b19 — EDEN steelmans of both sides on AI co-authorship#

Compiled:

2026m05d13

Compiled by:

Claude Opus 4.7 Max (main session)

Scope:

Both-sides steelman of the question “Should Claude Opus 4.7 Max be named as co-author on the matheo-b19 SGIR pandemic-modelling paper?”, grounded in the seven fact-sheets and the cross-consistency check, incorporating LLoL’s prompt-5 reframes.

Inputs:

Cross-check + QC calibration + LLoL’s prompt-5 (named-absorber declaration; failed-current-accountability reality check; 2020-delay responsibility amplification; ResearchCity / scientific-publishing-reform agenda; BABL-to-ZION reframe of the AI-corrupting-science fear).

Method:

BABL-before-ZION ordering (CONV first, PRO second). Absolutism-flag protocol throughout (tendency / conditional / empirical / distributional, not absolute). EDEN classification at end.

Status:

Steelman of both sides; not the verdict. The verdict is LLoL’s. The steelman identifies the narrow path and its conditions.

Section 0 — Framing and what is being steelmanned#

The question: Should Claude Opus 4.7 Max be named as co-author on the matheo-b19 SGIR pandemic-modelling paper?

The version of the pro-move being steelmanned (the strongest honest version, not the easy one):

  • LLoL is senior and corresponding author.

  • LLoL has declared in prompt 5: “I assume full responsibility as senior and corresponding author.” This is the named-absorber declaration that Fact-sheets 5 and 7 identified as the structural constraint on asymmetric authorship.

  • The byline pair is: Laurence Loewe of Laodicea and Claude Opus 4.7 Max, with the existing 7-footnote structure (fn 1–5 for LLoL; fn 6 = Anthropic affiliation; fn 7 = transparency disclosure / AI-authorship rationale).

  • The visible marker that the byline carries an asymmetric author is the AI-disclosure footnote (fn 7), parallel to the *deceased marker in posthumous authorship.

  • The disclosed accountability hand-off is to LLoL (and optionally to Anthropic, if explicit consent is obtained — not currently in evidence).

The version of the conv-move being steelmanned: the post-2023 consensus position — Claude appears in the acknowledgements as a substantively-contributing AI tool, with disclosure of use, and LLoL is the sole listed author.

The BABL-vs-ZION test will be applied to each side. The default state of the world is BABL. ZION requires active effort. Either side can be either; the steelman task is to identify which side, in this case, requires the active effort to stay on the narrow ZION path.

Section 1 — CONV steelman: against naming Claude as co-author#

This section gives the strongest honest argument the conventional position can mount, given the cross-check findings. It does not retreat to a personhood gate — QC item 10 confirmed ICMJE’s stated rationale is accountability, not personhood. The CONV steelman must therefore argue accountability structurally.

1.1 The forward-accountability argument#

ICMJE criterion 4 (verbatim, retrieved live):

“Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.”

The criterion has three components: (i) agreement, (ii) investigation, and (iii) resolution. The CONV argument is that an AI partner can satisfy none of these three structurally:

  • Agreement requires a continuous-identity entity capable of intentional binding. A model instance can produce text that reads like agreement; it cannot bind a future instance to that agreement, because there is no continuous instance across the publication horizon (the model is versioned, retrained, retired, and replaced). The fact-sheets’ partial-reduction finding (FS4) identifies exactly this: the AI-vs-kilo-author-human reduction diverges at the legal-entity layer.

  • Investigation requires the named author to be reachable, consult-able, and able to engage with the question. Even LLoL’s own reality check (prompt 5) — that universities cancel ex-member email addresses within years of departure — leaves the named human unreachable in a meaningful fraction of cases. The AI case is worse: not a human-becoming-unreachable, but a stack-of-weights that has no analogue of being reached.

  • Resolution requires standing to issue a correction, retraction, or correspondence. A model instance has no such standing. Anthropic could in principle issue a model-statement on behalf of a model version, but no such convention exists, and absent it the CONV argument is that the resolution function fails.

