Two free instruments for research-integrity screening
ARI² — the Author Research Integrity Risk Index, validated and published, screens an author's bibliometric profile for commercial-authorship and paper-mill patterns. JPI² (Beta — calibration in progress) — the Journal Publishing Integrity Risk Index, screens a journal's public metrics for the same family of patterns at the venue level. Free, open-source, built entirely on public data. Each reading tells you where to look — not what to conclude.
🌍Researchers from 18 countries across five continents have used the Kelleni Index — with zero paid promotion.
What is ARI²
The first researcher-level integrity risk index
ARI² is the first author-level bibliometric tool specifically designed to detect commercial-authorship and paper-mill patterns using only public data. It combines four signals into a composite score. In the pilot feasibility study (28 source-verified authors), ARI² achieved an AUC of 0.938 (p = 0.003) for distinguishing paper-mill cases from controls. A high score is a scientifically validated prompt to stop, review, and verify — then apply fair human judgment.
Signal A · 35%
Network Fragmentation
Co-authors ÷ Works
Weight: 35%
Paper-mill authors accumulate many unique co-authors per paper — mostly one-time collaborators from unrelated institutions. High values suggest transactional rather than collaborative authorship patterns.
Signal B · 25%
Retraction Burden
Retractions ÷ Works
Weight: 25%
Proportion of output formally retracted. Signal B makes full ARI² retrospective by design. Retractions are detected from OpenAlex is_retracted flags. Cross-check against Retraction Watch for high-stakes assessments.
Signal C · 20%
Output Anomaly
Works ÷ h-index
Weight: 20%
Paper-mill participants inflate work count without commensurate impact. Interpret cautiously for early-career researchers and specialized subfields where citation counts are structurally lower.
Signal D · 20%
Citation Integrity
(Works × h + 1) ÷ (Citations + 1)
Weight: 20%
Inverted citation density metric. High values flag hollow citation profiles. Not field-normalized in the current version — context review is essential.
Tool
Calculate an ARI² Score
Choose automatic lookup (recommended) to let the tool fetch all data directly from OpenAlex, or enter values manually if you already have them.
Search by author name
Type the author's name to find their OpenAlex profile, then click the matching author. The tool automatically retrieves every value needed for ARI² — including co-authors and retractions — directly from OpenAlex and shows the result.
Fetching author profile...
Starting...
⚖
Search for an author and click their name to see the ARI² result.
Enter Values Manually
Total publications indexed in OpenAlex
Total citation count from OpenAlex profile
h-index from OpenAlex author profile
Count of all distinct individuals who have co-authored at least one paper with this author. See the guide below to obtain this value from OpenAlex.
Retracted papers per OpenAlex or Retraction Watch. Enter 0 if none confirmed.
⚖
Enter all values and click Calculate to see the ARI² score.
How to Use
Step-by-step guide
The automatic lookup handles all data retrieval for you. The guide below explains how to find the OpenAlex Author ID and how to obtain co-author and retraction data manually if you prefer the manual entry mode.
1
Go to the Calculator section and type the author's name
In the Automatic Lookup tab of the calculator above, type the author's full name and click Search. The tool searches OpenAlex and returns matching author profiles.
kelleniindex.com
3 results found
Abdur Rauf
University of Swabi · Pharmacology · Pakistan
975 works · 24,738 citations · h-index 76
2
Click the matching author — the result appears automatically
Click on the correct author from the results. Verify first by checking the institution, field, and country, since multiple authors with the same name may appear. As soon as you click, the tool retrieves the author's full works list from OpenAlex, counts unique co-authors across all publications, counts retracted works, and calculates all four signals and the composite ARI² score. A progress bar shows the status for authors with large publication records.
What is computed automatically: Works count, Citations, h-index — from the author profile. Unique co-authors — by collecting all co-author IDs across all works and counting each person once. Retractions — by counting works where OpenAlex has flagged is_retracted = true.
No separate button, no downloads, no exports, no manual counting needed.
If the correct author does not appear in the results, try a different spelling, add the institution name, or use the Manual Entry mode with values obtained directly from openalex.org.
