Schema Study v0.1.0 — Four-category workflow and Mercury deployment

The four-category paper–dataset evidence workflow, interchangeable model profiles, a CUDA Docker image and the Mercury Apptainer launcher are available in this release.

English is the default documentation language. 简体中文发布说明 remains available as an optional translation.

• User guide: prepare PDF inputs, parse datasets independently, build a corpus, run/resume jobs and verify packets.
• Deployment guide: Docker Hub to SIF, mounts, model selection, checkpoints, configuration freezing and GPU checks.
• Docker Hub: plalelab/schema-study:0.1.0-cuda13 (Linux amd64).

Pull the exact released image:

docker pull plalelab/schema-study@sha256:c6abbb7f853e295dcce0634029eb991a1f1d488bb548c986b40998b1476db45e
apptainer pull schema-study-0.1.0.sif docker://docker.io/plalelab/schema-study@sha256:c6abbb7f853e295dcce0634029eb991a1f1d488bb548c986b40998b1476db45e

Runtime source commit: 07e5f989b127d25783d9c65a6b83c414b4562bf3. The OCI digest differs from the converted SIF file's SHA-256; freeze the actual SIF bytes for each experiment. The documentation language update does not change the runtime tag, image or release assets.

Validation: 113 Linux container tests passed; Windows had 112 passes and one skip for unavailable symlink privileges. Release GitHub Actions passed. The three Python preparation examples were executed successfully. The offline demo completed 11 mock dispatches and zero new dispatches on resume; packet integrity and derivation checks passed, with zero real model calls.

The image contains code and dependencies. Supply PDFs, datasets, model weights and credentials externally. Historical frozen file hashes remain unchanged. Mercury GPU/SIF qualification, real inference and independent semantic evaluation have not been performed; example profiles remain drafts.