Papers, reports and reproducible evidence
Explore what TrueOpen has studied, what the experiments measured, and what remains outside their guarantees.
Original preprints
Technical reports and protocol manuscripts
- Worker-Verifier Model Identity Verification with Logprobs and Routed Experts: An Experimental Study
What can probability traces tell us about the model behind an inference response?
- SLP Verifiable Inference: Proof-of-Concept Experiment Report
A version-specific record of sampled proof experiments and their coverage limits.
- Worker vs. Verifier Latency Benchmark Report
Measure generation and teacher-forced prefill separately before estimating verification overhead.
- Logprob Majority-Consensus Verification (LMCV): A Low-Cost Execution-Verification Protocol for Open Inference Networks
Combine a fixed-token probability replay with an independent-verifier protocol, while keeping the assumptions explicit.
Historical experiment summary
7B verification cost experiment — preserved EXP-02 summary, with reported configurations and cost ratios. Complete logs, absolute timings and repeat counts are not available in the current website files. This is a historical summary, not a new paper.
Experiment artifacts and data
These ZIP bundles contain published report files and result tables, not complete raw execution logs. Use the experiment repositories for available code, logs and reproduction instructions.
LMCV experiments · SLP experiments
Hugging Face datasets
TrueOpenAI/lmcv-experiments hosts the LMCV manuscript, experimental reports, tooling and data archives for logprob comparison, MoE routed experts and worker-verifier latency. Read the card and individual reports for configurations and evidence limits. Read the project explanation.
Models
No standalone TrueOpen model release is listed here. Qwen, TinyLlama and Llama models discussed in the experiments are third-party models, not models published by TrueOpen.
Related work by other researchers
External research bibliography — these publications provide context and are not TrueOpen publications.
Research reports are not independent security audits. Experimental measurements, protocol designs and production deployment are separate evidence states.