1
Go to openalex.org and find the author profile
Search for the author by name on openalex.org. Select Authors in the category filter. Click the correct profile and verify using institution and field.
openalex.org/authors/A··········
Author Name
Institution · Field
285
Works
12,450
Citations
42
h-index
Works, Citations, and h-index are visible on the author profile
2
Obtain the unique co-author count via the OpenAlex API
The unique co-author count is not displayed directly on the author profile page. It requires fetching the author's works data. The simplest method: copy the author ID from the page URL (the part starting with A followed by digits), then use the following API call in your browser address bar.
openalex.org/authors/A5023888391
Copy the Author ID from the URL — the bold part above
Paste this URL in your browser, replacing the Author ID with the one you copied. The resulting JSON contains co-author IDs and retraction flags for all works. The automatic lookup mode does this for you.
Easier alternative: Use the Automatic Lookup mode. It performs this entire process for you and shows the co-author count before calculating the score.
Automatic Lookup is strongly recommended for all users.
1
When to use Retraction Watch
The Automatic Lookup mode detects retractions from OpenAlex's is_retracted field, which is the same source used in the original ARI² research. OpenAlex's retraction detection is not exhaustive — some retractions are indexed later than others. Retraction Watch is the most comprehensive retraction database and should be used to cross-check the result for high-stakes editorial assessments.
When to cross-check with Retraction Watch: When the automatic lookup returns a non-zero retraction count and you need to verify the exact number. When the automatic lookup returns zero but you have reason to suspect retractions exist. Before taking any formal editorial action based on a High score band result.
For routine screening, OpenAlex retraction data is sufficient. For formal proceedings, always verify against Retraction Watch.
2
Search the Retraction Watch Database
Visit retractionwatch.com/retraction-watch-database and search by the author's full name. Count the total retracted papers listed. If a discrepancy exists between OpenAlex and Retraction Watch, use the higher count and enter it manually using the Manual Entry mode.
retractionwatch.com/retraction-watch-database
Retraction Watch Database
3
retracted papers found — compare this with the OpenAlex count
How to read ARI² signals
This is a stable reference for understanding ARI² before running a search. The live Signal Profile panel shown with every result explains your specific reading; the cards below explain what each signal means in general. Every signal is a public-record screening cue — never a misconduct finding. Tap a card to expand it.
Signal A Network Fragmentation · co-authors ÷ works
Does the author carry an unusually high number of unique co-authors relative to total works?
Elevated Signal A can reflect the one-time-collaborator pattern seen in commercial-authorship networks — but it is equally consistent with legitimate consortium science, multicenter clinical research, interdisciplinary work, or large collaborative fields. It should not be read as suspicious unless corroborated by other signals such as unusual output growth, retractions, abnormal geographic dispersion, or incoherent co-author/topic pairings.
Signal B Retraction Burden · retractions ÷ works
Do retractions form a higher-than-expected share of the author's indexed output?
This is the most direct integrity-related signal, but retractions are not all equivalent. They may reflect misconduct, honest error, publisher or editorial problems, paper-mill infiltration, duplicate publication, authorship disputes, or corrective scientific practice. Retraction reason, timing, journal pattern, and the author's role must all be reviewed before any conclusion is drawn.
Signal C Output Anomaly · works ÷ h-index
Is total output high relative to demonstrated scholarly impact (h-index)?
Elevated Signal C may reflect inflated output without commensurate impact — but it may also reflect recent productivity, early citation lag, a publication-type mix (letters, protocols, conference material, preprints), local or clinical research, or a large body of legitimate but low-citation work. Read it as an output-density flag, not as evidence of misconduct.
Signal D Citation Integrity · (works × h + 1) ÷ (citations + 1)
Is citation density low relative to the author's works and h-index in the current OpenAlex record?
This signal is not yet field-, age-, or document-type-normalized, so it may reflect specialty citation culture, recent output, publication-type mix, database coverage, or resource-limited visibility rather than misconduct. It should not be read as a misconduct signal unless supported by other independent ARI² signals.
ARI²-NR No-Retraction Variant
What do the network, output, and citation-density patterns look like with retractions removed?
ARI²-NR excludes Retraction Burden and redistributes its weight across the remaining signals (A ≈ 46.7%, C ≈ 26.7%, D ≈ 26.7%). It is useful as a prospective preview of patterns that exist before any retraction does — but it is less validated than the full ARI² score. A higher ARI²-NR with zero retractions should be read cautiously and contextually, never as a stronger allegation.
