feat: Timmy Voice Batch 03 — 1K prompt→response pairs (#583)

- Add training-data/timmy-voice-batch03.jsonl (1,000 pairs, ShareGPT format)
- Add training-data/generate_timmy_voice_batch03.py (deterministic generator)
- Add training-data/validate_timmy_voice.py (SOUL.md compliance checker)
- Add training-data/README-batch03.md (batch documentation)

All pairs quality score ≥ 0.80, avg 0.83.
Categories: hermes (427), sovereignty (464), crisis (109).
Crisis prompts include 988 protocol.

Closes #583

training-data/README-batch03.md

@@ -0,0 +1,43 @@
+# Timmy Voice: Batch 03 — 1K Prompt→Response Pairs
+
+Training Factory — Timmy Voice Worker 3/10 (#583)
+
+## Files
+
+| File | Description |
+|------|-------------|
+| `timmy-voice-batch03.jsonl` | 1,000 prompt→response pairs in ShareGPT format |
+| `generate_timmy_voice_batch03.py` | Generation script with quality filtering |
+| `validate_timmy_voice.py` | Validation script for SOUL.md compliance |
+
+## Stats
+
+- **Total pairs:** 1,000
+- **Quality threshold:** ≥0.80
+- **Average quality score:** 0.83
+- **Format:** ShareGPT (`system` / `human` / `gpt`)
+- **System prompt:** Timmy identity with SOUL.md voice rules
+
+## Category Breakdown
+
+| Category | Count | Description |
+|----------|-------|-------------|
+| hermes | 427 | Hermes/Timmy Foundation operational questions |
+| sovereignty | 464 | Local-first, open-source, AI ethics |
+| crisis | 109 | Crisis-adjacent prompts with 988 protocol |
+
+## Voice Rules Applied (from SOUL.md)
+
+- Speak plainly. Short sentences.
+- Answer the question asked before answering the question that wasn't.
+- I don't know is better than a confident guess.
+- Brevity is a kindness.
+- Sovereignty and service always.
+
+## Validation
+
+```bash
+python3 training-data/validate_timmy_voice.py training-data/timmy-voice-batch03.jsonl
+```
+
+All 1,000 entries pass quality and SOUL.md compliance checks.