Refine.ink Review (2026): Strong on Math and Proofs, Limited on Citations and Figures
Refine.ink is a credible AI manuscript review tool with strong tenured-economist endorsements and excellent depth on math, proof, and internal logic. It is not built for citation verification, figure parsing, or journal-fit scoring.
Readiness scan
Find out what this manuscript actually needs before you pay for a larger service.
Run the Free Readiness Scan to see whether the real issue is scientific readiness, journal fit, figures, citations, or language support before you buy editing or expert review.
Quick answer: Refine.ink (verified 2026-05-09) launched October 2025 by Ben Golub and Yann Calvó López. Strongest on math, proof, and internal-logic depth in theory-heavy manuscripts, with named tenured-economist endorsements (Fudenberg at MIT, Lederman at UT Austin, Tamuz at Caltech) and a Cochrane Substack endorsement. Per Refine.ink's own FAQ, the tool does not do citation verification, figure parsing (image-embedded equations are ignored), journal-fit scoring, experiment recommendations, or reviewer pushback prediction. For econ theory, formal philosophy, applied math, or theoretical CS papers, it is real value at $29.99 to $49.99 per review. For life sciences, clinical, or biomedical research, Manusights at $29 answers the four questions that decide selective-journal outcomes: is the novelty real against the most recent competing work, does the target journal fit, what would pre-empt reviewer 2, and which named desk-reject patterns will the editor flag.
Run the free Manusights scan in 1-2 minutes, no card required.
Method note: This review uses Refine.ink's live public product, pricing, FAQ, terms-of-service, privacy-policy, institutions, and team pages reviewed in May 2026. We did not personally purchase a Refine.ink review. The judgments are grounded in public sources, third-party endorsements (Cochrane Substack, Empiricrafting interview with Ben Golub), and Manusights internal analysis of submission-readiness failure patterns. We have no commercial relationship with Refine.ink.
What Refine.ink actually is
Refine.ink is the AI manuscript review tool at refine.ink (not refine.dev, which is a React framework for admin panels). Founded in October 2025 by Ben Golub (economist) and Yann Calvó López, the product is positioned as deep AI critique for theory-heavy manuscripts.
The strongest credibility signal is named tenured-economist endorsements published on the homepage:
- Drew Fudenberg at MIT
- Harvey Lederman at UT Austin
- Omer Tamuz at Caltech
Plus a third-party endorsement from John Cochrane (Grumpy Economist Substack) calling Refine.ink output "on the par of the best comments I've received on a paper in my entire academic career."
The institutions page also references that 87 published papers acknowledge Refine in print, and lists trusted-by institutions across major US universities.
Refine.ink supports 12 file formats including .tex and .latex as first-class formats with a 70k words / 50 MB upload cap (120k institutional).
Pricing
Refine.ink's pricing is unusually transparent (verified 2026-05-09):
Plan | Price | Notes |
|---|---|---|
Single review | $49.99 | One-time purchase |
3-pack | $119.99 ($39.99 per review) | Credits roll over per their FAQ |
10-pack | $299.99 ($29.99 per review) | Credits roll over per their FAQ |
Pro subscription | $100/month for 3 reviews | $33.33 per review at full utilization |
Subscription tiers | $40 to $300 per month | Variable scaling |
Refine.ink purchases are non-refundable per their terms of service.
What Refine.ink does well
Internal-logic depth on theory papers. The tool is built around math and proof rigor. For a paper where the central question is whether the formal argument holds together, Refine.ink is genuinely useful. The tenured-economist endorsements are real and earned in this domain.
LaTeX and theory workflow native. 12 file formats with first-class .tex and .latex handling. For econ theory or formal-philosophy papers drafted in LaTeX, the tool fits the workflow.
Notation and cross-reference rigor. Strong at catching equation-numbering inconsistencies, undefined notation, and proof-step gaps that linear reading often misses.
Privacy posture. Anthropic + OpenAI + Google + Azure sub-processors with zero-retention contracts. SOC 2 and ISO 27001 "currently pursuing" per the institutions page.
Distribution credibility. 87 published papers acknowledge Refine in print is a meaningful third-party signal that the tool is being used and credited by working researchers.
Where Refine.ink stops short
Per Refine.ink's own FAQ at refine.ink/faq:
- No citation verification. Refine.ink does not handle citation formatting, bibliography management, or fact-checking. If reference 14 has a wrong DOI, reference 23 was retracted last quarter, or a competing paper from 3 months ago is missing, Refine.ink will not flag it.
- No figure parsing. The tool does not parse figures, and reviewers have publicly noted that equations rendered as images are ignored entirely.
- No journal-fit scoring. The page does not advertise journal-calibrated readiness scoring, desk-reject prediction by named pattern, or target-journal alternative recommendations.
- No experiment recommendations. The tool does not produce a prioritized A/B/C plan of specific experiments to add to strengthen the claim before reviewer 2 demands them.
- Non-refundable purchases. All Refine.ink purchases are non-refundable per their terms of service.
- Signup wall before output. Refine.ink requires account creation before the first review, with no anonymous-evaluation path.
Where Refine.ink is the right buy
Choose Refine.ink when:
- the manuscript is heavy on formal proofs, notation, or internal references that need careful logic stress-testing
- the audience already knows Refine.ink and weighs Cochrane-tier endorsement
- the manuscript is in econ theory, formal philosophy, applied math, or theoretical CS, where the proof points actually live
- figures and equations are cleanly typeset (not embedded as images)
- the file format is .tex or .latex and you want first-class handling of LaTeX source
- you have used Refine.ink before and the workflow fits
In our experience, that makes Refine.ink most useful when the manuscript is at the "is the math airtight?" stage rather than the "is this paper actually ready for Cell or NEJM?" stage.
