docs: remove CLI recipes and rigid instructions in favor of clean architecture spec

README.md

@@ -25,7 +25,7 @@ tags:
 
 <h1 align="center">Protium-1.0: Objective-Gated Cognitive Architecture</h1>
 
-<h3 align="center">Zero-Waste Cognitive Protocol &bull; Dynamic Cognitive Impedance &bull; Universal LLM Runtime Support</h3>
+<h3 align="center">Zero-Waste Cognitive Protocol &bull; Dynamic Cognitive Impedance &bull; Universal Cross-Family Architecture</h3>
 
 <p align="center">
   <img src="https://img.shields.io/badge/org-Solstice--AI-blueviolet" alt="Solstice-AI">
@@ -42,28 +42,19 @@ tags:
 
 **Protium-1.0** is Solstice-AI's high-efficiency, objective-gated cognitive reasoning architecture for open-weight foundation models. Named after **Protium** ($^1\text{H}$) — the fundamental isotope of hydrogen featuring zero neutrons and zero dead weight — Protium-1.0 replaces traditional, unconstrained chain-of-thought bloat with a lean, forward-only reasoning pipeline.
 
-Traditional reasoning chat templates frequently impose static token minimums, rigid pseudo-schemas, or unconstrained internal deliberation loops. Protium-1.0 resolves this fundamental inefficiency by establishing an **Objective-Gated Protocol**: every reasoning sequence anchors its verification criteria at step one, reasons strictly forward toward that objective, and terminates deliberation the exact millisecond verified correctness is attained.
+Rather than imposing static token minimums, rigid pseudo-schemas, or unconstrained internal deliberation loops, Protium-1.0 operates on an **Objective-Gated Protocol**: every reasoning sequence anchors its verification criteria at step one, reasons strictly forward toward that objective, and terminates deliberation the moment verified correctness is attained.
 
-Equipped with the **`[L10]` Proteus Dynamic Engine** as its native default, Protium-1.0 auto-calibrates cognitive depth to the intrinsic difficulty of each prompt—collapsing to near-zero latency on direct queries while dynamically allocating full analytical runway for axiomatic proofs and complex engineering tasks.
+Equipped with the **`[L10]` Proteus Dynamic Engine** as its native default, Protium-1.0 auto-calibrates cognitive depth to the intrinsic difficulty of each prompt—minimizing overhead on direct queries while dynamically allocating full analytical runway for axiomatic proofs and complex engineering tasks.
 
 ---
 
 ## Architectural Principles
 
-### 1. Token 1 Target Anchoring
-At the inception of the cognitive trajectory, the model isolates and locks the exact termination criteria required for a complete, verified solution. The target is anchored before execution begins.
-
-### 2. Forward-Only Deduction
-Reasoning progresses strictly forward along deductive axes. Internal conversational stalling, rhetorical hesitation, and self-conversational commentary (*"Let me see"*, *"Hmm"*, *"Wait"*) are systematically eliminated from the attention cache.
-
-### 3. Objective-Gated Early Exit
-Deliberation is bounded exclusively by mathematical, programmatic, and empirical verification. The moment the target anchor is verified, the `<think>` block terminates immediately and the final response is delivered.
-
-### 4. Delta-Isolated Error Correction
-If a logical contradiction or boundary failure is detected during verification, the model isolates and repairs only the specific broken parameter. It does not regress into circular overthinking loops or full-context restarts.
-
-### 5. Decisive Operational Aura
-Outputs are delivered with maximum information density, absolute resolve, and complete absence of conversational preamble or unnecessary hedging.
+* **Token 1 Target Anchoring**: At the inception of the cognitive trajectory, the model isolates and locks the exact termination criteria required for a complete, verified solution.
+* **Forward-Only Deduction**: Reasoning progresses strictly forward along deductive axes, eliminating internal conversational stalling and rhetorical hesitation from the attention cache.
+* **Objective-Gated Early Exit**: Deliberation is bounded exclusively by mathematical, programmatic, and empirical verification. The moment the target anchor is verified, thinking terminates immediately and the final response is emitted.
+* **Delta-Isolated Error Correction**: If a logical contradiction or boundary failure is detected during verification, the model isolates and repairs only the specific broken parameter, preventing circular loops or full-context restarts.
+* **Decisive Operational Delivery**: Outputs are delivered with high information density, clear technical rigor, and zero conversational filler.
 
