Overview
The reproducibility crisis in computational research is fundamentally structural: research artifacts are scattered across disconnected tools. template/ applies Infrastructure as Code to the research lifecycle, making the manuscript, test suite, and provenance chain version-controlled, deterministically buildable, and independently verifiable.
reproducible researchinfrastructure as codebuild pipelineopen scienceModel Context Protocol
Overview source: Curated paper metadata.
Methods and contributions
Read the source for the full argument, qualifications, and evidence.
Findings and contributions
- All three exemplar projects (39, 505 and 65 tests) completed the pipeline, a reported 100% success rate with zero mock violations.
- Total pipeline duration across the three projects was ~125s on Apple M-series hardware, ~42s per project on average.
- The infrastructure suite of ~3,083 tests reached 83%+ coverage against a 60% threshold, with zero mock violations.
- The comparative feature analysis is reported to show template/ uniquely combining eleven distinctive capabilities in one enforced pipeline.
- Among stated limitations, the provenance layer offers SHA-256 tamper detection but not cryptographic non-repudiation, since it lacks private-key signatures.
Methods
- Two-Layer Architecture: 12 shared infrastructure subpackages + standalone projects — template/ separates a reusable infrastructure/ layer (~150 Python modules) from self-contained projects/ workspaces that consume it.
- Eight-stage build pipeline from environment setup to LLM review — Stages run sequentially: setup, tests with coverage, analysis scripts, Pandoc/XeLaTeX rendering, hashing/watermarking, PDF validation, LLM review.
- Zero-Mock testing policy with 90%/60% coverage gates — Tests use real filesystem operations and subprocesses instead of mocks, with coverage thresholds of 90% for projects and 60% for infrastructure.
- SHA-256 hashing with steganographic watermarking for provenance — Rendered PDFs carry SHA-256 hashes in metadata plus alpha-channel overlays and QR codes for tamper detection.
- Multi-project evaluation and feature comparison with peer tools — Three exemplar projects were run through the pipeline in a fast configuration that skips infrastructure tests and LLM review; template/ was also compared with peer tools on fourteen dimensions.
Summary sources: Paper metadata and evidence · Extracted source text.
PDF downloads
Archived files available directly from this site.
- template_daf_v1_03202026.pdf PDF · 1.10 MiB · extracted-text source
Citation
Citation metadata follows the unified bibliography.
Friedman, Daniel Ari. 2026. A template/ approach to Reproducible Generative Research: Architecture and Ergonomics from Configuration through Publication. Zenodo. DOI: 10.5281/zenodo.16903351. URL: https://doi.org/10.5281/zenodo.16903351.
Catalog details and resources
Related in Computational
Other catalogued works in the same domain.