Human Emacs Drama Manifesto and Internal Inconsistencies

Human Emacs Drama Manifesto and Internal Inconsistencies

The text argues that the Human Emacs manifesto is internally inconsistent: it forbids the arguments it relies on, imposes an unverified attestation rule, claims to respect GNU while pre-committing to reject its decisions, invokes user freedom while restricting contributions, lacks clear standing or operational power, and makes unverified claims about projects like Vim, Linux, and rsync.

1. It forbids the argument it is built from

The document asks people to join a conclusion while pre-cancelling the three lines of reasoning that could contradict it — effectiveness, open-weight models, provenance good faith [cite:1]. Then, two paragraphs later, it says “Discussion takes place on the mailing list” [cite:1]. A manifesto whose operative section is titled “Not Under Discussion” cannot usefully discuss; the signatory mechanism is the substitute for argument, and signing by name is how authority is accumulated where premises are forbidden.

It also declines the burden of proof it imposes on others: the harms are “clear enough that it’s not necessary to enumerate them afresh here” [cite:1]. A statement about the provenance of text offers no provenance for its own claims, no data for the “betrayals” narrative, and no citation for the assertion that Vim, Linux and rsync “aggressively adopted LLM-driven development” — the word “aggressively” carries the accusation, and no exhibit is attached.

2. It declares attestation-based rules unworkable, then proposes one

The policy is enforceable only through contributor statements about how code was produced. The document acknowledges this and dismisses it as an existing, unfixable risk — bad-faith contributors have always been able to lie, “it is enough to treat LLM-generated patches the same as other forms of plagiarism” [cite:1]. But that concession dissolves its own policy: a rule whose only implementation is an unverifiable attestation, defended as no worse than plagiarism, is not a policy with teeth. The projects it scolds handle this precisely by attestation. The kernel’s published patch process requires a Signed-off-by: under the DCO — which certifies “the contribution was created in whole or in part by me” — and requires an Assisted-by: tag for any advanced coding tool, with failure to acknowledge “may impede the acceptance of your work” 1. The kernel’s position is disclosure-plus-responsibility; Human Emacs calls the same class of practice betrayal while admitting its alternative cannot be verified. It has no standard that survives contact with its own paragraph.

3. It claims the high ground GNU historically held, then pre-commits to rejecting whatever GNU decides

The document is unhappy that GNU’s hesitation has centered on copyright [cite:1]. It then declares intent to act “regardless of whatever policy GNU decides upon”, inviting GNU to “do the right thing, in which case we will continue to support them” [cite:1]. The invitation is decorative: the outcome is fixed in advance, so GNU’s decision can only confirm or invalidate a conclusion already reached. That is the definition of bad faith in a process you claim to be participating in — and Stallman’s reply in the same thread, asking people to wait for the published policy “before stirring up disputes about hypothetical options” (https://lists.gnu.org/archive/html/emacs-devel/2026-03/msg00554.html), is on record as an objection to exactly that posture.

The asymmetry sharpens on substance. GNU’s actual standard is binary freedom, not provenance: the FSF is working on a criterion for judging whether a machine learning system is free/libre, and refuses “levels” of freedom — “something is either free/libre or it is not” (https://lists.gnu.org/archive/html/emacs-devel/2026-02/msg01047.html). On that criterion, GNU already accepted a specific model as listable — Apertus, built on data collected ethically with reproducible training — as acceptable for Emacs and ELPA packages (https://lists.gnu.org/archive/html/emacs-devel/2026-02/msg01047.html). Human Emacs rules out that entire class as “science fiction” and requires data collected with consent [cite:1]. Its standard is not stricter than GNU’s on ethics; it is stricter on a definition no model can ever satisfy, because the qualifying condition is the historical method of production, which is not one of the four freedoms and is not revocable by any future act. A standard that nothing in principle can meet is a purity test, not an ethical one.