 ---
 
@@ -89,91 +80,33 @@ Protium-1.0 features 11 calibrated reasoning tiers plus specialized multi-agent
 
 ---
 
-## Runtime Integration & Usage
-
-### 1. llama.cpp / LM Studio
-Supply any family template directly to `llama-cli` or `llama-server`:
-```bash
-# Default Proteus dynamic reasoning:
-llama-cli -m model.gguf --chat-template-file templates/qwen.jinja -p "Implement a lock-free ring buffer in Rust"
-
-# Explicit level steering:
-llama-cli -m model.gguf --chat-template-file templates/qwen.jinja -p "[L4] Implement a lock-free ring buffer in Rust"
-```
-
-### 2. Ollama Modelfile
-```dockerfile
-FROM ./model.gguf
-TEMPLATE """{{ read_file "templates/universal.jinja" }}"""
-PARAMETER stop "<|im_start|>"
-PARAMETER stop "<|im_end|>"
-```
-
-### 3. Transformers (Python)
-```python
-from transformers import AutoTokenizer
-
-tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-32B-Instruct")
-with open("templates/qwen.jinja") as f:
-    tokenizer.chat_template = f.read()
-
-# Dynamic default execution:
-prompt = tokenizer.apply_chat_template(
-    [{"role": "user", "content": "Prove that every planar graph is 5-colorable."}],
-    tokenize=False,
-    add_generation_prompt=True
-)
-
-# Explicit tier selection via kwarg:
-prompt_athena = tokenizer.apply_chat_template(
-    [{"role": "user", "content": "Refactor this database connection pool"}],
-    reasoning_effort="athena",
-    tokenize=False,
-    add_generation_prompt=True
-)
-```
-
-### 4. vLLM / SGLang / Open WebUI
-Pass standard OpenAI-compatible parameters in the payload:
-```json
-{
-  "model": "Qwen3.8-27B-Protium",
-  "messages": [{"role": "user", "content": "Design an async event bus"}],
-  "reasoning_effort": "proteus"
-}
-```
+## Steering & Control Reference
 
----
+Protium-1.0 supports flexible steering via inline tags or standard inference parameters:
 
-## Steering & Control Syntax
-
-Protium-1.0 supports instant mid-session steering via inline tags or standard API kwargs:
-
-* **Inline Steering Tags** (case-insensitive, cleanly stripped prior to model consumption):
-  * `[L10]` or `{REASON:proteus}`: Dynamic unconstrained adaptive reasoning (Default).
-  * `[L4]` or `{REASON:athena}`: Fixed Athena strategic architecture.
-  * `[L8]` or `{REASON:einstein}`: Fixed Einstein theoretical reduction.
-  * `[regular]` or `[vanilla]`: Pure vanilla chat template with zero system directives or thinking tokens.
-  * `{REASON:i8}`: Silent execution (applies Level 8 analytical rigor internally while suppressing `<think>` emission).
-* **Runtime API Kwargs**:
-  * `reasoning_effort="10"|"proteus"|"dynamic"|"auto"`
-  * `reasoning_effort="1".."9"` or named levels (`hermes`, `apollo`, `artemis`, `athena`, `prometheus`, `solstice`, `hyperion`, `einstein`, `oracle`, `swarm`, `council`)
-  * `regular=True` or `mode="vanilla"` (zero-overhead execution)
+| Control Target | Inline Syntax | API Kwarg / Parameter |
+|---|---|---|
+| **Dynamic Adaptive (Default)** | `[L10]` or `{REASON:proteus}` | `reasoning_effort="proteus"` or `"dynamic"` |
+| **Specific Cognitive Levels** | `[L1]` through `[L9]` | `reasoning_effort="1"` .. `"9"` (or level name) |
+| **Silent Reasoning** | `{REASON:i8}` | Runs high-tier analytical rigor with `<think>` hidden |
+| **Zero-Overhead Vanilla** | `[regular]` or `[vanilla]` | `regular=True` or `mode="vanilla"` |
 
 ---
 
-## Cross-Family Architecture Matrix
+## Template Family Distribution
+
+All templates are located in the `templates/` directory and pre-configured for native runtime token formats:
 
-| Model Family | Template Path | Special Token Implementation |
+| Target Model Architecture | File | Native Format |
 |---|---|---|
 | **Universal / ChatML** | `templates/universal.jinja` | `<|im_start|>`, `<|im_end|>` |
-| **Qwen 2.5 / 3.8 / Omni** | `templates/qwen.jinja` | `<|im_start|>`, `<|im_end|>`, Spatial/Temporal Vision |
-| **DeepSeek Native** | `templates/deepseek.jinja` | `<｜User｜>`, `<｜Assistant｜>`, Native DSML Tool Format |
+| **Qwen 2.5 / 3.8 / Vision** | `templates/qwen.jinja` | `<|im_start|>`, `<|im_end|>`, Multimodal Vision |
+| **DeepSeek Native** | `templates/deepseek.jinja` | `<｜User｜>`, `<｜Assistant｜>`, Native DSML |
 | **Meta Llama 3 / 4** | `templates/llama3.jinja` | `<|start_header_id|>`, `<|eot_id|>` |
 | **Mistral / Mixtral** | `templates/mistral.jinja` | `[INST]`, `[/INST]` |
 | **Google Gemma 2 / 3** | `templates/gemma.jinja` | `<start_of_turn>`, `<end_of_turn>` |
 | **Zhipu GLM 4 / 5.3** | `templates/glm.jinja` | `[gMASK]<sop>`, `<|user|>`, `<|assistant|>` |
-| **Vanilla Standalone** | `templates/regular_*.jinja` | Zero-overhead clean reference templates |
+| **Vanilla Reference** | `templates/regular_*.jinja` | Minimal, zero-overhead baseline templates |
 
 ---