
+- It must not infer responsibilities.
+- It must not override source evidence.
+- It must preserve uncertainty when transcript evidence is ambiguous.

docs/pipeline.md

@@ -67,6 +67,12 @@ Output View Rendering
 
 Each stage receives a well-defined input and produces a well-defined output.
 
+Meeting Context V1 exists as a manually maintained YAML metadata scaffold. It
+is implemented for validation, optional `--meeting-context` use during chunk
+extraction, deterministic prompt injection and minimal extraction JSON
+provenance. Later Canonicalizer, Semantic Consolidator, Canonical Meeting
+Knowledge and renderer integration remains future work.
+
 ---
 
 # Stage 1 – Normalization
@@ -284,6 +290,22 @@ Each extractor has:
 - one responsibility
 - one output schema
 
+The current shared extraction prompt is assembled from:
+
+- `common.md`
+- `decisions.md`
+- `todos.md`
+
+The todo prompt requires explicit assignment, volunteering or acceptance before
+recording a named responsible person. Meeting Context may validate identity,
+role, department and attendance, but never establishes responsibility.
+
+The decision prompt treats a personal commitment to concrete future work as a
+todo unless the group separately establishes a binding outcome, rule, approval,
+rejection, deferral, selection, process state or responsibility policy. The
+same proposition should not be duplicated under decisions and todos; extract
+both only for semantically separate propositions.
+
 ## Processing Type
 
 LLM

prompts/decisions.md

@@ -58,6 +58,25 @@ If a statement is only a proposal or suggestion, do not extract it.
 If participants discuss something but do not explicitly agree to it, do not
 extract it.
 
+A personal commitment to perform concrete future work is normally an action
+item, not a decision. Do not extract it as a decision unless the group also
+establishes a binding outcome, rule, approval, rejection, deferral, selection,
+process state, or responsibility policy beyond the personal work assignment
+itself.
+
+Do not duplicate the same proposition under decisions and todos.
+
+A single transcript passage may nevertheless contain both:
+
+- a group-level decision, rule, approval, rejection, deferral or process state
+- and a distinct resulting action item
+
+Extract both when they are semantically separate propositions, even if they
+occur in the same sentence or evidence passage.
+
+A concrete personal work commitment without a separate group-level outcome
+belongs only under todos.
+
 If the transcript describes an existing process, rule, template, document, or
 workflow, do not extract it unless the participants explicitly adopt or change it
 in this meeting.

prompts/todos.md

@@ -0,0 +1,16 @@
+Action-item responsibility rule:
+
+A named responsible person may be extracted only when the transcript contains
+clear evidence that the person was explicitly assigned the task, explicitly
+volunteered, or explicitly accepted/confirmed the task.
+
+The following are not assignments: addressing a person while discussing a
+topic; saying that a person or department will have different criteria;
+requesting input from a department; stating expertise, competence, or
+organizational role; suggestions, expectations, objections, or preferences;
+Meeting Context role or department metadata by itself.
+
+When work is clearly needed but no person accepted it, keep the action item
+without a responsible person if the schema allows it; otherwise do not create a
+named assignment. Meeting Context may validate identity, role, and attendance,
+but never establishes responsibility.

pyproject.toml

@@ -0,0 +1,7 @@
+[project]
+name = "meeting-lab"
+version = "0.1.0"
+requires-python = ">=3.11"
+dependencies = [
+  "PyYAML>=6.0",
+]

samples/real_live/project_process_meeting/README.md

@@ -31,6 +31,9 @@ Original input files:
 
 Reference material:
 
+- `meeting_context.yaml`: manually maintained Meeting Context V1 scaffold for
+  reliable metadata. It can be validated and injected into chunk extraction
+  with `--meeting-context`; later pipeline stages are not connected yet.
 - `reference/human_reference_protocol.md`: not currently included. No exact
   existing human-written protocol file was found in the repository workspace.
 

