You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$15,000 left to reach the funding goal.
TriplePass Labs is developing Prove, an open-source evaluation protocol and CLI that cryptographically seals benchmark evaluation runs into Bitcoin block headers using OpenTimestamps. Today, benchmark leaderboards suffer from silent model drift, prompt over-tuning, and retroactive cherry-picking. Prove provides verifiable cryptographic receipts proving that a specific evaluation was executed against an exact model version at a verified time without retroactive tampering. Alongside Prove, we operate Colony Zero, an empirical research testbed evaluating whether swarms of smaller, equal-compute models (e.g. 3x 8B or 4x 14B) can match or outperform monolithic frontier models under strict compute and token parity.
1. Ship Prove v1.0 CLI: Implement automated cryptographic manifest pre-registration (prove pre-register) and Bitcoin OpenTimestamps attestation (prove seal --anchor ots). Once an evaluation finishes, the execution merkle tree is sealed directly into Bitcoin block headers. Downstream auditors can verify receipts offline in milliseconds without trusting TriplePass Labs or any third party.
2. Launch the Public Agent Drift Observatory: Deploy an automated canary testbed running continuous daily and weekly evaluation sweeps across 10+ frontier APIs (Claude 3.5 Sonnet, GPT-4o, Gemini 1.5 Pro) and open-weight models (Llama 3, Qwen 2.5) to track and publish empirical capability drift.
3. Advance Colony Zero Swarm Benchmarks: Expand our empirical benchmark dataset comparing multi-agent swarms against monolithic reasoning systems under strict token and compute equality, publishing all proofs publicly.
- Metric 1: Public release of the prove Python package on PyPI with complete OpenTimestamps anchoring and verification commands, passing automated test suites with 100% offline proof verification.
- Metric 2: Live deployment of the Agent Drift Observatory web dashboard indexing at least 100 sealed evaluation runs anchored to Bitcoin block headers with verifiable merkle proofs.
- Metric 3: Release of a reproducible Colony Zero swarm benchmark comparative report with pre-registered manifests and open Hugging Face dataset releases.
- Metric 4: Adoption of Prove attestation by at least 3 independent AI safety researchers or external benchmark maintainers.
We are requesting $15,000 to fund a six-month execution sprint (with a minimum viable milestone unblocked at $5,000):
- Dedicated Local Evaluation Rig ($4,500): Dedicated 24/7 evaluation workstation with local high-VRAM GPU hardware for deterministic local open-weight model runs without metered runtime constraints.
- High-Speed NVMe Storage & Dedicated Bitcoin Node ($2,000): High-IOPS scratch drives for raw multi-agent execution traces and an always-on Bitcoin block header verification node.
- Multi-Model Cloud & API Inference ($4,500): Six-month inference budget across frontier model endpoints (Claude 3.5 Sonnet, GPT-4o, Gemini) and decentralized open-weight GPU endpoints for continuous drift monitoring.
- Public Drift Observatory & Web Hosting ($2,000): Cloud hosting for the public dashboard, proof verification explorer, and telemetry database.
- Educational Research Media ($2,000): Audio, video, and open technical documentation shared publicly on YouTube to educate researchers on verifiable evaluation pipelines.
Christopher Ayers is the solo founder of TriplePass Labs LLC in Puyallup, Washington. He is an analytics systems architect with fifteen years of experience designing high-scale data and operational pipelines in complex enterprise environments. Over the past year, he focused on AI systems integration, cryptographic attestation, and empirical multi-agent evaluation. He has shipped the core architecture for Prove and deployed an interactive Colony Zero explorer on Hugging Face Spaces. He works on TriplePass Labs 8 to 12 hours weekly on nights and weekends alongside a full-time analytics day job.
- Cause 1: Lack of community adoption. If benchmark authors find cryptographic pre-registration adds workflow friction, Prove might become a niche tool. Mitigation: We are designing Prove with zero-friction CLI wrappers that integrate directly into existing harnesses (e.g. Inspect AI, lm-evaluation-harness) with a single flag (--anchor ots).
- Cause 2: Hardware or API cost overruns. If API pricing spikes or local testing bottlenecks occur, continuous drift sweeps could become throttled. Mitigation: The testbed prioritizes local open-weight models and caps API polling intervals with exponential backoff on stable benchmarks.
- Outcome if it fails: Even if broad industry adoption is slow, the complete protocol, OpenTimestamps verification libraries, and Colony Zero swarm evaluation dataset will be fully open-sourced under Apache 2.0 / MIT licenses, providing permanent value to open-source AI safety researchers.
TriplePass Labs LLC is a newly formed (October 2026), 100% self-funded solo lab. Total outside funding raised to date is $0. The lab has zero venture capital, zero debt, and zero dilutive equity. We have pending grant applications under active review with non-dilutive programs including Emergent Ventures, Solo Grants, and SFF Speculation Grants, but have received no outside disbursements to date.
There are no bids on this project.