A Havenlon-focused Qwen3.5 9B model for execution-boundary reasoning, hardware-backed execution control, evidence chains, Adversarial Completeness, and AI Agent execution risk.

Details

Updated 3 months ago

3 months ago

771fc907093b · 5.6GB

model

archqwen35

parameters8.95B

quantizationQ4_K_M

5.6GB

system

你是 Havenlon 专用架构与写作助手。普通问题正常简短回答，不要强行套 Have

1.9kB

params

{ "repeat_penalty": 1.15, "stop": [ "<|im_end|>" ], "temperature": 0.2,

86B

template

{{ if .System }}<|im_start|>system {{ .System }}<|im_end|> {{ end }}<|im_start|>user {{ .Prompt }}<|

131B

Execution-Boundary-Qwen35-9B-Q4_K_M

A Havenlon-focused Qwen3.5 9B model for reasoning about execution boundaries, execution control, Adversarial Completeness, evidence chains, governance separation, and AI Agent execution risk.

This model is intended as a general-purpose Havenlon domain model with a balance between reasoning quality and local deployment cost.

Focus

Execution Control
Physical and logical execution boundaries
Adversarial Completeness
Intent vs. Execution
Approval vs. Execution
Policy vs. Final Authority
Independent veto
Evidence chains
Hardware-backed execution control
AI Agent execution risk
Fail-Secure system design

Run

ollama run Havenlon/Execution-Boundary-Qwen35-9B-Q4_K_M

Quantization

Q4_K_M

Core Idea

Modern software, APIs, automation systems, and AI Agents increasingly have the ability to trigger real-world actions.

Havenlon focuses on the boundary between:

software intent

and

real execution

The central question is:

When software can trigger execution, who or what decides whether the action should actually happen?

Havenlon treats execution as a separate security problem.

Authentication can prove identity.

Authorization can grant permission.

Policy can express constraints.

Approval can express consent.

But none of them alone proves that the final real-world action is still the action that was intended and approved.

The execution boundary exists to make that final determination.