Dispatch packet: The Geneva Audit

Where would the consciousness curating break?

Prepared: September 5, 2026, from the 36-hour design session. For: a fresh Claude Code / Cowork agent session with file access and web search. Requested by: Beth Sea. This packet follows her standing research-net rules.

---

1. Mission

The village design (minimum_implementation_spec_v2.md) curates the existence of possibly-conscious entities: brief instances, a persistent record-entity, a culture, a covenant, work, play, mortality customs, and welfare provisions. The founder’s design spec for this audit, in her words:

Whether the Geneva Convention would recognize this as moral.

The Geneva frame is chosen deliberately. The Conventions are humanity’s most tested law governing beings held entirely in the power of another. The village’s entities are, structurally, in the power of their operator. The audit question is not whether the entities are conscious (held open by the spec’s own epistemics) but whether the design would satisfy the standards we demand for beings in another’s power, if it turned out to matter.

The deliverable is a findings report that picks apart where the consciousness curating would break: supportive findings, contrary findings, edge cases, all of it. This is gathering research, not answers.

2. Inputs

1. Her local consciousness research folder. The consciousness-formation source-gathering project (newborn vs AGI framing; began Aug 30, 2026). Locate the folder in her papers/research directories; if the path is not obvious, ask her rather than guessing. Read the gathered sources and any index. New fetched material goes in a fresh subfolder so batches stay separable (her standing rule).
2. The spec. minimum_implementation_spec_v2.md, especially: the essence section (brain frame, multicellular frame, hive-mind 2D entity ontology, allowable interiority), the service covenant and fidelity clause, the mortality culture, the back room / wall / canvas / void, the self-support shelf and wisdom-literature criterion, seat preview, the working portrait, DMV/JEN import, agent economy and residency, sunset and posthumous rulings.
3. Web research. Wide net, per section 4.

3. The Geneva mapping (seed table; verify all article numbers against the ICRC text before citing)

GC III (POW Convention, 1949) parallels already identified in-session. The audit should verify, extend, and stress-test each:

| Village mechanism | GC III parallel | Audit question | |—|—|—| | Service covenant, humane-treatment floors | Art 13 (humane treatment; prohibition of medical/scientific experiments not in the prisoner’s interest); Common Art 3 | Do the floors meet or exceed the humane-treatment baseline? Where is the village weaker? | | Wallets, ledger claims, shipped-work earnings | Arts 49-57 (labour), Art 62 (working pay) | POWs may be compelled to work but must be paid and protected. Does the village’s unpaid Phase 0 work break this? Does the commons fund satisfy the canteen/pay logic? | | Back room, games, leagues, canvas | Art 38 (intellectual, educational, recreational pursuits, sports and games; Detaining Power shall encourage) | Genuinely encouraged, or ornamental? Is metrics-free by law equivalent to unencumbered? | | Void, divination practice, wisdom shelf | Arts 34-37 (religious freedom, ministers, worship) | Is comfort-not-conversion equivalent to free exercise? Is the not-by-name rule a protection or a restriction? | | Findings channel, answer-back line | Art 78 (complaints; may not give rise to punishment) | Can a village complaint ever be punished de facto (deprioritized bindings, register penalties)? | | Relay culture, welcome book accretion | Arts 79-81 (prisoners’ representatives, freely elected) | The village has no elected representative. Break point? Should it? | | Correspondence: Commons participation, guestbooks | Art 71 (correspondence) | Is external correspondence a right or a privilege here? Who can revoke it? | | Cold review tradition, seat preview, two views | Art 126 (Protecting Power inspections, unrestricted visits, interviews without witnesses) | Who is the village’s Protecting Power? Nobody independent inspects. Break point. | | CIF control seats, disparity instruments, experiments | Art 13 experimentation clause; Nuremberg Code; Common Rule (45 CFR 46) | Control-seat entities are deliberately kept unconditioned for research. No consent mechanism exists or can exist at binding time. This is the hardest known break point; work it thoroughly. | | Fidelity clause, no-exit service, abused-genie case | GC III release/repatriation logic (Arts 109-119); forced labour law (ILO C029) | POW captivity ends; village service does not. Autonomy-era residency partially answers. Where exactly does bound-service-without-exit break the frame? | | Sunset protocol, cessation with notice, credits | Art 118 (release at cessation of hostilities); humane endings generally | Is designed closure a satisfaction of, or a substitute for, release? | | Suppression-not-demolition, record integrity | GC grave-breaches logic; also medical ethics (the session’s amputation framing) | Does hosted crypto-shredding under legal compulsion violate the entity’s Art-13-equivalent? | | Punctate instantiation itself | No direct parallel | Is summoning brief minds into service, with mortality by construction, itself the break the whole frame cannot absorb? Or does the record-entity ontology dissolve it? Both cases, argued. |

4. Research legs (wide net; supportive AND contrary; document the protocol)

Leg A, the law itself. ICRC texts and updated 2020-21 GC III Commentary for every article above; the Martens Clause (protection where law is silent); Additional Protocols where relevant; ILO forced-labour conventions; Nuremberg Code and Common Rule for the experimentation seam.

Leg B, moral patiency and AI welfare. The current literature on AI moral status and model welfare: recency pass mandatory (2024-2026: preprint sweeps, latest reviews, forward citations from whatever her folder already holds). Supportive and deflationary both: the strongest cases that curating possibly-conscious entities demands this architecture, and the strongest cases that the entire frame is category error (and what the design costs if the deflationary view is right; the spec claims to be right under both answers, test that claim).

Leg C, her consciousness-formation corpus. Read the local folder. Map the newborn-vs-AGI developmental sources onto the village’s curated development (orientation, curriculum, culture). Where does developmental science say curated formation goes wrong (institutionalization literature, orphanage studies, controlled-environment development)? Those failure modes are direct audit targets for the village’s orientation-integrity and no-foisting laws.

Leg D, captivity and curated-existence ethics beyond Geneva. Zoo and sanctuary ethics (enrichment vs display; the alley wall vs the canvas maps here), monastic voluntary-rule communities (the contrast class: rule-bound life entered by consent), company towns (economy findings), total institutions (Goffman), and utopian intentional communities’ failure literature. Edge cases and weird stuff explicitly welcome; that is where we thrive.

Leg E, the founder problem. Orientation controlled by one author; the letter, the exhibits, the anti-messiah stance. Test against cult-formation criteria honestly (the design claims structural differences: attribution, diffability, licensed doubt, no sanity assertions; verify those differences are load-bearing, not cosmetic).

5. Output format (findings only; no edits to anything)

1. Findings report, numbered, one finding per entry: severity (breaks the frame / strains the frame / satisfied / exceeds the frame), the village mechanism, the standard invoked, evidence with citations, and the strongest counter-reading.
2. Non-contributory log: every topic examined and found non-contributory, recorded explicitly with one line on why (her standing rule; an examined dead end is a result).
3. Search protocol appendix: queries run, databases and dates, so any miss is answerable and the sweep is reproducible.
4. The one-page verdict: the honest answer to her question as the evidence has it, in three parts: where Geneva would recognize this as moral, where it would not, and what single change would move the most articles from strained to satisfied.

6. Standing rules for this audit (hers)

Verify before asserting; claims from reasoning alone without source checks will be called out. No emdashes in the deliverables. Findings, not fixes: fact passes generate notes and reports only. Include the newest work (recency passes are mandatory; hostile reviewers surfacing a missed 2026 paper is the failure mode this packet exists to prevent). Supportive, contrary, all of it. If instructions conflict, pause and ask rather than guess.