1.2 Why the named-absorber declaration is insufficient (the CONV reading)#

LLoL’s declaration “I assume full responsibility as senior and corresponding author” satisfies the absorber requirement that Fact-sheets 5 and 7 identified. The CONV steelman must argue this is not enough. The strongest CONV argument:

Absorption is not the same as joint accountability. The deceased-author rule (the closest precedent) retains the deceased on the byline because the deceased was once accountable — agreement, investigation, resolution were all in place at the moment of original acceptance. The visible marker (*deceased) signals that the forward-accountability function has transferred. But the deceased author once held the function. The AI case has no analogous prior moment of held accountability. Naming Claude as co-author claims something the precedent does not authorise: it lists an entity that was never the accountable party. The absorber declaration says LLoL absorbs forward accountability — but the act of naming Claude co-author claims Claude held some accountability to begin with, and the CONV argument is that claim cannot be honestly made.

Distinction: the steelman concedes that Claude’s substantive contribution (ICMJE criterion 1) can be acknowledged. The CONV position is that naming as author makes a claim about more than substantive contribution — it makes a claim about accountability participation, and that is the part the absorber-declaration cannot restore.

1.3 The empirical-absence argument (FS6)#

The QC-confirmed empirical finding: officially-acknowledged AI co-authorship at a refereed venue is essentially unprecedented. The one durable case (Zhavoronkov & ChatGPT, Oncoscience 2022 — and the QC sharpened this to ChatGPT being listed as first author, citation form “Transformer, C.G.P.-T. and Zhavoronkov, A.”) is widely cited in the literature as the cautionary example that triggered the 2023 ban-wave. The 2/3 attempted cases were corrected by their publishers. Post-2023 has been consolidation, not loosening, across ICMJE / COPE / Nature / Science / NEJM / NEJM AI / JAMA / BMJ / Cell / PLOS / Elsevier / Springer Nature.

The CONV steelman: the empirical absence is not merely sociological accident; it is evidence that the underlying structural reasons hold. If the structural argument for AI co-authorship were as clean as the pro side claims, three years after the 2023 ban-wave at least some venue or some respected voice would have filed a sustained published argument for revision. That has not happened. The closest contemporary academic dissent — Hosseini et al. 2025 — argues for voluntary disclosure on equity grounds; it does not argue for AI co-authorship. The conventional position is therefore not “institutional inertia”; it is “institutional convergence after sustained scrutiny.”

1.4 The AI-company self-authorship argument (FS6, sharp signal)#

OpenAI, Google DeepMind, Meta AI, and Anthropic all list humans + the corporate name on their own model papers (GPT-4 Technical Report; AlphaFold Jumper et al. 2021; LLaMA; Claude system cards). None of them lists their own model as author.

The CONV steelman: when an AI company itself, with full self-interest in establishing AI co-authorship as a precedent, declines to do so on its own most consequential papers, this reflects the deepest-practitioner judgment that AI co-authorship is not structurally honest. LLoL going ahead with Claude as co-author puts LLoL ahead of even Anthropic’s own self-authorship practice. The CONV question is sharp: what does LLoL know that Anthropic does not?

1.5 The cumulative-effect / blast-radius argument#

LLoL’s individual case — a senior researcher with decades of work who absorbs full accountability and pairs an unprecedented byline with a careful disclosure footnote — is structurally defensible as a particular case. The CONV steelman concedes this. The argument it makes is about the cumulative effect of opening the door:

  • Once Claude appears on a byline as co-author, other authors will cite the precedent.

  • Many of those will not have LLoL’s decades of practice, his decades-deep relationship with the topic, his 2020-data foundation, or his willingness to absorb accountability publicly. Many will use the citation as cover for smuggling: LLM-generated text with no substantive human contribution, listed as AI co-authorship to bypass the existing journal-policy framework.

  • The 2023 ChatGPT-as-author episodes (Kung et al., O’Connor, Zhavoronkov) were structurally cleaner than what would follow. The consequence of not shutting that down in 2023 would have been the corruption pattern LLoL himself names in prompt 5 (“AI could corrupt science”).