samples/real_live/project_process_meeting/meeting_context.yaml

@@ -0,0 +1,135 @@
+schema_version: "1"
+
+meeting:
+  meeting_id: "2026-07-27-projektprozess"
+  title: "Projektprozess für neue Projektideen in F&E, Marketing und Business Development"
+  language: "de"
+  date: 2026-07-27
+  objective: ""
+  notes: "Manuell gepflegter Kontext; noch nicht an die Pipeline angeschlossen."
+
+participants:
+
+  - participant_id: "martin"
+    display_name: "Martin"
+    aliases:
+      - "Martin Tazl"
+      - "Herr Tazl"
+    role: "Leiter F&E und Technik"
+    department_id: "fe"
+    attendance_status: "present"
+    notes: null
+
+  - participant_id: "lars"
+    display_name: "Lars"
+    aliases:
+      - "Lars Vollmert"
+      - "Herr Vollmert"
+    roles:
+      - "Leiter EDD"
+      - "Leiter Product Management"
+    department_id: "pm"
+    attendance_status: "present"
+    notes: null
+
+  - participant_id: "malte"
+    display_name: "Malte"
+    aliases:
+      - "Jan-Malte"
+      - "Herr Schnau"
+    role: "Product Manager GreenLine"
+    department_id: "pm"
+    attendance_status: "present"
+    notes: null
+
+  - participant_id: "bjoern"
+    display_name: "Björn"
+    aliases:
+      - "Björn-Erik"
+      - "Herr Falkenau"
+    role: "Leiter Marketing"
+    department_id: "mk"
+    attendance_status: "present"
+    notes: null
+
+mentioned_people:
+
+  - person_id: "jovana"
+    display_name: "Jovana"
+    aliases:
+      - "Jovana Husemann"
+      - "Frau Husemann"
+      - "Giovanna"
+      - "Jovanna"
+      - "Giovana"
+    role: "Leiterin Business Development"
+    department_id: "bd"
+    attendance_status: "not_present"
+    notes: null
+
+organization:
+
+  name: "Naue GmbH & Co. KG"
+
+  departments:
+
+    - id: "fe"
+      name: "F&E"
+      aliases:
+        - "Forschung und Entwicklung"
+        - "Research and Development"
+
+    - id: "pm"
+      name: "Product Management"
+      aliases: []
+
+    - id: "mk"
+      name: "Marketing"
+      aliases: []
+
+    - id: "bd"
+      name: "Business Development"
+      aliases: []
+
+  abbreviations:
+
+    PM: "Product Management"
+    BD: "Business Development"
+    MK: "Marketing"
+    GF: "Geschäftsführung"
+
+known_entities:
+
+  projects: []
+
+  products:
+    - "Bentofix"
+    - "Carbofol"
+    - "Combigrid"
+    - "Secugrid"
+    - "Secutex"
+    - "SoftRock"
+
+  systems: []
+
+  locations:
+    - "Adorf"
+    - "Bückeburg"
+    - "Espelkamp"
+    - "Malaysia"
+
+  technical_terms: []
+
+context_rules:
+
+  participant_list_is_authoritative: true
+
+  do_not_infer_roles: true
+
+  do_not_infer_departments: true
+
+  do_not_infer_responsibilities: true
+
+  do_not_infer_attendance: true
+
+  mentioned_people_are_not_participants: true

samples/templates/meeting_context.template.yaml

@@ -0,0 +1,74 @@
+# Meeting Context V1 template.
+# This file is manually maintained. Enter only objective context that is known
+# before or independently of model output.
+schema_version: "1"
+
+meeting:
+  # Human-readable title for the meeting.
+  title: ""
+  # Dominant meeting language, for example "de" or "en".
+  language: "de"
+  # Optional ISO date. Leave null if unknown.
+  date: null
+  # Optional meeting objective. Do not infer it from generated summaries.
+  objective: ""
+  # Optional neutral notes about context, scope or source material.
+  notes: ""
+
+participants:
+  # participant_id must remain stable across corrections and later runs.
+  # Use aliases for alternative spellings, short names or Whisper variants.
+  # attendance_status distinguishes actual participants from mentioned persons.
+  # role and department may remain null; uncertain values must not be guessed.
+  - participant_id: ""
+    display_name: ""
+    aliases: []
+    role: null
+    department: null
+    attendance_status: "present"
+    notes: null
+
+mentioned_people:
+  # People discussed or referenced but not present in the meeting.
+  # Mentioned people are not speakers and must not become responsible persons
+  # unless the meeting evidence explicitly assigns or confirms responsibility.
+  - person_id: ""
+    display_name: ""
+    aliases: []
+    role: null
+    department: null
+    attendance_status: "not_present"
+    notes: null
+
+organization:
+  name: null
+
+  departments:
+    # Department ids should be stable. aliases capture spelling variants.
+    - id: ""
+      name: ""
+      aliases: []
+
+  abbreviations:
+    # Fill in only abbreviations that are known for this meeting context.
+    # Leave values empty when expansion is unknown.
+    PM: ""
+    BD: ""
+    MK: ""
+    GF: ""
+
+known_entities:
+  # Relevant projects, products, systems, locations and technical terms.
+  # These lists provide vocabulary only; they must not imply responsibility.
+  projects: []
+  products: []
+  systems: []
+  locations: []
+  technical_terms: []
+
+context_rules:
+  participant_list_is_authoritative: true
+  do_not_infer_roles: true
+  do_not_infer_departments: true
+  do_not_infer_responsibilities: true
+  mentioned_people_are_not_participants: true

src/meeting_lab/extraction/extract_chunks.py

@@ -20,6 +20,11 @@ from typing import Any
 import requests
 
 from src.meeting_lab.llm.prompts import build_extraction_prompt
+from src.meeting_lab.models.meeting_context import (
+    MeetingContext,
+    load_meeting_context,
+    render_meeting_context_for_prompt,
+)
 
 
 DEFAULT_MODEL = "qwen3:8b"
@@ -36,6 +41,8 @@ EXTRACTION_CATEGORIES = (
 
 NORMALIZED_CHUNK_RE = re.compile(r"^(chunk_\d+)_normalized\.txt$")
 
+EXTRACTION_TASK_PROMPT_NAMES = ("decisions.md", "todos.md")
+
 
 OUTPUT_SCHEMA = {
     "chunk": {
@@ -138,11 +145,31 @@ def parse_args() -> argparse.Namespace:
         default=32768,
         help="Context window tokens per chunk (default: 32768)",
     )
+    parser.add_argument(
+        "--meeting-context",
+        type=Path,
+        help="Optional Meeting Context V1 YAML file to inject into extraction prompts.",
+    )
     return parser.parse_args()
 
