The September 17, 2026 report tests a seal-then-sample pipeline on GPT-2, TinyLlama-1.1B and Llama-2-70B. Commitments are fixed before the challenge; selected chunks are proved and checked. This page summarizes that report version, not every later paper revision.
For Llama-2-70B on a CPU server with 2 TB memory, the report records 163 sealed boundaries, five proved chunks, 1,259 seconds (about 21 minutes) for sealing plus proving, a 4.34 MiB proof and 46.3 seconds of verification. This excludes model registration and the separately reported 48,706-second integer-inference stage; it is not end-to-end latency or a production promise.
The report proves selected chunks of a 12-bit quantized canonical model, not the original floating-point serving model. Its 70B and TinyLlama proving experiments use context lengths of 8–16; production-context proving costs were not measured. Most reported figures are single runs without repeated-run variance (except E09 and E12). These limits matter when comparing the prototype with an inference service.
For a forgery confined to one eligible chunk fixed before the challenge, detection by a single audit is limited by sampling coverage. The report gives about 1.9% single-audit detection probability (3 randomly sampled chunks out of 161 eligible chunks, alongside 2 mandatory anchors) for one fixed invalid chunk in the 70B configuration and demonstrates challenge grinding with manifest-only sampling. External ordering and explicit threat assumptions are needed.
Original-source bibliography and citations
This page is an editorial overview, not the original paper. Cite the specific report or paper below for its findings. Bibliographic sources checked November 5, 2026.
SLP Verifiable Inference: Proof-of-Concept Experiment Report
TrueOpen.ai team
Report v1.0 · 2026-09-17
@techreport{trueopen_slp_report_2026,
author = {{TrueOpen.ai team}},
title = {SLP Verifiable Inference: Proof-of-Concept Experiment Report},
institution = {TrueOpen},
year = {2026},
month = sep,
number = {Version 1.0},
url = {https://trueopen.ai/reports/TrueOpen-SLP-PoC-Report.pdf},
note = {2026-09-17; source: TrueOpen.ai team, https://www.trueopen.ai/research.html}
}Seal, Then Sample: Sampled Layerwise Proofs for Verifiable LLM Inference from GPT-2 to 70B
Youki Lim; Sam Yong
Related later paper: arXiv:2609.27367v1 · submitted 2026-09-23. Not the September 17 report.
@misc{lim_yong_slp_2026,
author = {Lim, Youki and Yong, Sam},
title = {Seal, Then Sample: Sampled Layerwise Proofs for Verifiable LLM Inference from GPT-2 to 70B},
year = {2026},
eprint = {2609.27367},
archivePrefix = {arXiv},
primaryClass = {cs.CR},
doi = {10.48550/arXiv.2609.27367},
url = {https://arxiv.org/abs/2609.27367v1},
note = {Version 1; submitted 2026-09-23}
}Where are the raw logs?
The September 17 report describes a release containing raw logs; the website ZIP is a separate table export and does not include them. The public experiment repository contains raw-log files. We fetched the pinned 70B run log and checked its reported stages: 48,706.5 s inference, 1,259.1 s sealing plus proof, 46.3 s verification. This checks source consistency, not independent reproduction or completeness of every experiment.
70B raw log · pinned commit 592a5a6 ↗Download scope: the ZIP files contain published report tables and report files, not the complete raw run logs. Read the included README before attempting reproduction.
Sources and artifacts
- Original report (English PDF)
- SLP results tables (not raw logs)
- SLP experiment data repository
- Related paper: Seal, Then Sample
Cite this research overview
@misc{trueopen_sampled_layerwise_proofs_en,
author = {{TrueOpen}},
title = {What does a sampled inference proof actually prove?},
howpublished = {\url{https://www.trueopen.ai/en/research/sampled-layerwise-proofs/}},
note = {Research overview; original experiment sources linked on page}
}