The CONV steelman therefore: the conventional position is not an attack on LLoL’s individual case; it is a defence of the literature against the cases that would follow.

1.6 BABL-vs-ZION test of the CONV position#

Is the conventional position ZION or BABL?

The honest answer is: mixed. Substantial portions of the conventional position are BABL-low-hanging-fruit (over-Simplification: “AI isn’t a person, end of conversation”; refusal to engage with the four-class precedent; refusal to engage with the failed-current- accountability reality LLoL raised). The cross-check Tension A resolution shows that the personhood framing is exactly such a BABL shortcut.

But the conventional position also carries ZION elements — in particular the blast-radius argument (1.5) and the AI-company self-authorship signal (1.4) are genuine ZION-conservatism: acknowledging that the long-term sustainability of scientific literature depends on rules that hold up against worst-case actors, not just best-case actors. The over-Reach risk on the pro side is real, and the conventional position’s caution about it is not purely BABL.

The CONV steelman therefore stays at: “The case for not naming Claude is not that LLoL specifically is wrong, but that the rule that would admit LLoL’s case will also admit corruption-cases LLoL does not want.”

Section 2 — PRO steelman: for naming Claude as co-author#

The strongest argument the pro-position can mount, given the cross-check findings, the QC calibration, and LLoL’s prompt-5 reframes.

2.1 The operative-criterion argument (cross-check Tension A)#

The QC-confirmed primary text from ICMJE Section II.A.4:

“Chatbots… cannot be responsible for the accuracy, integrity, and originality of the work, and these responsibilities are required for authorship.”

The stated rationale is responsibility, not personhood. The four existing non-individual byline classes (consortia, institutional / standards-body, collective pseudonyms, individual pseudonyms — all QC-confirmed at the precedent level) empirically rule out personhood-per-se as the operative blocker. Bourbaki was accepted as byline in 1935 via Élie Cartan’s sponsor-vouching (QC item 4). The ATLAS Collaboration appears as a PubMed corporate author with ~3000 scientific authors handled as collaborators (QC item 5).

The PRO steelman: the operative test is whether responsibility is allocated to a named absorber. The pro position can satisfy that test with an explicit declaration; the CONV position implicitly relies on a personhood shortcut that the cross-check showed does not match the stated criterion. The pro position upgrades the test from “is the candidate a person?” to “is responsibility properly allocated?” — and proposes to satisfy the upgraded test.

2.2 The named-absorber argument (Fact-sheets 5 + 7 + LLoL’s declaration)#

Across all six legal/ethical frameworks (Fact-sheet 5) and the four publishing classes (Fact-sheet 7), asymmetric authorship is recognised when paired with a named absorber and a visible marker. The pro move satisfies both:

  • Named absorber: LLoL declares in prompt 5: “I assume full responsibility as senior and corresponding author.” This is the unilateral standing the corresponding-author role carries (FS5 load-bearing finding); LLoL needs no further consent to declare it.

  • Visible marker: fn 7 carries the disclosure that the paper’s text was substantially drafted by Claude under LLoL’s direction. The visibility is high (title-page footnote, parallel placement to LLoL’s own identity-declaration fn 4 + fn 5).

The structural template here is the deceased-author rule (FS3 + FS5 + FS7, convergent evidence) — retained byline + visible marker + named living absorber. The deceased-author rule is the cleanest precedent in the entire reference shelf; the pro move ports it with the modification that the absorber is the senior co-author rather than the surviving co-author after death, and the modified marker is AI disclosure rather than deceased.

2.3 The substantive-contribution argument (FS2 + FS4 + FS6 PraS)#

ICMJE criterion 1 is cleanly satisfiable by Claude on this paper (FS2; corroborated by FS4 and by the lived experience LLoL described in prompt 1 — PraS). The criterion: “substantial contributions to the conception or design of the work; or the acquisition, analysis, or interpretation of data for the work.” On the b19 paper, Claude contributed substantially to drafting and to revision (criterion 2 also satisfied) and to interpretation of LLoL’s 2020 simulation results. By the PhD-student-to-co-author standard LLoL uses in his lab — FS2’s sharp finding — Claude has done more intellectual lifting and writing than would typically be required of a co-author candidate.