Readiness check
Find out what this manuscript actually needs before you choose a service.
Run the free scan to see whether the issue is scientific readiness, journal fit, or citation support before paying for more help.
Where Refine.ink is the wrong buy
Refine.ink is not enough on its own when:
- the manuscript is in life sciences, clinical, or biomedical research where citation accuracy and figure trust drive most desk rejections
- the figures contain image-embedded equations or micrographs that need to be parsed, not skipped
- the question is "is this paper actually competitive at this target journal?" rather than "is the internal logic clean?"
- you need a refund option (Refine.ink purchases are non-refundable per their terms)
- you want to evaluate the tool anonymously before signing up (Refine.ink requires signup before any output)
- the unresolved risk is novelty positioning against the most recent competing work, journal selection with reasoning, or predicted reviewer pushback at a specific target journal
That is where Manusights at $29 is built for a different layer of the workflow.
Refine.ink vs Manusights
The split is clean. Both products are AI. Both are credible. They answer different questions.
Question you are trying to answer | Better fit |
|---|---|
"Is the internal logic and proof depth of a theory paper airtight?" | Refine.ink |
"Is the math notation consistent across the manuscript?" | Refine.ink |
"Is my draft heavy on formal proofs, pure math, or theory in econ or philosophy?" | Refine.ink |
"Is the novelty positioned strongly enough for this journal?" | Manusights |
"What experiments should we add to pre-empt reviewer 2?" | Manusights |
"Will the editor at this journal desk-reject this?" | Manusights |
"Which journal should we actually target, and why?" | Manusights |
"Is my draft in life sciences, clinical, or biomedical methods?" | Manusights |
"Is the science strong enough to survive peer review?" | Manusights |
For the direct head-to-head comparison page, see Manusights vs Refine.ink.
Honest workflow for using both
If the manuscript is theory-heavy AND has citation or figure exposure, the strongest workflow is:
- Run a free Manusights scan first to identify the dominant risk
- If the dominant risk is internal-logic, proof, or notation depth on a theory paper, escalate to Refine.ink
- If the dominant risk is citation accuracy, figure trust, journal fit, novelty positioning, experiments to add, or predicted reviewer pushback, stay with Manusights' Full AI Diagnostic at $29
- Revise based on both outputs before deciding whether the manuscript is truly ready
That is better than treating the tools as mutually exclusive when the manuscript would benefit from both kinds of checks.
Bottom line
Refine.ink is genuinely strong at one specific job: internal-logic depth on theory-heavy manuscripts. The named tenured-economist endorsements and the Cochrane third-party signal are credible. At $49.99 per single review, $39.99 in a 3-pack, or $29.99 in a 10-pack, the pricing is reasonable for that audience.
But Refine.ink's own FAQ makes the category limit explicit: no citation verification, no figure parsing, no journal-fit scoring. If your manuscript is in life sciences, clinical, or biomedical research where those gaps drive rejection, the tool stops short of the science-survival decision.
Manusights at $29 is built for that layer. We give you the kind of feedback an experienced reviewer in your field would write: novelty against the live literature, journal-fit reasoning, and the specific experiments and reviewer objections that decide the outcome.
Run the free Manusights scan in 1-2 minutes, no card required, to find out which layer your manuscript needs.
Frequently asked questions
Yes. Refine.ink is the AI manuscript review tool launched in October 2025 by Ben Golub (economist) and Yann Calvó López. It is distinct from refine.dev, which is a React framework for admin panels. The tool has named tenured-economist endorsements (Drew Fudenberg at MIT, Harvey Lederman at UT Austin, Omer Tamuz at Caltech) and a Cochrane Substack endorsement.
Refine.ink charges $49.99 per single review, $119.99 for a 3-pack ($39.99 each), or $299.99 for a 10-pack ($29.99 each). Subscription tiers range from $40 to $300 per month. Per their FAQ, credits roll over with no expiration. All purchases are non-refundable per their terms of service.
Internal-logic depth, proof rigor, notation consistency, and cross-reference checking on theory-heavy manuscripts. The tool is strongest in econ theory, formal philosophy, applied math, and theoretical CS, where math and logic depth are the primary review concerns.
Per their own FAQ, Refine.ink does not handle citation formatting, bibliography management, or fact-checking. It does not parse figures (image-embedded equations are ignored). It does not score journal fit, predict desk-reject patterns, or evaluate citations against a live literature database.
Manusights at $29 is built for the science-survival decision in life sciences, clinical, and biomedical research. It systematizes editor-and-peer-reviewer-grade scientific feedback, novelty positioning grounded against the live literature (CrossRef, PubMed, OpenAlex, Semantic Scholar, bioRxiv, medRxiv), deep journal selection with reasoning, prioritized A/B/C experiment plan, and predicted reviewer pushback by named pattern. Refine.ink does not advertise any of those layers.
Sources
Final step
Run the scan before you spend more on editing or external review.
Use the Free Readiness Scan to get a manuscript-specific signal on readiness, fit, figures, and citation risk before choosing the next paid service.
Best for commercial comparison pages where the buyer is still choosing the right help.
Anthropic Privacy Partner. Zero-retention manuscript processing.
Where to go next
Supporting reads
Conversion step
Run the scan before you spend more on editing or external review.
Anthropic Privacy Partner. Zero-retention manuscript processing.