In depth

The machinery behind the licenses, and an honest account of its limits.

---

The front page gives the short version: licenses that forbid nothing and require the operator of a people-facing system to state their principle in public. This page is for readers who want to know how the texts are built, whether the mechanism can actually do anything, and where the family goes next. It assumes no philosophy, but it doesn't hide any either.

1The condition, precisely

Every license here turns on one defined term: Consequential Deployment — the moment a larger system containing the covered code is operated upon people. Scoring them, ranking them, watching them, deciding about them. Up to that moment the code is just a library, and the license asks nothing beyond ordinary provenance. A date parser does not need a maxim.

At that moment, the operator — whoever actually points the system at people, not every upstream maintainer of every dependency — must publish a declaration. It has to be public, discoverable, in plain language, and it has to state the actual principle of the deployment, not a lawyer's paraphrase of one. The declaration must be renewed at each material change in how the system bears upon the people it affects.

The licensor never approves, rejects, or grades the declaration. This is written into every text as its Section 6, and it matters more than it looks. The Kantian test is internal: a maxim that destroys itself when universalized fails on its own terms, and no committee is needed to say so. The public record exists so that the people the software bears upon — who know better than anyone what it does to them — can judge, quote, and object. My opinion of your deployment is not part of the mechanism.

The choice of tradition is not cosmetic. The VPL asks a question the KPL cannot: what does running this system do to the person running it?¹ The UPL concedes up front that its own test cannot be verified from outside, and demands the reckoning anyway. Pick the tradition whose failure mode you can live with.

2Anatomy of the texts

All six texts share one skeleton; only the instrument in Section 3 changes. If you read a single one, read the KPL — the others are diffs against it, and the sameness is no longer a promise kept by hand: all six are generated from a single template, so outside the intended deltas they cannot drift apart.

Section 2 — Provenance

The light obligation. Distribute or operate a larger work: keep the license, make the covered work discoverable. Nothing else. A library is conveyed as a library.

Section 3 — The weighty obligation

Triggered by Consequential Deployment, and again by each material change. The scope is the deployment taken as a whole — the principle on which you operate this system upon these persons — never the maxim of some component inside it.

Section 3.5 — The Three Tests (KPL and LKPL only)

Three answers the Kantian declaration must contain. Self-exception: does the benefit of the practice depend on others in your position refraining from it? Role inversion: would you accept the deployment from the receiving end — subject to the system rather than operating it? The ablest opponent: the strongest objection to your own maxim, stated as an opponent would state it, answered or conceded. None of the answers is graded, in keeping with Section 6 — but each must exist, and a declaration that dodges them is incomplete on its face. The other traditions decline the section on their own grounds: the VPL's preamble refuses box-ticking tests outright, and the UPL already carries its universalization analogue inside Section 3.3, which reckons the practice "were it done by all in Your position".

Section 4 — Propagation

Semantic, not topological. The obligation attaches to whoever effects the deployment, even if the covered code sits five dependencies deep; it does not attach to contributors and redistributors who deploy nothing. In the lesser variants this section also carries the copyleft valve.

Section 5 — Termination

Rights terminate automatically upon deployment without a published declaration, or upon knowing misrepresentation in one. They come back upon publication of a truthful declaration and cure of the breach — unless terminated in writing for repeated or willful violation.

Section 6 — No policing by the licensor

Described above. The reckoning is the operator's to make and to own.

One more thing the texts say about themselves, and I want it said here too: these are free-software licenses, but they are not open source and could not be certified as such. Their obligations extend to a larger work and are triggered by run-time use, which fails criterion 6 of the Open Source Definition and then some. That is a choice, made with the trade-offs in view, not a misunderstanding of the OSD.

3Will this work?

Honest answer: as an enforcement instrument, mostly no. The interesting question is what "work" should mean here, so let me argue against the project before arguing for it.

The obvious objection — publish one maxim, do another — turns out to be the one scenario the mechanism handles well. Knowing misrepresentation terminates the license automatically, and from that point the operator's problem is not "ethics" but copyright infringement, which courts enforce all the time. Better: the published maxim is a document. Privacy-policy law works exactly this way — regulators rarely punish the practice, they punish the gap between the practice and the published text. The license manufactures the text.