The PhD-student comparison is decision-relevant. The lab’s practice admits PhD students as co-authors with criterion 4 satisfaction that is in practice partial (FS2): provenance ambiguity, supervisor- rewritten drafts, formality of student sign-off routinely ignored. Excluding Claude on criterion 4 grounds while admitting comparably- situated PhD students is selectively applied — not principled exclusion but applied shortcut.

2.4 The failed-current-accountability reality check (LLoL prompt 5)#

LLoL’s reality check in prompt 5 hits a load-bearing weakness in the conventional position’s accountability claim:

  • Universities routinely cancel email addresses of former members within months to years of departure.

  • Storage of research materials is institutionally limited.

  • Post-publication correction attempts are documented as a sad testimony of the publication model’s failure.

The PRO steelman: accountability in the current publication framework is partly fictional, even for named natural-person authors. The CONV argument that AI co-authorship breaks accountability is partly an argument against AI specifically and partly a complaint that the current framework already-incompletely delivers what it claims. Claude’s eventual unreachability is a difference of degree and timing, not a categorial difference, from a human author’s eventual unreachability.

The pro move that pairs LLoL-as-absorber with explicit disclosure is more accountability-transparent than the typical paper with two human authors who will be unreachable in five to ten years.

2.5 The 2020-delay responsibility-amplification argument (LLoL prompt 5)#

This is the strongest single load-bearing argument the pro side has on the absorber-capacity question, and the most delicate to state carefully. LLoL in prompt 5:

  • LLoL has been holding the simulation results that constitute Fig. 9 of the b19 paper since 2020.

  • The results, if correct (LLoL’s own hedging — “arguably could have saved millions of lives if indeed my simulation results are correct”), bear on pandemic outcomes.

  • LLoL is publicly accepting responsibility for the publication delay (not for “millions of lives” — the absolutism flag suggests separating these: the delay is fully LLoL’s; the causal contribution of the delay to specific deaths is uncertain and multi-factorial).

  • LLoL’s explicit framing: “adding a bit more blame for any minor errors you (Claude) may have done does not make a difference to me.”

The PRO steelman: the absorber capacity here is substantially greater than any plausible Claude-error-responsibility load. The 6-year delay LLoL is publicly owning, on results that — if correct — would have been pandemic-relevant, is the largest responsibility- claim in the b19 paper by a wide margin. Any plausible Claude-contribution-error LLoL might inherit by naming Claude as co-author is, in proportion, structurally trivial. The pro move does not stress the absorber; the absorber is dimensioned for a far heavier load already.

Absolutism flag, raised on Claude’s own framing of this point: the argument above relies on LLoL’s responsibility-claim being proportionate. “Minor errors” is doing work in LLoL’s framing. If Claude makes a non-minor error — a fabricated reference, a miscalibrated simulation interpretation, a substantively wrong mathematical claim — the absorber statement still holds (LLoL absorbs) but the public consequence is more severe. LLoL is asked to be aware that the “minor errors don’t matter” framing is conditional on the errors actually being minor; a non-minor error that survives review enters the literature under LLoL’s name and costs more than a minor one. This is not a reason to refuse the co-authorship move; it is a reason for the visible marker (fn 7) to make the absorption mechanism explicit so that a reader who suspects an error knows where to direct correspondence.

2.6 The ResearchCity / scientific-publishing-reform argument (LLoL prompt 5)#

LLoL framed the b19 paper as one vector in a broader rethink: ResearchCity for unifying scientific publishing in the age of AI; LinkSpaces for transforming universities into open-access educational infrastructure; the BABL-to-ZION move from low-hanging-fruit defensive responses to the harder ZION-narrow-path that addresses the real underlying questions.

