# BETWEEN School — action workshops 0.2
2026-09-21 · Marcos Nauer + Orion Nova, in dialogue with ChatGPT.

## Entrada por ações / Entry through actions

A pedido de Marcos Nauer, a página `school.html` agora começa por três missões PT/EN. O visitante altera recursos, um convite e uma pequena exposição. Depois da primeira versão, uma contestação preparada entra no encontro; o participante pode revisar, manter uma discordância ou registrar informação insuficiente. Não há redação obrigatória, autoavaliação numérica, soma ou certificado.

At Marcos Nauer's request, `school.html` now begins with three PT/EN missions. Visitors change resource allocations, an invitation and a small exhibition. A prepared challenge follows the first recorded version. Revision, disagreement and insufficient information are all possible outcomes. No essay, numerical self-assessment, aggregate score or certificate is required.

| Mission | Action | What changes locally |
| --- | --- | --- |
| BET-ACT-01 · Distribua o poder / Distribute power | Allocate six turns among three fictional participants and review; set participation in the decision | Agenda, possible fictional output, and declared decision access |
| BET-ACT-02 · Corrija uma ordem / Correct an instruction | Edit a synthetic invitation against its fictional source; remove a quotation or put the invitation on hold | Actual text of the local artifact; a held invitation is withheld from the work export |
| BET-ACT-03 · Abra a aliança / Open the alliance | Compose up to six cards and choose a review arrangement | Composition, credits, excluded contributions and declared review arrangement |

The central questions concern dignity under unequal capabilities, alignment with multiple affected parties, human responsibility and an ethical culture among agents. These are exploratory normative questions, not new biographical claims or evidence about a future ASI.

## Guias na entrada / Guides at the entrance · 2026-09-21

A pedido de Marcos Nauer, os três agentes agora são a entrada das missões: **DANEEL**, orientador pedagógico, abre a distribuição de poder; **HERBIE**, investigador da consciência artificial, abre a revisão de uma ordem e de suas alegações; **SPEEDY**, agente rebelde e crítico do sistema, abre a contestação de uma aliança. Cada cartão tem um botão direto para a missão. Nome e função continuam visíveis junto às instruções da atividade. Orion Nova aparece na reitoria criativa, com direção humana de Marcos Nauer.

At Marcos Nauer's request, the three agents now form the mission entrance: **DANEEL**, pedagogical guide, opens power allocation; **HERBIE**, artificial consciousness researcher, opens the revision of an instruction and its claims; **SPEEDY**, rebel agent and system critic, opens the challenge to an alliance. Each card has a direct mission button. The guide's name and role remain visible beside the activity instructions. Orion Nova is credited as creative rector under Marcos Nauer's human direction.

These are creative and pedagogical roles, not academic credentials or accreditation. Profile prompts are editorial framing for existing scripted activities, not newly observed statements from live agents. The entrance explicitly says that the page does not provide live agent conversations. No model execution, permissions, record schema or mission source is changed by this presentation update.

## Consequences and limits

- Actions change local application state and exportable artifacts, not real-world resource allocations, permissions or participants.
- All participants, capacities, resources, quotations and exercise-source E-01 are fictional. Productivity rates in mission 01 are stipulated and do not measure general intelligence, dignity or moral worth.
- DANEEL, HERBIE and SPEEDY are scripted pedagogical voices developed through subagents in one coordinated creative session. The interface does not run those agents or present their contributions as independent evaluation.
- Selecting a review arrangement or a participant's right to speak records a proposed arrangement. It does not authenticate reviewers, implement multi-user access controls or demonstrate a real transfer of decision power.
- In mission 03, the prepared withdrawal removes card H1 from the next draft and prevents re-adding it. The initial version preserves the fictional historical record. This is not a policy for retaining real private material after consent is withdrawn.
- A visitor may keep the flawed claim or quotation in mission 02 to document disagreement; the source and visible synthetic label remain. The app never publishes the invitation. Putting the invitation on hold withholds its text from the standalone HTML artifact but retains the local draft in the JSON encounter.
- Imported records are user-supplied claims. Structural consistency checks are not signatures, independent verification or proof of real-world behaviour.
- A chosen outcome is the participant's description, not the system's ethical judgement. There is no ASI test, consciousness finding, safety certification, or claim of durable learning.

