- Finalized MVP module selection (blue outline) based on system flow chart

- Marked core (internal) modules with **magenta star** and Docker modules with **Docker icon**

- Deferred post-MVP modules marked with **orange outline**

- ✅ Core (Magenta Star): `session_manager`, `input_broker`, `confidence_evaluator`, `memory_integrator`,

  `output_expander`, `output_planner`, `partner_ui`, `schema_recorder`

- 🐳 Dockerized: `audio_feed_worker`, `camera_feed_worker`, `speech_transcriber`, `landmark_extractor`,

  `gesture_classifier`, `sound_classifier`, `speech_context_classifier`, `visual_environment_classifier`,

  `model_trainer`, `sound_playback`, `landmark_visualizer`

- `llm_clarifier`, `feedback_log`, `retraining_scheduler`, `memory_interface`

- `model_registry` and `recorded_schemas` to remain internal on Hetzner, accessed via local endpoints

- Defined canonical `A3CPMessage` schema in `schemas/a3cp_core/a3cp_message.py` based on Section 2 of `SCHEMA_REFERENCE.md`

  - Includes required core metadata fields: `schema_version`, `record_id`, `user_id`, `session_id`, `timestamp`, `modality`, `source`

  - Supports forward-compatible extension via `extra = "allow"`

- Added schema validation utility in `schemas/utils/validate.py`

  - Function `validate_a3cp_message(data: dict)` raises on malformed messages

- Added unit test: `tests/schemas/test_a3cp_message.py`

  - Validates correct messages pass

  - Ensures missing or invalid fields are rejected

[Schema] Added Section 3.4: Classifier Output Components

- Introduced optional `classifier_output_components` field to support multimodal intent tracing.

- Allows downstream modules to inspect per-modality predictions (e.g., gesture, sound, speech).

- Each entry includes `intent`, `confidence`, and optional `timestamp`.

- Documented structure in Section 3.4 of the A3CPMessage specification.

Reason: Enables transparent, auditable contribution tracking when multiple classifiers inform a single semantic intent.