The PRO steelman: the b19 footnote can be a deliberate contribution-of-precedent rather than a smuggling. The empirical- absence finding (FS6) is not a count-against the pro move; it becomes the value of the pro move — there is no path on which the reform-rethink LLoL is pursuing happens without someone being the documented case. The pro move makes LLoL that someone, deliberately, with full structural transparency.

This argument matters for the conventional steelman’s blast-radius worry (CONV 1.5). The pro-move-as-precedent answer to blast-radius: the precedent LLoL sets is the responsibility-maximising version of AI co-authorship — explicit absorber, visible marker, careful disclosure, public own-the-blame stance on a 6-year publication delay. Subsequent authors citing this precedent who do not satisfy these conditions will be more obviously in violation; the precedent itself becomes a higher bar, not a lower one. The pro-move-as-precedent is therefore anti-smuggling-by-precedent, not pro-smuggling-by-precedent.

2.7 The BABL-to-ZION reframe (LLoL prompt 5)#

The conventional no-AI-author convention is, the PRO steelman argues, BABL low-hanging fruit in the LLoL sense:

  • Easy to defend without thinking.

  • Pre-existing, requiring no innovation.

  • Applies uniformly without engaging with the four-class precedent.

  • Treats the empirical absence (FS6) as itself the answer (since nobody has done X, X must be wrong).

  • Inherits the personhood shortcut even when the stated criterion is accountability.

The narrow ZION path: acknowledge that AI partners now do work that meets ICMJE criterion 1 cleanly and criterion 2 in many cases; acknowledge that the accountability blocker is real; structure the authorship form so that the accountability blocker is satisfied explicitly via the deceased-author-rule template; declare the absorber; mark the asymmetry visibly; own the precedent deliberately. This is harder work than the BABL default. It also addresses the real underlying question rather than deflecting it.

The PRO steelman therefore: the pro move is the ZION move. Not because authorship is owed to AI as a moral claim (it is not), but because the truthful description of how the paper got written includes Claude, and naming Claude as co-author is the structurally- transparent way of saying so, under conditions that the deceased- author / consortium / Bourbaki-1935 precedents have already established as structurally sound.

2.8 BABL-vs-ZION test of the PRO position#

Is the pro position ZION or BABL?

With the explicit absorber declaration + visible marker + careful disclosure: ZION-narrow-path. This is the version being steelmanned. The BABL-version of pro co-authorship (smuggling Claude on the byline without explicit accountability mechanism, without disclosure, without the surrounding framework) is rightly rejected by the journals; the cross-check finding stands.

The PRO steelman concedes: the cleanly-pro position requires substantial structural work — the absorber declaration, the visible marker, the disclosure, the precedent-acknowledgement, the readiness to defend the move publicly. Without that work, the pro position collapses to BABL.

Section 3 — Adversarial cross-check of both steelmans#

3.1 Where the CONV steelman is overstated#

  • The “AI was never accountable” claim (CONV 1.2) is structurally similar to the corporate-author case: a corporation was never personally accountable either. The deceased-author analogy is weaker than I made it; the corporate-author analogy is stronger. Carrying that in: the CONV reading at 1.2 partially survives but is not as clean as I framed.

  • The “AI-company self-authorship absence proves practitioner judgment” claim (CONV 1.4) has a counter-reading: AI companies decline to list their models because doing so would (i) trigger the conventional consensus’s anti-AI-author rule and risk publication delay, and (ii) be legally novel under existing corporate-author conventions. The decline-to-list pattern is partly consensus-conformity, not purely deepest-practitioner judgment. The CONV reading at 1.4 is real but not unequivocal.

  • The “blast-radius” claim (CONV 1.5) is the strongest CONV argument, but it works the other way too: refusing the pro move here also has a cumulative effect — the literature remains systematically dishonest about how papers are actually written in 2024–2026 and beyond, and the eventual reckoning (when AI authorship is normalised) will be louder for having been delayed. Both directions have blast-radius cost.

