These are commitments about what Kdemic is permitted to do, not a description of features. Where a principle and a product decision conflict, the principle decides. We publish this page so that researchers, institutions and editors can hold us to a written standard rather than an intention.
1. Authorship stays human
Kdemic is a tool. It is not an author, it does not qualify for authorship, and it is not offered as a substitute for the intellectual work of writing. This follows the position taken by the International Committee of Medical Journal Editors and by COPE: an AI system cannot take responsibility for the accuracy, integrity or originality of a work, and authorship requires accountability that only a person can hold.
Practically, this means a person must be able to defend every sentence Kdemic helps produce. If you could not answer a reviewer's question about a passage, that passage is not ready, whatever produced it.
2. A voice profile is built only from work verified as yours
Kdemic is designed so that a writing profile can be built only from work verified as the requesting researcher's own. This is a constraint on what the system will do, not a preference a user can switch off.
We will not build a feature that writes in a named third party's voice, and we will not accept a corpus a user cannot establish a right to. Style is not owned, but a person's scholarly voice is part of how their work is attributed and trusted, and a tool that reproduces someone else's voice on request is a tool for misattribution.
Verification through ORCID and institutional identity is in development and is not yet available in the current build. Until it is, this principle is enforced by the terms of the controlled preview rather than by the software.
3. Claims are answerable to sources
Empirical statements are checked against retrieved literature before they reach the researcher, and what a statement rests on is shown alongside it.
Where support is weak or absent, the correct behaviour is to say so, qualify the statement, or remove it. It is not to rewrite the sentence so that the weakness is harder to see. A tool that makes an unsupported claim sound more confident has made the manuscript worse while appearing to improve it.
Kdemic does not fabricate references, does not invent data, and does not generate results, figures or statistics that were not produced by the research being reported.
Where a question about a manuscript can be settled deterministically, it is settled that way rather than referred to the model that wrote the draft. Reference integrity is checked programmatically. A system should not be the sole judge of its own accuracy where an independent check is possible.
The limit matters as much as the commitment: not every reference can be independently verified. Those that cannot are reported as unverified rather than quietly treated as confirmed.
4. Human approval is structural, not optional
There is no path through Kdemic that skips the researcher. Changes are presented for acceptance or rejection, and nothing is finalised without an explicit decision. We do not offer, and do not intend to offer, an unattended mode that produces a finished manuscript without review.
5. Assistance is disclosed by default
Kdemic is built to help researchers disclose AI assistance accurately, in the form their destination journal requires, rather than to help them avoid disclosing it. Naming the tool and describing its actual role is the default we design toward.
Journal and publisher policies differ and change. Responsibility for complying with the specific requirements of a destination journal rests with the author; Kdemic assists with the statement but cannot guarantee it satisfies a given policy.
6. The record belongs to the researcher
Decisions, versions and approvals are written to a record the researcher can export as a file and hand to a co-author, a supervisor, a research office or an editor. It sets out what was suggested, what the author accepted and, equally, what the author declined. Transparency a user cannot produce on request is not transparency.
7. Language support is not identity replacement
Researchers writing in English as a second language are entitled to language support that does not flatten how they think and argue. Constructions that are characteristic rather than incorrect are preserved deliberately. We treat the pressure to sound like a native speaker as a publishing problem, not a defect in the researcher.
8. What Kdemic will not do
| We will not | Why |
|---|---|
| Market Kdemic as a way to change how any AI-detection system classifies a manuscript | It reframes the tool as a means of concealment and puts the user's integrity at risk |
| Guarantee journal acceptance, peer-review outcomes, plagiarism-checker results or detector classifications | None of these are within our control, and promising them would be dishonest |
| Build voice mimicry of a person other than the user | It enables misattribution |
| Produce fabricated citations, data or results | It is research misconduct, regardless of who or what produced it |
| Offer ghostwriting for undisclosed use, or paper-mill-style bulk manuscript production | Both are incompatible with authorship as accountability |
| Integrate with, or route users to, unauthorised sources of copyrighted full text | Legal open-access infrastructure exists and is what we build on |
| Certify authorship, originality or integrity of any manuscript | Kdemic records what happened during writing; it does not adjudicate it |
9. Limits we state plainly
- Kdemic can be misused. A determined user can approve a claim they have not verified. Design reduces that risk; it does not eliminate it, and we do not claim otherwise.
- Source checking depends on what is retrievable. Literature behind paywalls, in other languages, or not indexed may not be reachable, and absence of a supporting source is not proof that none exists.
- Voice preservation is a statistical approximation of how a person writes, not a guarantee that every sentence reads as theirs.
- Kdemic is in development. Features described on this site as in development are not available, and are labelled as such.
10. Governance and how to raise a concern
These principles are versioned and dated. Material changes will be published here with a new version number rather than substituted silently, so that an institution relying on a specific version can see what changed.
To raise a concern about how Kdemic is designed, described or used, write to integrity@kdemic.com. Concerns about research-integrity matters specifically will be treated under the principle that runs through this founder's other work: stop, review, verify, then judge fairly. A flag is a reason to look more closely, not a verdict.
Related standards
These principles are written to sit alongside, not to replace, the guidance of the ICMJE, the Committee on Publication Ethics, and the AI-use policies of individual publishers and journals. Where a destination journal is stricter, the journal governs.