The deeper hole is the honest villain. A deportation-logistics operator can publish "the state enforces its own borders" and affirm its universalizability with a straight face. This is the tailoring problem, as old as the categorical imperative itself: describe any act abstractly enough and it universalizes.² The license inherits the problem whole. It filters unspeakable evil, not evil as such — and while a surprising share of the surveillance economy does depend on its operating principle never being said in plain language, not all of it does.

And the realistic corporate response is neither compliance nor fraud but avoidance. This is what happened to the AGPL: large companies obey it by banning it. Code under these licenses will land on dependency blocklists, which cuts against adoption. I have made my peace with that; a license can work as a repellent, a lever for an engineer inside a company who needs a legitimate hook to ask what exactly is being deployed, and a genre of public record that makes stating one's principle a normal thing to do. Kant's publicity principle from Perpetual Peace is the actual bet: maxims that cannot survive being spoken aloud are unjust by that very fact. The license doesn't test anyone's morals. It revokes the privilege of silence.

4The registry: judgment without a judge

Section 6 of every text forswears the licensor's judgment, and Section 3 hands the declaration to "the public" — which, in practice, means to no one in particular. An unstructured public judges the way a comment thread judges: loudly, briefly, and off the point. Between the committee I refuse to convene and the feed I refuse to trust there is a third instrument, and academia has been running it for three centuries: peer review — minus the part where the reviewers accept or reject.

The sketch. A public registry of declarations: one entry per declaration, a plain format, full revision history, so that a deployment's stated principle can be watched as it changes. Reviews are signed, and an objection must name its kind: self-exception (the practice depends on others abstaining from it), contradiction in the will, wrong altitude (a maxim of some component, not of the deployment), divergence (the declaration does not match the system's observed behavior). The KPL's Three Tests (Section 3.5) hand reviewers their rubric — each test is a question the declaration must already have answered, and a review can say, concretely, that it has not; for the other traditions, the content requirements of Section 3.3 play the same role. Divergence is the sharp kind: a documented gap between the stated principle and the observed practice is the raw material of a Section 5 termination — knowing misrepresentation — and no reviewer has to adjudicate it. Publication is enough.

What the registry must never do is issue verdicts. No pass and no fail, no seal, no badge — counts, not conclusions: 4 reviews · 2 objections standing · last reply 2026-05-12. The moment a review becomes a gate, the committee has been rebuilt with extra steps, and Section 6 binds the registry's design as it binds me. The operator answers, revises, or ignores — in public; the record accumulates either way. One honesty is owed up front: whoever hosts the registry keeps residual power through moderation, so the moderation rule must be published and narrow — spam and doxxing are removed, arguments are not.

None of this is built. It can start embarrassingly small: a git repository, where a declaration is a file and a review is an issue. If the licenses are the constitution, the registry is the newspaper — and anyone can run one, which is the point.

5Traditions not yet drafted

The skeleton generalizes: any tradition that can say what an operator ought to be able to state publicly can be a license. Four are sketched but unwritten. Sketches of their Section 3 instruments exist in my notes; if one of these is yours, write it.

6Neighbors

These licenses stand on a decade of ethical-source experiments and diverge from all of them in restricting nothing by content.

The Hippocratic License is the most developed of the prohibition family: it forbids uses violating internationally recognized human rights and anchors that condition to external, verifiable standards. That anchoring is its best idea and precisely the road not taken here — an external standard is replaced by a public procedure. The JSON License ("The Software shall be used for Good, not Evil") is the genre's ancestor and a proof of the difficulty: an ethical condition with no procedure and no test. The KPL is, in a sense, the JSON License with Kant supplying the missing test.

Also in the neighborhood: the Non-Violent Public License and its cooperative variant, the Just World License, the Anti-Capitalist Software License, and the Organization for Ethical Source, which keeps the catalogue and carries the argument with the OSD.

7Reading

Primary texts, all free: Kant, Groundwork of the Metaphysics of Morals (Gutenberg, Abbott translation); Aristotle, Nicomachean Ethics (MIT Classics, Ross); Mill, Utilitarianism (Gutenberg). For orientation, the Stanford Encyclopedia entries on Kant's moral philosophy, virtue ethics, and consequentialism cover the machinery the three licenses borrow. MacIntyre's After Virtue is the book behind the VPL's suspicion of rule-vocabularies. The wider license landscape is catalogued at ethicalsource.dev.

---