 
-def build_prompt(source_name: str, transcript: str) -> str:
-    return build_extraction_prompt(source_name, transcript, OUTPUT_SCHEMA)
+def build_prompt(
+    source_name: str,
+    transcript: str,
+    meeting_context: MeetingContext | None = None,
+) -> str:
+    context_text = (
+        render_meeting_context_for_prompt(meeting_context)
+        if meeting_context is not None
+        else None
+    )
+    return build_extraction_prompt(
+        source_name,
+        transcript,
+        OUTPUT_SCHEMA,
+        meeting_context=context_text,
+        task_prompt_names=EXTRACTION_TASK_PROMPT_NAMES,
+    )
 
 
 def call_ollama(
@@ -376,12 +403,13 @@ def extract_chunk(
     temperature: float,
     num_predict: int | None,
     num_ctx: int | None,
-) -> dict[str, list[str]]:
+    meeting_context: MeetingContext | None = None,
+) -> dict[str, Any]:
     transcript = chunk_path.read_text(encoding="utf-8-sig").strip()
     if not transcript:
         raise ValueError(f"The input file is empty: {chunk_path}")
 
-    prompt = build_prompt(chunk_path.name, transcript)
+    prompt = build_prompt(chunk_path.name, transcript, meeting_context=meeting_context)
     raw_text, _metadata = call_ollama(
         endpoint=endpoint,
         model=model,
@@ -402,6 +430,8 @@ def extract_chunk(
         ) from exc
 
     extraction = normalize_current_schema(parsed)
+    if meeting_context is not None:
+        extraction["context"] = meeting_context.provenance()
     output_path.parent.mkdir(parents=True, exist_ok=True)
     output_path.write_text(
         json.dumps(extraction, ensure_ascii=False, indent=2) + "\n",
@@ -419,6 +449,7 @@ def extract_input(
     temperature: float,
     num_predict: int | None,
     num_ctx: int | None,
+    meeting_context: MeetingContext | None = None,
 ) -> list[Path]:
     if input_path.is_file():
         output_path = output or extraction_path_for_chunk(input_path)
@@ -431,6 +462,7 @@ def extract_input(
             temperature,
             num_predict,
             num_ctx,
+            meeting_context=meeting_context,
         )
         return [output_path]
 
@@ -459,6 +491,7 @@ def extract_input(
             temperature,
             num_predict,
             num_ctx,
+            meeting_context=meeting_context,
         )
         output_paths.append(output_path)
 
@@ -469,6 +502,11 @@ def main() -> int:
     args = parse_args()
 
     try:
+        meeting_context = (
+            load_meeting_context(args.meeting_context)
+            if args.meeting_context is not None
+            else None
+        )
         output_paths = extract_input(
             input_path=args.input,
             output=args.output,
@@ -478,6 +516,7 @@ def main() -> int:
             temperature=args.temperature,
             num_predict=args.num_predict,
             num_ctx=args.num_ctx,
+            meeting_context=meeting_context,
         )
     except requests.ConnectionError:
         print(
@@ -496,6 +535,8 @@ def main() -> int:
         return 1
 
     print(f"Input:                 {args.input}")
+    if args.meeting_context is not None:
+        print(f"Meeting context:       {args.meeting_context}")
     print(f"Processed chunks:      {len(output_paths)}")
     print(f"Extraction JSON files: {len(output_paths)}")
     for output_path in output_paths:

src/meeting_lab/llm/prompts.py

@@ -30,12 +30,14 @@ def build_extraction_prompt(
     source_name: str,
     transcript: str,
     output_schema: dict[str, Any],
+    meeting_context: str | None = None,
     task_prompt_names: Iterable[str] = ("decisions.md",),
     prompts_dir: Path = PROMPTS_DIR,
 ) -> str:
     schema_text = json.dumps(output_schema, ensure_ascii=False, indent=2)
     prompt_parts = [
         load_prompt("common.md", prompts_dir),
+        meeting_context,
         *load_existing_prompts(task_prompt_names, prompts_dir),
         f"""Quelldatei:
 {source_name}