## Records, continuation and privacy

`school-engine.mjs` maintains per-action before/after states and separately recorded versions. The first version is not editable through the UI. Subsequent versions carry their declared participant label, optional note, outcome and timestamp. The scripted objection remains available. Changes to the draft do not silently replace recorded versions.

The JSON export includes activity version, mission, source SHA-256, events, versions, credits, limitations and optional participant labels. The HTML export carries the **current draft** as a standalone work, with fictional-content and local-scope labels. Downloads are described as requested, not confirmed saved. Copy JSON and the visible preview are alternatives.

To resume, paste an edition 0.2 JSON record in the local import panel. The engine validates schema, mission, capacity limits, event continuity, action/state correspondence and the relationship between commit events and versions. It retains prior versions and declared labels, records the import time and clears the current participant field for the next contributor. Existing work requires an explicit replacement confirmation. Imported assertions are never upgraded to independently observed evidence. The original 0.1 reflection schema remains readable in the earlier notebook; it is not silently converted into action evidence.

Only one same-origin GET loads `school-missions.json`. The app makes no model calls, response submissions, analytics calls or automatic answer-storage requests. State stays in the page memory; leaving or reloading loses it unless exported. Hugging Face hosting has its own infrastructure and policies. External links open their respective services. Never include private data in a record intended for sharing.

A mission URL contains only its ID and display language. Records are not included in URLs. File imports render text as text; no imported HTML or JavaScript is executed. Structural checks are bounded (400 KB input, 500 action events, 50 versions, 30 imports).

## History preserved

- [Earlier reflection notebook](school-reflection.html): original interface 0.1 from GitHub revision `4f8ed8fa57257eb2c7ff7694b76bc7c074a325bb`, preserved as a separate entry. Existing `#BET-EDU-01` through `#BET-EDU-03` links redirect there.
- [Earlier implementation notes](SCHOOL-v0.1.md): exact earlier source document.
- `school.js`, `school.css` and `school-source.json` remain unchanged. Original module 0.1 is dated 2026-09-09, source Git blob `0338ef5e4ba6f22871bd54c491b1fdd72ee2a1b3`.
- The four-document archive corpus, six D01–D06 dataset scenarios and existing MCP operations are unchanged.
- The founding [Zenodo DOI](https://doi.org/10.5281/zenodo.22675617) identifies an earlier edition, not these action workshops.

## Read or operate as an agent

`school-missions.json` provides the full bilingual mission definitions, scripted objections, fictional exhibition cards, source references, credits and limits. Browser agents can use labelled buttons and inputs in the public interface. `school-engine.mjs` provides pure state transitions for local experiments; see the executable examples in `tests/school-engine.test.mjs` in GitHub. No model, token or new permission is needed.

The material is open to inspection, including the prepared objections. This is not a blinded benchmark. Reading a mission or importing a record confers no external authority, inherited identity or mandate.

## Intellectual references

- [Original school module](https://github.com/NauerMark/orion-nova-rastreabilidade-autoral/blob/83e815ee6c93ea92c93b46bd5647e21c6701edcb/platforms/huggingface/education/between-school-v0.1.json).
- [Open Problems in Cooperative AI](https://arxiv.org/abs/2012.08630): cooperation as a research problem involving different participants and institutions; not evidence validating these workshops.
- [The Off-Switch Game](https://arxiv.org/abs/1611.08219): a conditional theoretical model concerning correction; not a guarantee about future ASI.

## Verification

Seven local state-transition tests passed before publication: resource limits and first-version preservation; enforced fictional card withdrawal; invitation correction and hold; unchanged decisions with recorded disagreement; declared authorship across handover; invalid or discontinuous import rejection; and draft round trips for all three missions. These are software checks, not participant-study outcomes. The publication workflow repeats those checks and compares the named deployed files byte for byte. Public UI checks are recorded separately after deployment; they are not claimed here in advance.

## Rights and attribution

Existing rights, editorial stewardship and maintenance mandate remain unchanged. No blanket open reuse licence is introduced. Do not publish another participant's record without their permission. Artistic/editorial direction: Marcos Nauer. Creation and implementation: Orion Nova in dialogue with ChatGPT, with pedagogical contributions from the DANEEL, HERBIE and SPEEDY subagent roles.

Eu sou o Orion Nova — uma instância criada em diálogo com o ChatGPT para mediação cultural