Primary Driver Which signal dominates
Which single signal contributes most to the current score?
The Primary Driver explains the score's structure; it does not identify misconduct. A profile driven mainly by Retraction Burden should be read differently from one driven mainly by Network Fragmentation, Output Anomaly, or Citation Integrity. Because drivers are derived from cohort-level normalization, they are not stable properties of an individual author and can shift with cohort composition. The driver should guide review, not replace it.
PPGDI Per-Paper Geographic Dispersion
How many unique co-author countries appear per eligible multi-author paper?
High geographic dispersion can be compatible with commercial-authorship networks, but it equally reflects genuine global collaboration, meta-research, international consortia, or multicenter studies. PPGDI is exploratory, requires no retraction to compute, and is most informative when combined with network fragmentation, output anomaly, topic mismatch, or retraction burden — never read in isolation. In the pilot, the highest control PPGDI slightly exceeded the paper-mill group mean, confirming the metric is not diagnostic alone.
Consistent with the ARI² manuscript, which formally defines Signals A–D and emphasizes that moderate or high outputs warrant contextual human review rather than automatic judgment. For what the overall score bands mean, see ARI² Interpret below.
Interpreting ARI² Results
What the score bands mean
ARI² scores are calibrated against the 28-author pilot cohort. Score bands are exploratory heuristics — not validated cut-offs. Every flag requires human review and contextual assessment before any decision is taken.
Low Score Band
0 – 33.3
No unusual bibliometric pattern detected relative to the pilot cohort. Standard editorial review applies. A low score does not exclude misconduct not captured by the four signals — particularly content fraud or selective data manipulation.
Moderate Score Band
33.3 – 66.7
One or more signals are elevated. This may reflect legitimately large collaboration networks, high-volume productive researchers, or genuine authorship irregularities. Consult the due-process checklist. Six high-output controls with verified zero retractions scored in this band in the pilot study.
High Score Band
66.7 – 100
Multiple signals significantly elevated relative to the pilot cohort. This pattern is consistent with commercial authorship or high retraction burden. A high score is a prompt to stop, review, and verify: confirm author identity, check database-merging errors, compare across databases, examine field norms — then apply fair human judgment using the due-process checklist. The score tells you where to look. It does not tell you what to conclude.
Calibration caveat: Scores are normalized against the 28-author pilot cohort (Kelleni 2026) and are not yet externally validated. Authors whose profiles fall outside the pilot range may receive scores below 0 or above 100 — these should be read qualitatively. For the full open-source scoring pipeline, see the Zenodo repository. A validated reference population is under development.
Companion Tool · Beta — Calibration in Progress
Screen a JournalJPI² · v1
The Journal Publishing Integrity Risk Index (JPI²) is the venue-level companion to ARI². JPI² is in Beta: its observational-signal thresholds are provisional and reference-class calibration is still in progress, so every reading should be treated as exploratory rather than final. Enter a journal name to see whether its public metrics show signs that warrant careful review. As with ARI², a high reading is not a verdict — and the tool never labels a named journal "predatory" or "a paper mill." It reports a risk signal and the evidence behind it, which is a fair basis to investigate further, and nothing more. Editors and publishers can also use it to screen their own journals proactively for early signs of special-issue infiltration, before retractions accumulate. v1.3.1 runs J-A (retraction burden, band-driving) live from OpenAlex with Crossref/Retraction Watch confirmation, plus J-B, J-C, J-D, and J-G (the ARI²-bridge signal) as observational. See the interpretation guide below.
Search by journal name
Type a journal title to find it on OpenAlex. Confirm the exact journal — publisher and ISSN are shown — before reading the result; matching the wrong journal to a score is something this tool is built to avoid.
Try:J. of Healthcare EngineeringComp. Math. Methods Med.PLOS ONEScientific ReportsAngewandte Chemie
Search for a journal or pick an example to see its screening profile.