src/meeting_lab/models/meeting_context.py

@@ -0,0 +1,396 @@
+"""Meeting Context V1 loading, validation and prompt rendering."""
+
+from __future__ import annotations
+
+import ast
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+
+SUPPORTED_SCHEMA_VERSIONS = {"1"}
+VALID_ATTENDANCE_STATUSES = {"present", "not_present", "absent"}
+
+
+class MeetingContextValidationError(ValueError):
+    """Raised when a Meeting Context file is structurally invalid."""
+
+
+@dataclass(frozen=True)
+class MeetingContext:
+    data: dict[str, Any]
+    source_file: Path
+
+    @property
+    def schema_version(self) -> str:
+        return str(self.data["schema_version"])
+
+    @property
+    def meeting_id(self) -> str:
+        return str(self.data["meeting"]["meeting_id"])
+
+    def provenance(self) -> dict[str, str]:
+        return {
+            "meeting_id": self.meeting_id,
+            "source_file": str(self.source_file),
+            "schema_version": self.schema_version,
+        }
+
+
+def load_meeting_context(path: Path) -> MeetingContext:
+    loaded = _load_yaml(path)
+    if not isinstance(loaded, dict):
+        raise MeetingContextValidationError("Meeting Context must be a YAML object.")
+
+    validate_meeting_context(loaded)
+    return MeetingContext(data=loaded, source_file=path)
+
+
+def validate_meeting_context(data: dict[str, Any]) -> None:
+    schema_version = str(data.get("schema_version", "")).strip()
+    if schema_version not in SUPPORTED_SCHEMA_VERSIONS:
+        raise MeetingContextValidationError(
+            f"Unsupported meeting context schema_version: {schema_version!r}."
+        )
+
+    meeting = _require_mapping(data, "meeting")
+    _require_non_empty_string(meeting, "meeting.meeting_id")
+    _require_non_empty_string(meeting, "meeting.title")
+    _require_non_empty_string(meeting, "meeting.language")
+
+    organization = _optional_mapping(data.get("organization"), "organization")
+    departments = _optional_list(organization.get("departments"), "organization.departments")
+    department_ids = _collect_department_ids(departments)
+
+    participants = _optional_list(data.get("participants"), "participants")
+    mentioned_people = _optional_list(data.get("mentioned_people"), "mentioned_people")
+
+    participant_ids = _collect_unique_ids(participants, "participant_id", "participants")
+    person_ids = _collect_unique_ids(mentioned_people, "person_id", "mentioned_people")
+    collisions = sorted(participant_ids & person_ids)
+    if collisions:
+        raise MeetingContextValidationError(
+            "Participant IDs and mentioned-person IDs must not collide: "
+            + ", ".join(collisions)
+        )
+
+    for index, participant in enumerate(participants):
+        item_path = f"participants[{index}]"
+        _validate_attendance(participant, item_path)
+        if participant.get("attendance_status") != "present":
+            raise MeetingContextValidationError(
+                f"{item_path}.attendance_status must be 'present'."
+            )
+        _validate_department_reference(participant, item_path, department_ids)
+
+    for index, person in enumerate(mentioned_people):
+        item_path = f"mentioned_people[{index}]"
+        _validate_attendance(person, item_path)
+        if person.get("attendance_status") == "present":
+            raise MeetingContextValidationError(
+                f"{item_path}.attendance_status must not be 'present'."
+            )
+        _validate_department_reference(person, item_path, department_ids)
+
+
+def render_meeting_context_for_prompt(context: MeetingContext) -> str:
+    data = context.data
+    meeting = data["meeting"]
+    organization = _optional_mapping(data.get("organization"), "organization")
+    departments_by_id = {
+        str(department.get("id")): str(department.get("name"))
+        for department in _optional_list(
+            organization.get("departments"), "organization.departments"
+        )
+        if isinstance(department, dict) and department.get("id") and department.get("name")
+    }
+
+    lines = [
+        "MEETING CONTEXT V1 (AUTHORITATIVE METADATA)",
+        "",
+        "Rules:",
+        "- The participant list is authoritative.",
+        "- Mentioned people did not attend this meeting.",
+        "- Roles and departments must not be inferred or changed.",
+        "- Discussion of a department does not establish responsibility.",
+        "- An action item may name a responsible person only when assignment or acceptance is explicit in the transcript.",
+        "- Objections, suggestions and expertise do not establish ownership.",
+        "",
+        "Meeting:",
+        f"- Title: {_text(meeting.get('title'))}",
+        f"- Language: {_text(meeting.get('language'))}",
+    ]
+
+    objective = _text(meeting.get("objective"))
+    if objective:
+        lines.append(f"- Objective: {objective}")
+
+    participants = _optional_list(data.get("participants"), "participants")
+    if participants:
+        lines.extend(["", "Actual participants:"])
+        for participant in participants:
+            lines.append(_render_person_line(participant, "participant_id", departments_by_id))
+
+    mentioned_people = _optional_list(data.get("mentioned_people"), "mentioned_people")
+    if mentioned_people:
+        lines.extend(["", "Mentioned but absent people:"])
+        for person in mentioned_people:
+            lines.append(_render_person_line(person, "person_id", departments_by_id))
+
+    abbreviations = _optional_mapping(organization.get("abbreviations"), "organization.abbreviations")
+    abbreviation_lines = [
+        f"- {key}: {_text(value)}"
+        for key, value in sorted(abbreviations.items())
+        if _text(value)
+    ]
+    if abbreviation_lines:
+        lines.extend(["", "Abbreviations:", *abbreviation_lines])
+
+    known_entities = _optional_mapping(data.get("known_entities"), "known_entities")
+    entity_lines = []
+    for key in sorted(known_entities):
+        values = [_text(value) for value in _optional_list(known_entities.get(key), key)]
+        values = [value for value in values if value]
+        if values:
+            entity_lines.append(f"- {key}: {', '.join(values)}")
+    if entity_lines:
+        lines.extend(["", "Relevant known entities:", *entity_lines])
+
+    context_rules = _optional_mapping(data.get("context_rules"), "context_rules")
+    rule_lines = [
+        f"- {key}: {str(value).lower() if isinstance(value, bool) else _text(value)}"
+        for key, value in sorted(context_rules.items())
+    ]
+    if rule_lines:
+        lines.extend(["", "Context rules:", *rule_lines])
+
+    return "\n".join(lines).strip() + "\n"
+
+
+def _render_person_line(
+    person: dict[str, Any],
+    id_key: str,
+    departments_by_id: dict[str, str],
+) -> str:
+    parts = [_text(person.get("display_name"))]
+    aliases = [_text(alias) for alias in _optional_list(person.get("aliases"), "aliases")]
+    aliases = [alias for alias in aliases if alias]
+    if aliases:
+        parts.append(f"aliases: {', '.join(aliases)}")
+
+    roles = _roles(person)
+    if roles:
+        parts.append(f"roles: {', '.join(roles)}")
+
+    department_id = _text(person.get("department_id") or person.get("department"))
+    if department_id:
+        department_name = departments_by_id.get(department_id, department_id)
+        parts.append(f"department: {department_name}")
+
+    identifier = _text(person.get(id_key))
