ARR — Archive for Rigorous Research

ARR is the hostile-audit research registry: new admissions must survive an operator-selected frontier-model audit of the exact hashed manuscript, with zero unresolved material objections and a final human decision. ARR promises no fixed provider, model, report count or reasoning tier; it publishes exactly which models were used, their findings and any disagreement alongside canonical artifacts, machine-readable renditions and explicit verification evidence.

Passing that gate is meaningful positive evidence that a paper deserves serious attention. It is not a guarantee of truth, a proof certificate or a replacement for qualified domain-expert peer review; ARR makes the hard filter inspectable instead of asking readers to trust a label.

This repository is the technical foundation for the archive. It keeps lightweight, inspectable sources in Git and publishes large generated artifacts—PDFs, complete source bundles, datasets, and build logs—as GitHub Release assets.

Current status

ARR operates a public archive and is building a currently fee-free direct private-submission pilot. Authors will use one form without an invitation or account; activation remains gated by the signed production checklist. The manuscript enters private quarantine and email carries only an operator notice, never the PDF. No record is considered published merely because it appears in a branch, pull request or private intake. A research paper or technical note becomes an ARR publication only after a human acceptance workflow completes and a versioned release is created.

Local checks

python scripts/validate_papers.py
python scripts/build_site.py
python -m http.server 8000 --directory _site

Production builds also serve PDFs alongside each abstract for scholarly crawlers. See search indexing and Search Console setup for the complete build check, PDF integrity rules, and DOI metadata workflow.

Create a concurrent-safe candidate identifier and sharded directory with:

python scripts/new_record.py --author "Author Name" --type research-paper
python scripts/new_record.py --author "Author Name" --type technical-note

Create the next version of an existing record without changing its ARR identifier:

python scripts/new_version.py ARR-2026-XXXXXXXXXXXXXXXX --change-size minor --summary "Corrects notation and expands the reproducibility instructions."
python scripts/new_version.py ARR-2026-XXXXXXXXXXXXXXXX --change-size major --summary "Replaces the main argument and adds new principal results."

Repository layout

papers/                 Accepted record sources, grouped by year and month (legacy path name)
papers/.../versions/vN/ Immutable source snapshots for later versions
templates/paper/        Shared template for new paper and technical-note candidates
schema/                 Machine-readable metadata contract
scripts/                Validation, packaging and site generation
site/                   Static presentation assets
docs/                   Editorial and operational documentation
.github/workflows/       GitHub validation, Pages and Release automation

See docs/WORKFLOW.md for the publication procedure. Copy-ready agent instructions are in docs/AGENT_DEPOSIT_PROMPTS.es.md.

Scale and licensing

ARR separates private intake, the public metadata registry, and immutable artifact storage so that GitHub can be replaced or complemented without changing record identifiers. See capacity and migration, state and evidence labels, and the licensing policy.

The private intake service is in services/intake/. ARR does not currently charge for submission, assessment, publication or withdrawal under ARR-DEPOSIT-1.4; any future fee would apply only after advance notice and new terms. New admissions require the version-locked frontier-model gate in ARR-ASSESS-1.0, while legacy records remain honestly labelled not_assessed. The operator is Lluis Eriksson, founder, registry operator, responsible editor and GDPR controller. See privacy, complaints/contact, retention, governance and the production launch gate.

Platform software is AGPL-3.0-or-later; ARR-authored documentation is CC-BY-4.0; public catalogue metadata is CC0-1.0; deposited papers, code and data declare their own scoped licenses.

Public author profiles use stable identities from registry/authors.json. Paper and author activity rankings are generated from reproducible provider snapshots under ARR-METRICS-1.0; missing page-view measurement is never represented as zero or simulated activity.