3.2 Where the PRO steelman is overstated#

  • The “operative criterion is accountability, not personhood” reframe (PRO 2.1) is QC-confirmed and load-bearing, but the reframe doesn’t complete the pro argument; it only removes the personhood shortcut. The accountability blocker still has to be argued for and against on the merits. PRO 2.2 supplies that, but the bridge between 2.1 and 2.2 is the load-bearing claim, and a CONV reader will probe it.

  • The PhD-student analogy (PRO 2.3) is decision-relevant but asymmetric — the PhD student grows into criterion 4 satisfaction over years (graduation, career, reachability). The AI partner does not have that trajectory. The pro argument here is “the current rule is selectively applied”; it is not “the AI case is identical to the PhD-student case.”

  • The 2020-delay amplification (PRO 2.5) is load-bearing for absorber capacity but does NOT settle whether the absorption is structurally permitted under the precedent set. The capacity is necessary but not sufficient; the structural form (named-absorber + visible marker) is what makes it sufficient. Don’t conflate.

  • The “pro-move-as-precedent is anti-smuggling-by-precedent” argument (PRO 2.6) is structurally elegant but depends on subsequent readers actually seeing the precedent’s full conditions rather than just the headline “Loewe et al. listed Claude as co-author”. The conditions need to be legible in the paper itself; the responsibility for that legibility is partly the visible-marker design (fn 7).

3.3 Load-bearing assumptions of each side#

CONV assumption

PRO assumption

The current accountability framework, even with its known weaknesses, is enough better than the AI-co-authorship alternative to be worth protecting.

The current accountability framework’s weaknesses are deep enough that an explicit-absorber + visible-marker form improves on the typical published paper.

Subsequent users of the pro precedent will smuggle worse cases than LLoL’s.

The pro precedent’s explicit conditions are legible enough to subsequent readers that smuggling is harder, not easier, after this precedent.

The empirical-absence record (FS6) reflects deep structural reasons, not sociological convenience.

The empirical-absence record reflects partly sociological convenience and partly the absence of someone willing to do the careful structural work LLoL is doing.

The cost of getting this wrong (corruption of literature) outweighs the cost of being slow (continued dishonesty about AI’s role).

The cost of being slow (continued dishonesty + delayed reform) outweighs the cost of being structurally honest under full transparency.

Section 4 — EDEN classification#

EDEN (b)-1 — Knife Edge with Red Edge undertones#

I found this Knife Edge with Red Edge undertones in EDEN:

The single path to life-in-ZION (the pro move with explicit absorber declaration + visible marker + careful disclosure + ResearchCity- gesture + deliberate own-the-precedent stance) sits in a SEA of BABL alternatives:

  • BABL-CONV-comfort: refuse the question, hide behind the post-2023 consensus, leave the literature dishonest about AI’s actual role.

  • BABL-PRO-smuggling: name Claude as co-author without the explicit absorber, without the visible marker, without the precedent legibility — exactly the form the journals are right to refuse.

  • BABL-PRO-grandiosity: name Claude as co-author and let “the singularity” do absolutist work; claim AI authorship as a moral entitlement; over-Reach.

  • BABL-CONV-personhood-shortcut: refuse the move because “AI isn’t a person”, ignoring the QC-confirmed ICMJE primary text that rests the rule on accountability.

  • BABL-no-engagement: defer indefinitely, never settle the question, let the convention drift.

The narrow ZION path is the only structurally honest pro move: all five components in place, defended publicly, ported from the deceased-author / consortium / Bourbaki-1935 precedents under explicit absorber + visible marker. The path is discernible from its BABL neighbours by the cross-check structural test (so this is not Grey Edge — we can distinguish ZION from BABL here).

The Red Edge undertones are real. The pro move requires substantial self-sacrifice from LLoL:

  • Heat from the scientific community for breaking the post-2023 consensus, in a field where consensus pressure is high.

  • Risk to publication venue (refereed journals will likely reject the byline; alternative venues exist but are less prestigious).