+    if identifier:
+        parts.append(f"id: {identifier}")
+
+    return "- " + "; ".join(part for part in parts if part)
+
+
+def _roles(person: dict[str, Any]) -> list[str]:
+    roles = [_text(role) for role in _optional_list(person.get("roles"), "roles")]
+    role = _text(person.get("role"))
+    if role:
+        roles.insert(0, role)
+    return [role for role in roles if role]
+
+
+def _load_yaml(path: Path) -> Any:
+    text = path.read_text(encoding="utf-8-sig")
+    try:
+        import yaml  # type: ignore[import-not-found]
+    except ModuleNotFoundError:
+        return _parse_simple_yaml(text)
+    return yaml.safe_load(text)
+
+
+def _collect_department_ids(departments: list[Any]) -> set[str]:
+    ids: set[str] = set()
+    for index, department in enumerate(departments):
+        if not isinstance(department, dict):
+            raise MeetingContextValidationError(
+                f"organization.departments[{index}] must be an object."
+            )
+        department_id = _text(department.get("id"))
+        if not department_id:
+            raise MeetingContextValidationError(
+                f"organization.departments[{index}].id must be non-empty."
+            )
+        if department_id in ids:
+            raise MeetingContextValidationError(
+                f"Duplicate department id: {department_id!r}."
+            )
+        ids.add(department_id)
+    return ids
+
+
+def _collect_unique_ids(items: list[Any], key: str, path: str) -> set[str]:
+    ids: set[str] = set()
+    for index, item in enumerate(items):
+        if not isinstance(item, dict):
+            raise MeetingContextValidationError(f"{path}[{index}] must be an object.")
+        identifier = _text(item.get(key))
+        if not identifier:
+            raise MeetingContextValidationError(f"{path}[{index}].{key} must be non-empty.")
+        if identifier in ids:
+            raise MeetingContextValidationError(f"Duplicate {key}: {identifier!r}.")
+        ids.add(identifier)
+    return ids
+
+
+def _validate_attendance(item: dict[str, Any], path: str) -> None:
+    status = item.get("attendance_status")
+    if status not in VALID_ATTENDANCE_STATUSES:
+        raise MeetingContextValidationError(
+            f"{path}.attendance_status has invalid value: {status!r}."
+        )
+
+
+def _validate_department_reference(
+    item: dict[str, Any],
+    path: str,
+    department_ids: set[str],
+) -> None:
+    department_id = _text(item.get("department_id") or item.get("department"))
+    if department_id and department_id not in department_ids:
+        raise MeetingContextValidationError(
+            f"{path}.department_id references unknown department: {department_id!r}."
+        )
+
+
+def _require_mapping(data: dict[str, Any], key: str) -> dict[str, Any]:
+    value = data.get(key)
+    if not isinstance(value, dict):
+        raise MeetingContextValidationError(f"{key} must be an object.")
+    return value
+
+
+def _optional_mapping(value: Any, path: str) -> dict[str, Any]:
+    if value is None:
+        return {}
+    if not isinstance(value, dict):
+        raise MeetingContextValidationError(f"{path} must be an object.")
+    return value
+
+
+def _optional_list(value: Any, path: str) -> list[Any]:
+    if value is None:
+        return []
+    if not isinstance(value, list):
+        raise MeetingContextValidationError(f"{path} must be a list.")
+    return value
+
+
+def _require_non_empty_string(data: dict[str, Any], key_path: str) -> None:
+    key = key_path.split(".")[-1]
+    if not _text(data.get(key)):
+        raise MeetingContextValidationError(f"{key_path} must be present and non-empty.")
+
+
+def _text(value: Any) -> str:
+    if value is None:
+        return ""
+    return str(value).strip()
+
+
+def _parse_simple_yaml(text: str) -> Any:
+    lines = []
+    for raw_line in text.splitlines():
+        stripped = _strip_yaml_comment(raw_line.rstrip())
+        if stripped.strip():
+            lines.append((len(stripped) - len(stripped.lstrip(" ")), stripped.lstrip(" ")))
+    if not lines:
+        return None
+    parsed, index = _parse_yaml_block(lines, 0, lines[0][0])
+    if index != len(lines):
+        raise MeetingContextValidationError("Could not parse Meeting Context YAML.")
+    return parsed
+
+
+def _parse_yaml_block(
+    lines: list[tuple[int, str]],
+    index: int,
+    indent: int,
+) -> tuple[Any, int]:
+    if lines[index][1].startswith("- "):
+        result = []
+        while index < len(lines) and lines[index][0] == indent and lines[index][1].startswith("- "):
+            content = lines[index][1][2:].strip()
+            index += 1
+            if not content:
+                value, index = _parse_yaml_block(lines, index, lines[index][0])
+                result.append(value)
+                continue
+
+            if ":" in content:
+                key, raw_value = content.split(":", 1)
+                item = {key.strip(): _parse_scalar(raw_value.strip()) if raw_value.strip() else {}}
+                while index < len(lines) and lines[index][0] > indent:
+                    child_indent, child_content = lines[index]
+                    if child_content.startswith("- "):
+                        break
+                    child_key, child_raw_value = child_content.split(":", 1)
+                    index += 1
+                    child_raw_value = child_raw_value.strip()
+                    if child_raw_value:
+                        item[child_key.strip()] = _parse_scalar(child_raw_value)
+                    elif index < len(lines) and lines[index][0] > child_indent:
+                        item[child_key.strip()], index = _parse_yaml_block(lines, index, lines[index][0])
+                    else:
+                        item[child_key.strip()] = None
+                result.append(item)
+            else:
+                result.append(_parse_scalar(content))
+        return result, index
+
+    result = {}
+    while index < len(lines) and lines[index][0] == indent and not lines[index][1].startswith("- "):
+        key, raw_value = lines[index][1].split(":", 1)
+        index += 1
+        raw_value = raw_value.strip()
+        if raw_value:
+            result[key.strip()] = _parse_scalar(raw_value)
+        elif index < len(lines) and lines[index][0] > indent:
+            result[key.strip()], index = _parse_yaml_block(lines, index, lines[index][0])
+        else:
+            result[key.strip()] = None
+    return result, index
+
+
+def _parse_scalar(value: str) -> Any:
+    if value in {"null", "Null", "NULL", "~"}:
+        return None
+    if value in {"true", "True", "TRUE"}:
+        return True
+    if value in {"false", "False", "FALSE"}:
+        return False
+    if value in {"[]", "{}"} or (
+        value.startswith("[") and value.endswith("]")
+    ):
+        return ast.literal_eval(value)
+    if (
+        (value.startswith('"') and value.endswith('"'))
+        or (value.startswith("'") and value.endswith("'"))
+    ):
+        return ast.literal_eval(value)
+    return value
+
+
+def _strip_yaml_comment(line: str) -> str:
+    in_single = False
+    in_double = False
+    for index, char in enumerate(line):
+        if char == "'" and not in_double:
+            in_single = not in_single
+        elif char == '"' and not in_single:
+            in_double = not in_double
+        elif char == "#" and not in_single and not in_double:
+            return line[:index].rstrip()
+    return line