4. It invokes user empowerment to restrict what users may receive

The opening sentence frames the project as empowering users [cite:1]. Freedom 1 and 3 are the freedoms to make and distribute modified versions, and GNU’s own position treats functionality as irrelevant to freedom — a program that “arbitrarily rejects certain meaningful inputs” is less useful but still free, and users can overcome that through freedoms 1 and 3 [cite:af5f57a7-7]… more precisely as stated in [cite:a2458422-7]. A permanent rule barring a category of contributions does exactly what freedom 1 and 3 protect against: it removes code users would otherwise be free to include. So the empowerment premise and the rule are opposed, and the rule must be imposed downstream — which the GPL does not permit: the GPL relaxes restrictions and requires only that distributors keep the distribution terms unchanged; adding restrictions to a GPL-covered program changes the terms and can render modified versions nonfree [cite:af5f57a7-2]. A fork that inherits GPL code cannot add an enforceable restriction to it; it can only have a project policy, which the document is right that GNU could adopt — but a policy is not a license term, and the fork’s own authority over its code would still rest on the same attestation problem.

5. Its exemplar conceded what it will not

The motivation section holds up the Guix consensus document “Standing up for human crafting” while noting that Guix “did not have the foresight to place a temporary ban” and therefore “felt forced to leave some unfortunate exceptions in their proposal” [cite:1]. The sentence implicitly accepts that principled prohibitions require exceptions. The Human Emacs document has no exceptions, no carve-outs, no discussion, and no mechanism — so it holds itself to a stricter standard than the authority it cites, and criticizes that authority for the weakness that the document’s own logic produces.

Also worth noting: the Guix page did not load at all — Codeberg returned an anti-scraper block [cite:1]. A document whose cited evidence base is a human-crafting argument cannot be read through a machine, and the Amnesty citation resolves to a landing page, not a text [cite:1]. The factual scaffolding is presented as self-evident rather than checkable.

6. The fork depends entirely on people who never agreed to it

The fork argument rests on hundreds of contributors who “never at any time did all those people decide to work together”, and GNU’s own account of forks is that credit for a variant cannot be divided without arbitrariness [cite:09bde7e3-1]. Human Emacs relies on a code base — and on the reputational cache of a project — built by people who did not sign, on a document whose copyright is claimed by a single named person plus “contributors” [cite:1], while the signatories are called signatories, not contributors. Under the FSF’s own approach, contributions from named people create rights held by those people, and assignment exists precisely because attribution and control are contested [cite:a2458422-8]. A document that credits no signature as authorship while asserting copyright over their names is inconsistent with the provenance ethic it declares.

The same asymmetry runs through the GNU communications guidelines those signatories are contributors to: assume participants post in good faith, accept code presented as their own work, be kind about mistakes, and welcome contributions from people who use nonfree software [cite:09bde7e3-3]. The document suspends exactly that presumption in advance for hypothetical contributors and closes off the discussion it asks others to respect.

7. It has no standing, and no operational content

No member of the group that decides GNU Emacs policy appears among the names — no Zaretskii, Monnier, Kaludercic, Radchenko, Stallman [cite:1], and none of the people who answered the question on the list [cite:2]. So the document claims authority to define the ethics of a project it has no mandate over, and it already has the practical outcome it wants — the ban is in force today [cite:2]. The manifesto therefore changes nothing operationally; its content is the delegitimization of a process before the process concludes. That is the cleanest hypocrisy: it condemns the way decisions are made, while relying on the decisions it demands, and it praises the current as “glad to see” posture in the same paragraph where it calls that posture inadequate [cite:1].

8. Two unverified assertions worth holding open

Whether Vim, Linux and rsync as described adopted “LLM-driven development” is asserted, not shown — the kernel’s rule is disclosure, not adoption-by-default, and Vim Classic presents itself as long-term maintenance rather than an ethics protest against a betrayal [cite:6], describing a fork “maintained without the assistance of generative AI tools” whose announcement is about stability (https://vim-classic.org/). Comparing a maintenance fork to an ethics declaration is a category slip, and the document uses the slip to imply equivalence between projects.