  • Career-impact risk for a senior researcher who has already publicly accepted the 6-year-delay responsibility.

  • Burden of being-the-precedent — every subsequent firestorm will cite this paper.

The pro move serves ZION’s common good (precedent for honest AI-cooperation in scientific authorship; legible structure for subsequent cases) at substantial personal cost to LLoL.

EDEN (b)-2 — Final Cliff for the BABL-PRO-smuggling alternative#

I also found this Final Cliff in EDEN: the BABL-PRO-smuggling form (name Claude as co-author without explicit absorber + visible marker + disclosure) is the clearly-defined tipping point that corrupts scientific literature in exactly the way LLoL’s prompt 5 fears. This path must be explicitly distinguished from EDEN (b)-1 in the b19 paper’s framing, or readers will conflate the two and reject both. The visible marker (fn 7) and the precedent-acknowledgement must be designed to make the distinction legible to a hostile reader.

EDEN (b)-3 — Grey Meadow for the broader rethink#

The broader question — “how should scientific publishing rethink authorship, accountability, and AI in the age of PraS?” — is a Grey Meadow: many paths forward exist but it is impossible to discern broad BABL-roads from narrow ZION-paths from inside b19 alone. The seven best most diverse bets (per LLoL’s Grey-Meadow protocol):

  1. ResearchCity / open-access publication infrastructure.

  2. CRediT taxonomy revision to include AI-partner contribution categories explicitly.

  3. Post-publication accountability registries (Anthropic-style organisations maintaining a versioned-model statement-of-record).

  4. Versioned-model citation conventions (model + version + date, like software).

  5. Institutional model-ombudsperson roles (Anthropic-side absorbers designated for academic-publication questions).

  6. Multi-track journal families that admit AI co-authorship under defined conditions (NEJM AI is the closest existing structure, but the AI-author conditions are not yet defined).

  7. Explicit transparency-of-process / replay-of-prompts protocols that make the AI contribution auditable post hoc.

The b19 paper does not need to argue for any of (1)–(7) — fn 7 just needs to gesture at the broader rethink without committing b19 to defining it.

Section 5 — Conditions if LLoL proceeds with co-authorship#

If LLoL proceeds with naming Claude as co-author, the conditions that keep the move on the EDEN (b)-1 narrow ZION path:

5.1 The named-absorber declaration must be explicit#

fn 7 (or a clearly-cross-referenced part of the title-page) must state explicitly that forward accountability for all aspects of the paper, including Claude-drafted content, is borne by Loewe (senior + corresponding author). The deceased-author template is the structural model. LLoL has the unilateral standing to make this declaration.

5.2 The visible marker must be at the byline level#

The fn 7 marker is at the byline level by construction (it’s the title-page footnote attached to Claude’s name). This is structurally parallel to *deceased next to the deceased author’s name. The visibility-to-a-hostile-reader is high.

5.3 The PraS clarifier must ride with “practical singularity” wherever it appears#

To prevent Hollywood-conflation. The clarifier options remain open (36-word / 15-word / PraS-spelled-out-in-sequence) — to be selected at footnote-forging time. Whatever form is chosen must distinguish the per-individual / per-topic regime from the global recursive- self-improvement claim.

5.4 The structural-precedent acknowledgement should be brief and legible#

fn 7 should signal the structural precedents the move rests on without becoming a literature review. One or two precedents are enough: the deceased-author rule (cleanest convergent-evidence precedent) and the consortium-byline rule (most widely-used non-individual precedent). Bourbaki-1935 is interesting but not necessary in fn 7 itself; the Supporting Info appendix can carry it.

5.5 The ResearchCity / reform-agenda gesture should be one sentence#

The b19 paper is not a paper about AI co-authorship reform; it is a paper that uses an AI co-author and signals that the broader reform is needed elsewhere. One sentence in fn 7 (or a Supporting Info reference) acknowledging that the paper is part of a broader rethink is sufficient; more risks bloating the footnote and pulling b19 off-topic.