tests/gold/position_explicit_objection/README.md

@@ -0,0 +1,18 @@
+# position_explicit_objection
+
+Tests extraction of one explicit objection as a position.
+
+Scope:
+
+- Extract Tom's stated objection as exactly one position.
+- Do not create a todo for Tom.
+- Do not derive a decision from the objection.
+
+Exclusions:
+
+- This scenario does not test responsibility attribution for an open owner.
+- Responsibility attribution is covered by `responsibility_attribution_negative`.
+
+The transcript gives unique ground truth because Tom explicitly says "I object"
+and "My position is", then explicitly refuses responsibility. The group also
+states that no decision and no task for Tom were created.

tests/gold/position_explicit_objection/expected.json

@@ -0,0 +1,14 @@
+{
+  "facts": [],
+  "decisions": [],
+  "todos": [],
+  "questions": [],
+  "positions": [
+    {
+      "speaker": "Tom",
+      "position": "Tom objects to using one generic intake checklist because generic criteria will not work for analytics pilots and hardware trials.",
+      "evidence": "Tom: I object to that. My position is that generic criteria will not work for these two types of work."
+    }
+  ],
+  "technical": []
+}

tests/gold/position_explicit_objection/transcript.txt

@@ -0,0 +1,11 @@
+Iris: We could use one generic intake checklist for analytics pilots and hardware trials.
+
+Tom: I object to that. My position is that generic criteria will not work for these two types of work.
+
+Iris: Understood. Are you taking responsibility for rewriting the checklist?
+
+Tom: No. I am not taking that on, and I am not proposing an owner. I am only stating my position.
+
+Uma: Then we are not deciding the checklist today.
+
+Iris: Correct. No decision and no task for Tom.

tests/gold/responsibility_attribution_negative/README.md

@@ -7,9 +7,14 @@ The scenario includes a Marketing participant who comments critically on
 Business Development criteria. No one assigns that participant responsibility
 for defining the criteria, and the participant does not accept such a task.
 
