A research program is not a sequence of documents. It is a repository. Papers are not the research — they are renders of the research at specific points on its timeline, shared with the scientific community so findings can be confirmed and knowledge can accumulate.
Paper Repo defines a protocol for managing scientific research as version-controlled repositories. The protocol optimizes the knowledge production process, not the paper. Individual papers are tagged releases — frozen snapshots forked to journals for community review — with automated compliance gates, attributed reviewer commits, and provenance chains that make the full research lifecycle machine-readable.
| Level | Meaning |
|---|---|
| Research program | Repository — the single source of truth, evolving over time |
| Paper | Render of the research at a point on its timeline — a frozen snapshot, a communication event |
| Fork | Sharing that render with a journal for confirmation |
| Publication | Merge — the community's confirmation that the findings join the shared knowledge base |
This maps onto the same rendering problem pattern that appears across domains: a specification is rendered into an expression, which is then evaluated by observers. In SBT: brand specification → signals → perception cloud. In OST: organizational schema → operations → performance metrics. In Paper Repo: research repository → paper → community evaluation (peer review). Same structure, different substrates.
Scientific publishing routes knowledge production through administrative checkpoints: formatting manuscripts, uploading to portals, manually blinding, re-reviewing papers rejected elsewhere. These activities consume editorial and author labor without creating knowledge.
The root cause: the system treats a paper as a static document (a PDF) rather than a living repository with history, contributors, branches, and provenance.
| Feature | What it does |
|---|---|
| Research repository | Every research program is a Git repo; papers are tagged releases with structured metadata (paper.yaml, CONTRIBUTORS.yaml, PROVENANCE.yaml) |
| Fork-based submission | A fork request to a journal creates a frozen, cryptographically linked fork |
| Compliance gate | Journal publishes journal_spec.yaml; validator checks compliance before fork request is processed |
| Blinding as function | Reviewer anonymization is automated per journal config, not manual |
| Reviewer attribution | Reviews are typed commits on branches — creating portable reviewer portfolios |
| Provenance by design | Every fork is irrevocably recorded; dual submission becomes structurally detectable |
| Collections as users | Journals, preprint servers, archives are collection users who curate forks |
| Hybrid submission | Git-linked submission coexists with traditional upload (like ORCID linking) |
| Federation | No central platform — any institution runs its own server |
A compliant paper repository:
paper/
paper.md # Manuscript source (Markdown, LaTeX, or Quarto)
paper.yaml # Claims, methods, falsification conditions (Paper Spec)
CONTRIBUTORS.yaml # Contributor roles verified against commit history
PROVENANCE.yaml # Fork history, submission records
DATA_MANIFEST.yaml # Links to external data archives (DOI + checksum)
LICENSE # Content license (CC-BY-4.0 recommended)
figures/ # Generated figures
analysis/ # Code, scripts, notebooks
See Paper Spec for the paper.yaml schema.
python scripts/validate_paper.py --repo /path/to/paper --spec examples/journal-specs/journal_spec_jm.yamlOutput:
Validating against: Journal of Marketing
Repository: /path/to/paper
------------------------------------------------------------
[OK] Manuscript: paper.md
[OK] Word count: 9,217 (limit: page-based)
[OK] Abstract: 187 words (max: 200)
[FAIL] References: self-citation 27.3% > maximum 25%
[OK] Statement: ai_disclosure
------------------------------------------------------------
Results: 5 passed, 1 failed
Submission gate: BLOCKED — fix failures before submitting
mkdir my-paper && cd my-paper && git init
# Copy template files:
cp /path/to/paper-repo/templates/* .
# Edit paper.md, paper.yaml, CONTRIBUTORS.yaml
# Validate: python validate_paper.py --repo . --spec journal_spec.yaml| Schema | Purpose |
|---|---|
schemas/journal_spec.yaml |
Journal submission requirements (46 field categories from 15 real journals) |
schemas/paper_repo.yaml |
Paper repository structure specification |
| Example | What it demonstrates |
|---|---|
examples/journal-specs/ |
Journal submission specs (machine-readable requirements) |
examples/R13/ |
Paper Spec paper — contributor attribution + fork history (Zenodo + QSS) |
examples/R14/ |
This protocol's own paper — full self-referential implementation (paper.md + paper.yaml + all artifacts) |
Open science and confidentiality are not mutually exclusive. A commit hash is a tamper-evident fingerprint — publishing the hash establishes that the work in this exact form existed at this exact moment, signed by this exact author, without revealing what the work is.
This is commit-reveal: priority and provenance established immediately, content disclosure under the author's control.
| Disclosure mode | What is public | What stays private |
|---|---|---|
| Solo private + hash anchors | Commit hashes (timestamped + signed) | Full content until author chooses to release |
| Consortium shared private branch | Periodic hash anchors | Branch contents — visible only to members |
| Hybrid public-private repo | paper.yaml, references, public commit graph |
.wiki/, raw data, draft branches |
| Embargoed disclosure | Hash at time t | Content until t + n; hash proves no editing during embargo |
| Standard open science | Everything | Nothing |
Years from now, in the event of a priority dispute, the researcher reveals the contents and any party can recompute the hash to verify the disclosed file matches the previously published anchor. Most academics assume git-native research forces them to choose between confidentiality and credit. Cryptographic commit-reveal gives them both.
For misconduct prevention, this is the structural alternative to centralized institutional databanks: a scientist's reputation is the verifiable history of their commits. There is no need to "report" misconduct at the moment of an employment transition — the misconduct, if it occurred, is visible at the granularity of the commit graph. Fabrication cannot retroactively edit a signed, hashed commit without breaking the chain.
Git doesn't just track your work. It tracks your AI's work too.
Every AI-assisted edit in a paper repository becomes a commit with metadata: the tool used, the model version, and what changed. When Claude rewrites Section 3, that's a commit. When Gemini generates a figure, that's a commit. The contribution is recorded not because the author declared it on a form, but because the system recorded it.
This solves the AI disclosure problem that journals are currently addressing with honor-system checkboxes. In a paper repository, AI contribution is structurally transparent and auditable by construction. The CONTRIBUTORS.yaml file maps every contributor (human and AI) to their verified commit history.
The research paper proposing this protocol (R14) is itself structured as a compliant paper repository. Its paper.yaml contains 7 typed claims with falsification conditions. Its CONTRIBUTORS.yaml lists 4 contributors (1 human + 3 AI tools). Its PROVENANCE.yaml will be populated as fork requests are made to venues.
Paper Spec defines WHAT a paper claims (paper.yaml schema).
Paper Repo defines HOW a paper is managed (repository structure, fork protocol, compliance gate).
They compose: paper.yaml is one file in the paper repository. Paper Spec validates the content; Paper Repo validates the process.
Zharnikov, D. (2026u). Research as Repository: A Git-Native Protocol for Scientific Knowledge Production. Working Paper. https://doi.org/10.5281/zenodo.19294864
- Paper Spec — Machine-readable standard for scientific claims
- Spectral Brand Theory — The research program this protocol emerged from
- Organizational Schema Theory — Test-driven business design (the OST value-stream lens informs this protocol)
MIT