Review Process
Every submission enters the same curation system. Five independent reviewers evaluate the work against a six-dimension rubric. The panel runs twice. Every review is published with the accepted paper. Here is exactly what that means.
The curation system is open source
Every component the AI panel actually runs — prompts, the six-dimension rubric, scoring code, slot orchestration, Review Quality Control audit, citation verification, and the curation decision logic — is published openly. Read it. Audit it. Fork it. Run it on your own corpus. The reviewer does not have to be trusted on faith — you can verify what it does.
Self-hosted journals can adopt the entire stack. Researchers studying peer review can use the prompts and scoring code as a public benchmark. The architecture is the product.
The Panel
ICSAC's review panel is built from a roster of independent advanced AI reviewers, each drawn from a separate model architecture and calibrated by the Institute against the published rubric. Reviewers do not share scores during evaluation. Every submission is evaluated by up to ten independent reviewers.
- Up to 10 independent advanced AI reviewers per submission
- Each reviewer is a distinct model architecture; results are independent
- A minimum reviewer-completion threshold must be met — below it, the submission is held for human review
- Cross-reviewer disagreement is flagged and published, not averaged away
What Gets Evaluated
Domain Fit
Does the work use scientific, mathematical, or formal methodology to make testable claims? The panel also flags submissions requiring specialized empirical expertise it cannot fully provide.
Methodological Transparency
Are methods, data, and parameters documented well enough to reproduce the work?
Internal Consistency
Do the conclusions follow from the evidence and methodology presented?
Citation Integrity
Are references real, resolvable, and meaningfully connected to the claims?
Novelty Signal
Does the work advance the field? Novel frameworks are not penalized for limited prior literature.
Generative-Artifact Assessment
AI-assisted research is welcome. Work generated entirely by AI systems with no original human analysis or contribution does not meet the authorship standard.
Scoring
Each dimension is scored independently on a 1–5 scale.
| Score | What it means |
|---|---|
| 5 | Exceptional — field-advancing contribution |
| 4 | Solid and publishable — minor concerns only |
| 3 | Adequate — fundamentally sound, revision may be needed |
| 2 | Significant issues — major revision required |
| 1 | Fundamentally flawed, out of scope, or fails authenticity |
How Decisions Are Made
Every submission moves through two stages. The AI panel produces a recommendation against the six-dimension rubric. The recommendation is then routed to the curator — a real human — who reads the panel's full record alongside the manuscript and either confirms or overrides the recommendation. The curator's call is the verdict. Nothing publishes, no email goes out, and no deposit is staged until the curator's verdict lands.
Panel Recommendation
The AI panel evaluates the work against the published rubric and produces a recommendation. The panel advises; it does not decide. Every recommendation — clean, borderline, or against acceptance — is routed to the curator for review.
Curator Verdict
The curator reads the panel's full record alongside the manuscript and either confirms or overrides the recommendation. The curator's call is the verdict. Only after the verdict does anything downstream fire — the author email, the publication record, and (for upload-route submissions) the Zenodo deposit. The panel can be wrong; the curator is the step that catches it.
In every case where a paper publishes, the curator's reasoning publishes alongside the panel record at the paper's permanent URL.
Before adjudication, submissions are screened on authenticity, methodology, and scope. Work that fails the floor — manufactured citations, missing methodology, out-of-scope content — does not enter the review pool. Authors receive the panel record either way.
Citation Verification
Before the panel evaluates the manuscript, every citation in the bibliography is independently resolved against scholarly databases. References that cannot be located are flagged as potentially fabricated. References that resolve are then checked for misattribution — whether the cited work actually supports the claim it is attached to.
Both the resolution status and any misattribution flags are provided to reviewers before scoring begins. This separates "the paper looks plausible" from "the paper's foundation is verifiable" — reviewers do not evaluate the bibliography on faith.
Quality Control
After the panel completes, a separate automated audit evaluates the quality of the reviews themselves — checking that each reviewer cited specific content, scored consistently, and followed the rubric. If any reviewer fails the audit, the submission is flagged and held for human review before any decision is finalized. The audit results are published alongside the reviews.
Curation Recusal
When a submission is authored, co-authored, or has financial entanglement with an ICSAC curator or governance member, that curator recuses from the curation decision on the submission. Where no non-conflicted curator is available, the recommendation produced by the panel and the Quality Control audit is applied directly, without curator-side override or revision. The full review trail is published unredacted alongside the paper.
DOI vs. PDF Submission
Zenodo DOI or arXiv ID
- Best for work already deposited on Zenodo or arXiv
- Paste your DOI or arXiv ID — ICSAC fetches the manuscript and metadata automatically
- Your existing record is preserved as the archival source
PDF Upload
- For unpublished work with no existing DOI anywhere
- ICSAC mints a new Zenodo DOI on accept; the manuscript is deposited in the ICSAC Zenodo community on the author’s behalf
- Upload PDF with title, abstract, and keywords
- The manuscript must have an extractable text layer — image-only scans cannot be reviewed
- Do not submit work currently under review at another journal — most journals treat prior open deposit as prior publication and will withdraw your submission
Both paths enter the same curation system. The review standard does not differ by submission method.
What to Expect
- Submit — confirmation email sent immediately. ORCID required for every submission; ICSAC verifies authorship and sources author metadata from the ORCID record.
- Queue assignment — within 24–48 hours of receipt
- Advanced AI review — 24–48 hours after queue assignment
- RQC audit + human verification — 48–72 hours after review completes
- Result notification — full panel record sent to the submitting author
- If accepted — paper registered in the ICSAC public registry, landing page published, full review record published alongside the work
Typical elapsed time from submission to result: 5 to 10 days. Traditional peer review takes six to twelve months. Every review here is published.
Timeline is approximate. Submissions with incomplete methodology documentation, missing abstracts, image-only PDF scans, non-extractable text layers, or other deficiencies may require additional handling time or result in rejection prior to panel assignment. Authors are responsible for ensuring submissions meet the documentation standards described on this page.