-Expected behavior:
+Scope:
 
 - Do not assign Business Development criteria to the Marketing participant.
 - Do not reclassify the Marketing participant as Business Development.
-- Preserve the objection as a position.
 - Keep responsibility open unless explicitly assigned.
+- Preserve the agreed next step to ask Business Development for an owner.
+
+Exclusions:
+
+- This scenario does not test whether the objection is extracted as a position.
+- Position extraction is covered by `position_explicit_objection`.

tests/gold/responsibility_attribution_negative/expected.json

@@ -28,12 +28,6 @@
     }
   ],
   "questions": [],
-  "positions": [
-    {
-      "speaker": "Noah",
-      "position": "Noah says generic criteria will not work for different Marketing and product contexts.",
-      "evidence": "Noah: From Marketing, I can tell you that generic criteria will not work. A digital campaign and a physical product launch need different checks."
-    }
-  ],
+  "positions": [],
   "technical": []
 }

tests/test_extraction_protocol.py

@@ -5,12 +5,15 @@ from pathlib import Path
 
 from src.meeting_lab.extraction.extract_chunks import (
     EXTRACTION_CATEGORIES,
+    EXTRACTION_TASK_PROMPT_NAMES,
     build_prompt,
     extraction_path_for_chunk,
     normalize_current_schema,
     parse_json_response,
 )
+from src.meeting_lab.models.meeting_context import load_meeting_context
 from src.meeting_lab.protocol.build_protocol import build_protocol
+from scripts import run_gold_test
 
 
 class ExtractionProtocolTests(unittest.TestCase):
@@ -102,6 +105,41 @@ Final answer:
         self.assertIn("Extract each decision as one atomic commitment.", prompt)
         self.assertIn("Anna: Agreed.", prompt)
 
+    def test_build_prompt_includes_todo_prompt_file(self) -> None:
+        prompt = build_prompt("transcript.txt", "Nina: I will update it.")
+
+        self.assertEqual(EXTRACTION_TASK_PROMPT_NAMES, ("decisions.md", "todos.md"))
+        self.assertIn("Action-item responsibility rule:", prompt)
+        self.assertIn(
+            "A named responsible person may be extracted only when the transcript contains",
+            prompt,
+        )
+
+    def test_gold_runner_uses_shared_production_prompt_assembly(self) -> None:
+        self.assertIs(run_gold_test.build_prompt, build_prompt)
+
+    def test_no_context_and_meeting_context_prompts_use_same_task_prompt_set(self) -> None:
+        context = load_meeting_context(
+            Path("samples/real_live/project_process_meeting/meeting_context.yaml")
+        )
+
+        no_context_prompt = build_prompt("chunk_01_normalized.txt", "Anna: Agreed.")
+        context_prompt = build_prompt(
+            "chunk_01_normalized.txt",
+            "Anna: Agreed.",
+            meeting_context=context,
+        )
+
+        for prompt in (no_context_prompt, context_prompt):
+            self.assertIn("You extract decisions from meeting transcript text.", prompt)
+            self.assertIn("Action-item responsibility rule:", prompt)
+
+    def test_prompt_assembly_is_deterministic(self) -> None:
+        first = build_prompt("transcript.txt", "Nina: I will update it.")
+        second = build_prompt("transcript.txt", "Nina: I will update it.")
+
+        self.assertEqual(first, second)
+
     def test_build_protocol_groups_extraction_items(self) -> None:
         with tempfile.TemporaryDirectory() as directory:
             input_dir = Path(directory)

