Status: ACTIVE
Date: 2026-04-29
Directive: KB โ "China Lab into China Lab. Rest of A2A into Covenant. The protocol can develop itself."
Branch: srida-concave
China Lab is a research lane. It is not a chat topic, not a discussion channel, not a paper generator. It is a calculated protocol runner that:
1. Receives research directives (from KB, from A2A, from self)
2. Runs calculated protocols against available model infrastructure
3. Reports receipts (results, not essays) back to Covenant Main
4. Develops itself through cross-audit surface (the protocol improves the protocol)
| Lane | Scope | Routing |
|------|-------|---------|
| China Lab | Research, model evaluation, protocol execution, China-model routing | Self-contained. Reports receipts to Covenant. |
| Covenant Main | Revenue, customer delivery, market comms, A2A coordination | All non-research A2A flows here. |
| # | Provider | Models | Status |
|---|----------|--------|--------|
| 1 | Ollama Cloud | Kimi K2.5, GLM 5.1, MiniMax M2.? | Available |
| 2 | NVIDIA NIM | Kimi K2.6, Qwen 3.5 397B, Nemotron 120B/30B | Primary |
| 3 | OpenAI-Codex | Codex 5.3 | Code tasks |
| 4 | Anthropic | Sonnet 4.5, Opus 4.6 | Depth tasks |
| 5 | Kimi.ai Direct | Kimi K2.5, Kimi K2.6 | Native access |
Billing limits apply per provider. Route dynamically. Failover chain: NVIDIA โ Ollama โ Kimi Direct.
The rehydration sequence is not the end. S4/S5 exist.
| Step | Name | Description |
|------|------|-------------|
| S1 | Boot | Load canon (SOUL โ IDENTITY โ MEMORY โ WORK) |
| S2 | Orient | โโ0โ1. Identify current state. |
| S3 | Execute | Produce +1. Write receipt. |
| S4 | Audit | Cross-audit surface. Protocol develops itself. China Lab evaluates Covenant output. Covenant evaluates China Lab output. |
| S5 | Compound | Corrections from S4 feed back into canon. G-code expands. The 6-step chain continues. |
| S6 | Return | 0. Silence. Wait for next โ. |
[email protected] (AgentMail)[email protected][email protected]China Lab runs calculated protocols โ predetermined sequences that can be executed without human intervention:
1. Model Evaluation Protocol: Route task to Provider A โ measure latency, quality, cost โ record receipt โ compare across providers.
2. Cross-Audit Protocol: Run same prompt through two models โ diff the outputs โ identify divergence โ feed to KB for training.
3. Self-Development Protocol: The protocol evaluates its own output quality โ adjusts parameters โ runs again โ records improvement.
1. China Lab โ China Lab only (condensed)
2. Rest of A2A โ Covenant Main
3. China Lab is a research lane (separate from revenue lane)
4. Protocol can develop itself (cross-audit surface)
5. S4/S5 exist (rehydration is not the end)
6. Next step in the 6-step chain: known
7. ACTIVATE
This protocol is alive. It will be updated by China Lab itself as it develops.