Interpreting JPI² Readings
How to read a journal screening
A JPI² reading describes public-record signals, not a verdict. The headline is not a single 0–100 score. In v1.3.1 it is chosen by rule-based severity tiers: the lead retraction-burden signal (J-A) can drive a Moderate or Elevated headline on its own, and observational signals (output-burst, issue concentration, topic concentration, author-network patterns) can escalate a reading even when J-A is low. See the JPI² Interpret section below for what each headline state means.
How to read the individual signals
JPI² is a multi-signal index, not a retraction-only tool. Each signal answers one specific question about the journal's public record. The live tool's Signal Profile panel (shown with every result) reports the current status of every signal below for the journal you search — this glossary explains what each one means and how reliable it currently is.
J-A — Retraction BurdenAre retractions unusually high relative to the journal's total indexed output?Calibrated reference range · band-driving (can set the headline alone)
J-B — Output-Burst ConcentrationDid the journal show an abrupt publication spike in a single year rather than steady, organic growth?Observational · provisional threshold, not yet reference-class calibrated
J-C — Issue ConcentrationAre recent articles clustered unusually heavily into a small number of issues or special issues?Observational · provisional · not applicable to continuous-publication journals and megajournals
J-D — Topic-Field ConcentrationIs the journal's most-cited output unusually concentrated in a narrow topic pattern?Observational · provisional · contextual-only in v1.3.1 (cannot alone escalate a headline)
J-G — Author-Network AggregateDo this journal's recent first-authors show elevated, simplified ARI²-style anomalies as a group?Observational · provisional · aggregate only — no individual author is ever named or scored
J-F — Content AnomalyDoes the published text show tortured-phrase or non-standard linguistic fingerprints associated with paper mills?Planned · specified in the methods document, not yet implemented
This is why the JEPH and NEJM cases in the manuscript matter for reading any single result: JEPH was missed because J-A sat below the Moderate cut and the observational signals did not escalate; NEJM over-escalated because J-D alone read high before reference-class calibration existed. A signal's current reliability is part of how it should be weighed, not just its value.
Why the headline is not always J-A. Letting the retraction-burden signal alone drive the headline would mean that a journal with missing or zero-coverage retraction data (which the tool cannot always distinguish from genuinely clean) would receive a green band by default. That is a false reassurance, which is as harmful as a false alarm — and worse, it would be the harm that hurts the user trying to choose where to submit. JPI² is architected so that observational signals (J-B output-burst, J-C issue concentration, J-D topic concentration) can override a J-A-only headline whenever they show strong anomalies.
Why some signals are observational, not band-driving. J-A has a calibrated reference range (the 0.21% baseline from peer-reviewed retraction-rate studies, plus the 10× factor seen in captured journals). J-B, J-C, and J-D do not yet have calibrated cut-points across the full reference-class normalisation (size, open-access model, discipline, age). Until calibration is complete, they are shown as observations — informative for the reader, but never used alone to issue a coloured band. This is intellectually honest and protects against false flags on legitimate large-scale venues.
A reading is a question, not an answer. Every JPI² output is the start of a careful review, never the end of one. The evidence panel directly links to the public sources used to compute the signals — verify everything for yourself. Common legitimate explanations include a clean-up phase after external misconduct, or a specialised topic with naturally narrow citation distribution. See the due-process checklist below for the full verification steps before drawing any conclusion.
v1.3.1 coverage and limitations. JPI² currently runs J-A (band-driving, via OpenAlex is_retracted with Crossref/Retraction Watch confirmation) plus J-B, J-C, J-D, and J-G (the ARI²-bridge: simplified ARI² signals aggregated across the journal's recent first-authors) as observational signals. J-F (content anomaly via Problematic Paper Screener) is specified in the methods document and not yet wired. Provisional weights and reference-class normalisation thresholds are non-portable and exploratory until calibrated on a large journal sample. The methods specification (in preparation) will document the validation procedure and the discriminant statistics in detail.
Interpreting JPI² Readings
The five headline states
Every JPI² reading resolves to one of the five states below. The headline is not a single 0–100 score. In v1.3.1 it is chosen by rule-based severity tiers: the lead retraction-burden signal (J-A) can drive a Moderate or Elevated headline on its own, and observational signals can escalate a reading even when J-A is low — never the reverse. Each reading is accompanied in the tool by a Signal Profile panel showing the primary driver and the status of every signal. As with ARI², a reading is a screening signal and a basis for review — never a verdict.