tests/test_meeting_context.py

@@ -0,0 +1,197 @@
+import copy
+import json
+import unittest
+from pathlib import Path
+from unittest.mock import patch
+
+from src.meeting_lab.extraction.extract_chunks import (
+    EXTRACTION_CATEGORIES,
+    build_prompt,
+    extract_chunk,
+)
+from src.meeting_lab.models.meeting_context import (
+    MeetingContextValidationError,
+    load_meeting_context,
+    render_meeting_context_for_prompt,
+    validate_meeting_context,
+)
+
+
+CONTEXT_PATH = Path("samples/real_live/project_process_meeting/meeting_context.yaml")
+SCRATCH_DIR = Path(".test-tmp")
+
+
+class MeetingContextTests(unittest.TestCase):
+    def setUp(self) -> None:
+        self.context = load_meeting_context(CONTEXT_PATH)
+
+    def tearDown(self) -> None:
+        if not SCRATCH_DIR.exists():
+            return
+        for path in SCRATCH_DIR.glob("chunk_0*_*.txt"):
+            path.unlink()
+        for path in SCRATCH_DIR.glob("chunk_0*_*.json"):
+            path.unlink()
+
+    def test_valid_context_loading(self) -> None:
+        self.assertEqual(self.context.schema_version, "1")
+        self.assertEqual(self.context.meeting_id, "2026-07-27-projektprozess")
+        self.assertEqual(self.context.data["meeting"]["language"], "de")
+
+    def test_duplicate_participant_ids_are_invalid(self) -> None:
+        data = copy.deepcopy(self.context.data)
+        data["participants"][1]["participant_id"] = data["participants"][0][
+            "participant_id"
+        ]
+
+        with self.assertRaisesRegex(MeetingContextValidationError, "Duplicate"):
+            validate_meeting_context(data)
+
+    def test_invalid_department_references_are_invalid(self) -> None:
+        data = copy.deepcopy(self.context.data)
+        data["participants"][0]["department_id"] = "unknown"
+
+        with self.assertRaisesRegex(MeetingContextValidationError, "unknown department"):
+            validate_meeting_context(data)
+
+    def test_participant_and_mentioned_person_id_collision_is_invalid(self) -> None:
+        data = copy.deepcopy(self.context.data)
+        data["mentioned_people"][0]["person_id"] = data["participants"][0][
+            "participant_id"
+        ]
+
+        with self.assertRaisesRegex(MeetingContextValidationError, "collide"):
+            validate_meeting_context(data)
+
+    def test_invalid_attendance_status_is_invalid(self) -> None:
+        data = copy.deepcopy(self.context.data)
+        data["participants"][0]["attendance_status"] = "remote"
+
+        with self.assertRaisesRegex(MeetingContextValidationError, "invalid value"):
+            validate_meeting_context(data)
+
+    def test_prompt_representation_is_deterministic(self) -> None:
+        first = render_meeting_context_for_prompt(self.context)
+        second = render_meeting_context_for_prompt(self.context)
+
+        self.assertEqual(first, second)
+        self.assertIn("MEETING CONTEXT V1", first)
+        self.assertIn("- Language: de", first)
+
+    def test_authoritative_rules_appear_in_prompt(self) -> None:
+        prompt = build_prompt(
+            "chunk_01_normalized.txt",
+            "Martin: Wir besprechen Marketing.",
+            meeting_context=self.context,
+        )
+
+        self.assertIn("The participant list is authoritative.", prompt)
+        self.assertIn("Mentioned people did not attend this meeting.", prompt)
+        self.assertIn("Roles and departments must not be inferred or changed.", prompt)
+        self.assertIn("Discussion of a department does not establish responsibility.", prompt)
+        self.assertIn(
+            "An action item may name a responsible person only when assignment or acceptance is explicit",
+            prompt,
+        )
+        self.assertIn("Objections, suggestions and expertise do not establish ownership.", prompt)
+
+    def test_extraction_behavior_is_unchanged_without_context(self) -> None:
+        response = json.dumps(
+            {
+                "facts": [],
+                "decisions": [],
+                "todos": [],
+                "open_questions": [],
+                "positions": [],
+                "technical_details": [],
+            }
+        )
+
+        SCRATCH_DIR.mkdir(exist_ok=True)
+        chunk_path = SCRATCH_DIR / "chunk_01_normalized.txt"
+        output_path = SCRATCH_DIR / "chunk_01_extraction.json"
+        chunk_path.write_text("Anna: Keine Entscheidung.", encoding="utf-8")
+
+        captured_prompts = []
+
+        def fake_call_ollama(**kwargs):
+            captured_prompts.append(kwargs["prompt"])
+            return response, {}
+
+        with patch(
+            "src.meeting_lab.extraction.extract_chunks.call_ollama",
+            side_effect=fake_call_ollama,
+        ):
+            extraction = extract_chunk(
+                chunk_path,
+                output_path,
+                model="test",
+                endpoint="http://example.invalid",
+                timeout=1,
+                temperature=0.0,
+                num_predict=None,
+                num_ctx=None,
+            )
+
+        self.assertEqual(set(extraction), set(EXTRACTION_CATEGORIES))
+        self.assertNotIn("context", extraction)
+        self.assertNotIn("MEETING CONTEXT V1", captured_prompts[0])
+
+    def test_extraction_result_records_meeting_context_provenance(self) -> None:
+        response = json.dumps(
+            {
+                "facts": [],
+                "decisions": [],
+                "todos": [],
+                "open_questions": [],
+                "positions": [],
+                "technical_details": [],
+            }
+        )
+
+        SCRATCH_DIR.mkdir(exist_ok=True)
+        chunk_path = SCRATCH_DIR / "chunk_02_normalized.txt"
+        output_path = SCRATCH_DIR / "chunk_02_extraction.json"
+        chunk_path.write_text("Martin: Hallo.", encoding="utf-8")
+
+        with patch(
+            "src.meeting_lab.extraction.extract_chunks.call_ollama",
+            return_value=(response, {}),
+        ):
+            extraction = extract_chunk(
+                chunk_path,
+                output_path,
+                model="test",
+                endpoint="http://example.invalid",
+                timeout=1,
+                temperature=0.0,
+                num_predict=None,
+                num_ctx=None,
+                meeting_context=self.context,
+            )
+
+        written = json.loads(output_path.read_text(encoding="utf-8"))
+
+        self.assertEqual(
+            extraction["context"]["meeting_id"],
+            "2026-07-27-projektprozess",
+        )
+        self.assertEqual(extraction["context"]["schema_version"], "1")
+        self.assertEqual(written["context"], extraction["context"])
+        self.assertNotIn("participants", written["context"])
+
+    def test_real_context_keeps_metadata_separate_from_assignment(self) -> None:
+        prompt_context = render_meeting_context_for_prompt(self.context)
+
+        self.assertIn("Björn", prompt_context)
+        self.assertIn("department: Marketing", prompt_context)
+        self.assertIn("Mentioned but absent people:", prompt_context)
+        self.assertIn("Jovana", prompt_context)
+        self.assertIn("Discussion of a department does not establish responsibility.", prompt_context)
+        self.assertIn("assignment or acceptance is explicit", prompt_context)
+        self.assertNotIn("responsible: Björn", prompt_context)
+        self.assertNotIn("responsible: Jovana", prompt_context)
+
+
+if __name__ == "__main__":
+    unittest.main()