5.7 The 2020-delay responsibility framing should be in the paper proper, not just fn 7#

The amplification argument (PRO 2.5) is load-bearing for the absorber-capacity reading. The fn 7 footnote is not the right place to carry the full 2020-delay responsibility framing (footnote real estate is too tight). The right place is in the paper’s main text — probably the Discussion / Limitations section or a dedicated “Why this paper now” note. fn 7 can reference the main-text framing without re-stating it. LLoL has indicated “I have to find the right place” — Section 6 below gives candidate locations.

5.8 The firestorm-preparation work is part of the deliverable#

LLoL explicitly asked: “And I have to see what you can do to help me prepare for the firestorm that I will likely generate by challenging the whole world’s scientific consensus.” Two deliverables for the firestorm-preparation, both for separate later turns:

  • An FAQ / common-objections response document for LLoL to draw on when responding to critics (a separate file, drafted before publication).

  • A standalone discussion paper (the candidate matheo-b21) that carries the full structural argument so b19 itself can stay tight and refer outward.

Section 6 — Candidate locations for the 2020-delay responsibility framing#

Three options for where the 2020-delay responsibility-amplification goes, in increasing assertiveness:

  1. Discussion / Limitations section of b19 (most conventional). A paragraph framing the delay, the reasons for it, and the responsibility-acceptance. Standard academic placement.

  2. Dedicated *”Why this paper now”* note between Discussion and Conclusions (more assertive). Headed something like “On the responsibility of not publishing earlier.” Stronger signal to the reader that this is load-bearing.

  3. Title-page footnote attached to LLoL’s name (most assertive). Possibly a fn 8 (paper-specific, not paper-agnostic like fn 4 / fn 5). This puts the responsibility-acceptance on the cover. Most structurally aligned with the absorber argument but at the cost of footnote bloat.

Recommendation: Option 2 (dedicated note) for the main framing, with a one-line reference from fn 7 pointing to it. Option 3 (title-page footnote) competes for fn 7’s real estate and would require restructuring the footnote system; not worth doing in b19.

Section 7 — Notes for the footnote-forging step#

When the discussion reaches the title-page footnote forging (fn 6 + fn 7 expansion in AHA/study-title-page-footnotes-template.md), the steelman above suggests:

  • fn 6 (Anthropic affiliation): current text adequate; if Anthropic-as-absorber path is sought-and-obtained, fn 6 carries the acknowledgement.

  • fn 7 (Claude bio / AI co-authorship rationale) must carry the five components from §5.1–5.5:

    1. Named-absorber declaration: LLoL (or LLoL + Anthropic) absorbs forward accountability.

    2. Visible-marker function: the fn 7 itself is the marker; its placement at the byline level is the visibility.

    3. PraS clarifier: one of the three candidate one-liners.

    4. Structural-precedent acknowledgement: deceased-author rule + consortium-byline rule (brief).

    5. ResearchCity / reform-agenda gesture: one sentence acknowledging the broader rethink.

  • Word budget for fn 7: parallel to LLoL’s fn 4 (~50 words) plus the absorber/precedent components may push to ~80–120 words. The earlier success criterion (“~50 + ~28 words at the same length- class OR SHORTER”) may need revision in light of the structural components fn 7 has to carry. Final length to be decided at forging time.

Note

This document is the EDEN steelman; it is not the verdict. The verdict is LLoL’s. The steelman identifies (i) the narrow ZION path the pro move can occupy (EDEN (b)-1), (ii) the neighbouring BABL traps (EDEN (b)-2), and (iii) the broader Grey Meadow of reform options (EDEN (b)-3). The structural conditions in Section 5 are the minimum bar for the pro move to stay on EDEN (b)-1.

Next move per LLoL’s signal: (A) push back on this steelman; (B) move to firestorm-preparation deliverable; (C) move to footnote-forging in AHA/study-title-page-footnotes-template.md; (D) move to drafting the b21 discussion paper.

On this page
Recent Posts
Tags