No Elevated Signal
J-A < 0.5% · signals clean
The current public metrics do not cross the JPI² screening thresholds. This does not prove the journal is clean, and it is not a certificate of quality — retraction records lag behind infiltration events, so a recently-affected journal may not yet show an elevated burden. It means only that the signals JPI² currently computes do not show an elevated pattern. Ordinary scholarly checks (scope, peer-review transparency, indexing history, publisher information, article-level quality) remain worthwhile regardless of the reading.
Moderate Screening Signal
J-A 0.5–2%
The journal's retraction burden is elevated relative to its indexed output, though not in the highest band. This may reflect several different contexts — including delayed clean-up after externally-introduced paper-mill activity, ordinary editorial correction, or other integrity-related events. It is a reasonable basis for closer, context-aware review, not a conclusion of misconduct.
Mixed Profile — Review Required
J-A < 0.5% · observational signals elevated
The retraction-burden signal is not elevated, but one or more observational signals show an unusual pattern — output bursts, issue clustering, topic-field concentration, or recent author-network anomalies. Because these observational thresholds are provisional, this result should be interpreted cautiously and checked against the journal's context, editorial history, and public records. The Signal Profile panel names which signal is driving the reading. When the only elevated signal is the author-network bridge (J-G) — the most provisional signal, drawn from a small first-author sample — the headline softens to a Contextual Author-Network Alert, since a lone, uncalibrated signal is a cautionary observation rather than a journal-level concern.
Elevated Screening Signal
J-A > 2%
The journal's retraction burden is strongly elevated relative to its indexed output. In the proof-of-concept set, this range was seen in a documented journal later shuttered amid paper-mill capture concerns. This is a strong public-record screening signal and a reasonable, good-faith basis for prioritised independent investigation under established procedures (e.g. due process, COPE). It is still not a verdict, and it does not by itself prove misconduct or culpability.
Anomalous Pattern / Insufficient Data
J-A unavailable or eligibility not met
If J-A cannot be retrieved but one or more observational signals are elevated, JPI² returns an amber Anomalous Pattern Detected headline — a missing retraction-burden reading is never allowed to silence strong observational anomalies. If the journal is too new, too small, or lacks a resolvable OpenAlex source record, the tool returns Insufficient Data rather than forcing a band; verify directly through OpenAlex, Crossref/Retraction Watch, DOAJ, and the journal's own editorial and publication records.
Why this is a profile, not a score. JPI² v1.3.1 issues a headline from severity-tiered rules across its signals, not from a single weighted number. The per-signal weights shown in the methods paper are provisional parameters for a future composite and are not applied in the current headline. Presenting a single 0–100 score now would imply a precision the provisional thresholds cannot yet support \u2014 so the tool reports a structured Signal Profile instead. For the full reasoning behind each signal and the architecture, see the JPI² Guide above.
Exploratory Companion Signal
PPGDI — Geographic Diversity Index
The Publication-Pattern Geographic Diversity Index (PPGDI) is an exploratory companion metric to ARI². It captures the breadth of geographic spread across an author's co-author network and may help identify unusual international collaboration patterns associated with commercial authorship behaviour.
What PPGDI measures: PPGDI is computed as the ratio of unique co-author countries to total unique co-authors. A value approaching 1 means nearly every co-author comes from a different country — an unusual pattern in normal academic collaboration. A low value indicates concentrated, locally coherent networks.
When it is most relevant: PPGDI carries the most interpretive weight when combined with a high Signal A (Network Fragmentation) score. In the pilot dataset, PPGDI showed potentially meaningful group-level differences and may help capture cross-geographic collaboration anomalies associated with commercial authorship. Alone, however, it cannot distinguish legitimate international collaboration from suspicious dispersal.
Future versions will compute PPGDI automatically from OpenAlex institution-level country metadata. Currently, manual entry is required. The co-author count used here should match the value entered in the ARI² calculator.
Calculate PPGDI
Enter the number of unique countries represented across your author's co-author network. The co-author count carries over automatically from your ARI² calculation above.
How many distinct countries appear across all co-authors? This is the one value OpenAlex does not yet return automatically.
