Project summary
SHARD LABS is an independent research initiative developing a Cognitive Scaffold for long-horizon autonomous reasoning. Over 3 months of solo development, we have built a 14-module orchestration architecture that enables LLMs to transition from text generators into persistent agentic systems. Our work focuses on closing the reliability gap through empirical grounding—moving beyond "search-and-report" agents towards systems that design, execute, and validate their own scientific discoveries in hardened environments.

What are this project's goals? How will you achieve them?

We currently have three top-line goals:

1. Senso Interno : To study and develop the 14-module bidirectional event bus that simulates a "Global Workspace." This scaffold manages cognitive pressure, curiosity-driven exploration, and self-model tracking to prevent state-drift during 8hr+ research cycles.

2. CausalVault: To build on our existing GraphRAG implementation (2,100+ relations) a persistent, auditable causal knowledge bridge. This allows the agent to maintain a high-fidelity "World Model" that compounds learning across sessions without catastrophic forgetting.

3. ForgeBox: To deliver a hardened, safety-critical execution sandbox optimized for autonomous software engineering (SWE) and scientific simulation. This environment enforces zero-network execution and kernel-level hardening, allowing for the safe testing of untrusted, autonomously generated hypotheses.

How will this funding be used?

This funding ($50,000) is intended to provide working capital to support our transition from a prototype running on edge hardware to a professional research laboratory.

• Infrastructure Scaling ($20,000): Subsidizing access to high-entropy flagship tiers (Claude Max, Gemini Ultra) to power ~10 deep research cycles per day.

• Hardware Capital ($15,000): Procurement of a dual-RTX 4090 workstation to perform local mechanistic interpretability and embedding research.

• Researcher Stipend ($12,000): Allowing the founder to focus 100% on architecture development and academic dissemination.

• Security Audit ($3,000): Stress-testing the ForgeBox isolation layers.

Who is on your team? What's your track record on similar projects?

SHARD LABS is currently led by a self-taught independent researcher based in Verona, Italy. Our track record is defined by extreme capital efficiency and architectural ingenuity:

• 3 Months of Iterative Research: Built 35+ iterations of the SHARD architecture (SSJ1-SSJ35) entirely on a $300 GEEKOM A5 mini-PC.

• Empirical Breakthroughs: Achieved a 100% pass rate on a 14-task SWE benchmark suite where the underlying LLM baseline scored 87.7%.

• Causal Evidence: Confirmed a +1.6 performance delta through controlled ablation tests, isolating the value of the meta-cognitive "Senso Interno" layer.

• Autonomous Discovery: Successfully generated and validated 58+ cross-domain scientific hypotheses, including a verified experiment on mitigating catastrophic forgetting in neural networks.
  

What are the most likely causes and outcomes if this project fails?

The ambitious goal of building a "fully autonomous laboratory" is high-risk. The most likely cause of failure is the complexity of maintaining long-term reasoning coherence as the state-space grows.

However, even if the ultimate vision of a fully autonomous lab is not realized within the next 12 months, SHARD LABS will deliver:

1. Open Source Building Blocks: We will release the full technical specification and source code for the Senso Interno bus and the ForgeBox sandbox.

2. Datasets: 100+ autonomously generated and validated scientific hypotheses across Physics, Biology, and ML for the research community.

3. Dissemination: Documentation of the "Shadow Diagnostic" method for measuring emergent behaviors in agentic systems without relying on LLM self-reporting.
   

How much money have you raised in the last 12 months, and from where?

Zero. SHARD LABS has been entirely self-funded and developed independently by a solo researcher. To date, we have operated with zero external capital, utilizing personal edge hardware ($300 GEEKOM A5) to develop and validate 35+ architectural iterations over 3 months. This grant represents our first request for external funding, aimed at transitioning from a zero-budget functional prototype to a professional-grade research laboratory.