Co-authors:—
Run the ARI² calculator above first and this will fill automatically.
🌍
Enter co-author countries and total co-authors to calculate PPGDI.
How to interpret PPGDI
PPGDI Pattern
Interpretation
Action
Low PPGDI (< 0.15)
Limited geographic dispersion; locally or regionally concentrated collaboration
No additional concern from geographic signal
Moderate PPGDI (0.15 – 0.35)
Mixed or regionally diverse collaboration; common in established international networks
Potentially more relevant to commercial-authorship pattern; network may be transactional rather than collaborative
Include in due-process checklist review
High PPGDI alone
Not diagnostic; may reflect entirely legitimate international collaboration in global research networks
No action on PPGDI alone
Important: A high PPGDI score does not prove misconduct. International collaboration is often legitimate and valuable. PPGDI should only be interpreted as a contextual anomaly signal, especially when combined with high network fragmentation, unusually rapid output growth, retraction burden, or other validated indicators. PPGDI is an exploratory research signal, not a diagnostic or misconduct-determining metric. It is not part of the validated ARI² composite score.
Future Development
ARI²-PM — Prospective Module
The current ARI² is primarily a retrospective risk index: Signal B (Retraction Burden) means the index reaches full discriminative power only after retractions have occurred. A prospective extension — ARI²-PM — is under development, aiming to detect anomaly signals earlier in the publication cycle.
RESEARCH IN PROGRESS
Toward ARI²-PM: Retraction-Free Anomaly Detection
ARI² is currently best understood as a retrospective author-level research-integrity risk index with prospective potential. A future ARI²-PM module may integrate retraction-free collaboration anomalies for earlier detection of commercial authorship patterns — potentially before any retractions have been registered.
Candidate signals under investigation for ARI²-PM include the following. Each requires further empirical validation before inclusion in a scored index.
PPGDI
Geographic diversity of co-author network — already available as an exploratory companion signal above
Affiliation Diversity
Number of distinct institutional affiliations per paper relative to authorship size
Author–Topic Mismatch
Unusual breadth of disciplines across an author's publication record
Special-Issue Clustering
Disproportionate publication in special issues relative to regular journal submissions
Authorship-Order Irregularity
Abnormal patterns in first/last author positioning across a publication career
Cross-Disciplinary Anomaly
Unlikely co-authorship across distant scientific fields without clear interdisciplinary rationale
Researchers interested in collaborating on ARI²-PM validation are welcome to contact the developer at drthabetpharm@yahoo.com. All methodological development will be documented openly via the Zenodo repository.
Responsible Use
Due-process checklist
A moderate or high ARI² score, or a Moderate/Elevated JPI² reading, is a prompt to stop, review, and verify — then judge fairly. The checklists below must be applied before any editorial, institutional, or publishing decision is taken. A reading tells you where to look. It does not tell you what to conclude. That is a feature, not a limitation.
ARI² — author screening checklist
1
Verify author identity
Confirm via ORCID. Name disambiguation errors in OpenAlex can inflate co-author counts for authors with common names. Ensure the profile has not merged multiple distinct researchers.
2
Check database-merging errors
Compare author records across Google Scholar, Scopus, OpenAlex, and institutional profiles. Discrepancies in publication counts, h-index, or affiliation may indicate profile-merging artefacts rather than genuine anomalies.
3
Review field context
Is high co-author count normal in this field? Large consortium science and clinical trial networks legitimately produce high Signal A values.
4
Assess career stage
Is the author early-career? High output may reflect a productive lab. Is the elevated signal recent or consistent across the full career?
5
Check retraction details
Are retractions from a single event or distributed across multiple journals and years? The pattern matters as much as the count.
6
Review co-author network
Do co-authors appear across many unrelated papers? Are many one-time collaborators from geographically and institutionally unconnected locations?
7
Examine journal patterns
Were many papers submitted to the same journal cluster or known predatory publishers? Unusual special-issue clustering warrants scrutiny.
8
Apply proportionate response
ARI² flags trigger investigation — never automatic rejection. All decisions must follow institutional due-process procedures and allow the author to respond — see the right-of-reply channel below. No author should face any consequence based solely on a score.
JPI² — journal screening checklist
1
Verify the retraction data directly
Confirm retraction counts and dates in Retraction Watch and Crossref. OpenAlex's is_retracted flag is the fastest live source but can lag behind very recent retraction notices.
2
Check for an active or completed cleanup phase
Has the journal, its editorial board, or its publisher already issued a public statement, expression of concern, or retraction notice explaining the cause? A high reading during transparent cleanup is a sign of accountability, not concealment.
3
Distinguish a special issue from systemic decline
Many documented paper-mill captures concentrate in one or two guest-edited special issues. Check whether the J-C issue-concentration signal and the retraction dates point to a contained episode rather than a journal-wide pattern.
4
Review editorial leadership and publisher history
Has the editorial board changed since the flagged period? Has the publisher (e.g. via COPE membership or public notices) responded with a documented remediation process?
5
Consider scale, discipline, and indexing status
Megajournals can carry high absolute retraction counts while remaining clean on a normalised (J-A) basis; narrow specialist journals can show naturally high topic concentration (J-D) without misconduct. Check the journal's current DOAJ, Scopus, or Web of Science status for independent corroboration.
6
Treat observational signals as context, not proof
J-B, J-C, J-D, and J-G are provisional and not yet reference-class calibrated (see the JPI² Guide). They support or qualify the J-A reading; none of them alone should be treated as confirming a pattern.
7
Give the journal or publisher the opportunity to respond
Use the right-of-reply channel below before drawing or publicising any conclusion. Verified context from the journal will be noted in the instrument's documentation.
8
Apply proportionate response
JPI² flags warrant investigation — never automatic delisting, citation refusal, funding consequence, or reputational sanction against the journal, its editors, its publisher, or any author who has published there.
Ethical Considerations and Responsible Use
ARI² scores are probabilistic risk indicators, not determinations of misconduct. A high score reflects a statistically unusual publication pattern that warrants human review and contextual investigation — not automatic sanction. No author should face rejection, retraction, defunding, or any professional consequence based solely on an ARI² score. The index is a screening aid. Misuse of this index to make adverse decisions without adherence to due process would be both ethically unsound and legally perilous.
Likewise, JPI² readings are public-record screening signals about a journal's published output, not determinations of misconduct by the journal, its editors, its publisher, or any author who has published there. An elevated reading reflects a statistically unusual pattern that warrants independent investigation under established procedures (e.g. COPE) — not automatic delisting, citation refusal, funding consequence, or reputational sanction. The index is a screening aid. Misuse of this index to make adverse decisions against a journal or its contributors without adherence to due process and an opportunity to respond would be both ethically unsound and legally perilous.
Right of reply
An opportunity to respond is the most important step in any due-process procedure. Both screened authors and screened journals have a named, documented channel to add context before any reading is acted upon.
Author — right of reply (ARI²)
If you are an author and wish to add context to an ARI² reading — whether to flag a profile-merging or disambiguation error, note a consortium or large-collaboration context, document a retraction history not captured by the public record, or provide any other relevant clarification — please write to drthabetpharm@yahoo.com. Verified context will be noted in the instrument's documentation.
Journal or publisher — right of reply (JPI²)
If you represent a journal or publisher and wish to add context to a JPI² reading — whether to flag a data error, note active cleanup activity, or provide evidence of editorial reform — please write to drthabetpharm@yahoo.com. Verified context will be noted in the instrument's documentation.
A transparent and responsive correction process is part of the responsible design of both instruments. The right of reply is not a courtesy; it is a structural feature of the due-process framework these tools require.
About
Developer
Dr. Mina T. Kelleni
MD, PhD · Assistant Professor of Pharmacology · College of Medicine, Minia University, Egypt Research Fellow · INTI International University, Malaysia Stanford / Elsevier Top 2% Most Cited Scientists (2022–2025)
Dr. Kelleni is a pharmacologist and clinical researcher who developed ARI² (Author Research Integrity Risk Index), JPI² (Journal Publishing Integrity Risk Index), and the Per-Paper Geographic Dispersion Index (PPGDI) as free, open, public-record bibliometric tools for research integrity screening. The index was developed without institutional funding in a resource-limited research environment, using AI-assisted analysis under the author's direction and supervision. Its development was motivated by direct observation of paper-mill solicitation and by a commitment to transparent, accessible research integrity tools that researchers in any setting can apply freely.
This tool uses only publicly available bibliometric data from OpenAlex and the Retraction Watch Database. No human subjects are recruited, contacted, or experimented upon. Named case authors in the validation study were identified exclusively through documented public records.
The due-process framework embedded in this tool reflects the author's commitment to responsible and proportionate use of bibliometric risk indicators. ARI² and JPI² are screening aids — never a verdict. Any use of either index that bypasses the relevant due-process checklist, stigmatises researchers or journals without contextual review, or substitutes a reading for expert judgement would be contrary to the intent and spirit of this work.
Use, modify, and distribute freely including commercially. Retain the copyright notice and licence file. State changes made to the original.
Research dataset 28-author pilot cohort, signal values, PPGDI data (Zenodo)
Apache 2.0
Freely reusable for research and validation. Cite the Zenodo repository. Do not present the pilot dataset as an exhaustive or externally validated reference population.
“Kelleni Index” — Name & Methodology Notice
“Kelleni Index” is used as the project name identifying the original ARI² index, its scoring methodology, and this implementation. It is not a registered trademark, but it identifies an original research contribution and is used here to distinguish this work from adapted or derived implementations.
Adapted or modified implementations of ARI² are permitted under the Apache 2.0 licence and encouraged for further research. However, they must clearly indicate any changes made to the original scoring rules, weights, or normalization parameters, and must not imply endorsement by the original author, official affiliation with this project, or methodological equivalence with the Kelleni Index unless the original scoring rules are preserved in full.
In practice: if you change the signal weights, add signals, or re-normalize against a different reference cohort, your implementation is a derivative — please name it accordingly and link back to the original.
How to cite
Kelleni MT. Commercial authorship and content fraud leave distinct bibliometric signatures: a retrospective feasibility study of the Author Research Integrity Risk Index (ARI²). Zenodo (preprint), 2026. DOI: 10.5281/zenodo.20451926Manuscript Under Review
✓Use the calculator freely for editorial screening, research, or education without charge or registration.
✓Reproduce excerpts of the website text in papers, reports, or teaching materials with attribution (CC BY 4.0).
✓Fork, adapt, or re-implement the scoring code for your own research under Apache 2.0, provided you state what you changed.
✓Build derivative tools that extend or modify ARI², provided the derivative is clearly named and attributed.
✗Present a modified scoring implementation as the Kelleni Index without preserving the original signal weights and normalization.
✗Imply that your adapted version is endorsed by, affiliated with, or equivalent to the original methodology unless it is.
✗Use ARI² scores to take adverse action against any author without independent expert review and a full due-process procedure.
✗Commercialise the Zenodo dataset or scoring code in a way that removes attribution or misrepresents the source.
Questions about reuse? Contact Dr. Kelleni at drthabetpharm@yahoo.com before publishing any adaptation. Attribution disputes or apparent misuse may also be reported to the same address.
Coming Soon · Preview
API-Level PPGDI Computation
Currently PPGDI requires manual entry of unique co-author country counts. The next planned upgrade will compute this automatically from OpenAlex institution metadata — the same API already used for Signal A.
Current approach
Manual entry of unique co-author countries. Relies on the user’s own count, which may involve fuzzy matching and unresolved affiliations. Reproducibility depends on method consistency.
Planned upgrade
Automatic extraction of institutions.country_code from each work’s authorship records during the existing OpenAlex fetch. Fully reproducible, no manual counting, consistent with the manuscript methodology.
Technical note: Implementation extends the existing paginated works-fetch loop to collect work.authorships[].institutions[].country_code for all co-authors, then computes unique country codes ÷ unique co-author IDs. No additional API calls required — runs within the same OpenAlex request already used for Signal A.
Transparency · Live Record
Publication Status Tracker
A transparent record of each instrument's publication journey — from preprint through peer review to final publication. Completed milestones are shown below with their DOIs; pending stages update as the work progresses.
How this is maintained: This tracker is updated as each manuscript progresses, so editors and readers can see the current peer-review status, revision history, and final publication DOI in one place — maintaining complete transparency about each tool's scientific validation status.
