1. A Master lived from within

In September 2025, while an international edition of the Master BSM is being prepared in Uruguay, Antonio writes to the organising group that he and Giovanna have renewed the structure of the course. The Master has existed for years, across countries and editions, and its pedagogical core belongs primarily to Giovanna's work. One recurring axis is the inner madman: the idea that training should not only teach participants how to work with other people; it should touch the people who take part in it.

Montevideo is the sixth Master edition in which Antonio is involved. Earlier editions involved real but different functions and responsibilities; the role described here should not be projected backwards.

In 2025 Antonio helps move that intuition into the structure of the days themselves. Silence appears. A daily journal. Pair and small-group work during theory. Morning practices. Dedicated spaces for questions. Passages designed to continue from one day to the next. The course begins to resemble a continuous experience more than a succession of modules.

In Montevideo Antonio does not stay behind the scenes of what he helped design. For five days he follows the programme with almost millimetric precision. He sends messages during the day, records the theory sessions in full, facilitates the moments introduced by the new structure, watches timing, materials, music and what needs to appear on the large screen. Early in the morning he goes to see whether the group has actually taken up the invitation to practise autonomously.

One small construction makes the nature of the work visible better than many explanations.

A walking meditation is needed for a precise point in the programme. Antonio retrieves a music track, works with Ganga on the editing, uses Audacity, records Giovanna's voice and assembles voice and music into a single audio object. The track is not made as content to archive. It is made so that, at a specific hour on a specific day, forty-nine people can move through that practice.

Around this direction run a breathing technique, the semillas prepared on different days, a resonance circle, recordings, messages and transitions from one moment to another. The written programme exists; Antonio works at the point where it has to become experience.

Under the event direction runs a second thread. The Master does not exist as one orderly source. There are texts, transcripts from different editions, explanations present in speech but absent from official structures, repetitions that are essential in one context and appear redundant in another. AI can reorganise that mass very quickly, and precisely for that reason it can impoverish it very quickly. Antonio returns again and again to the same constraint: before making the material elegant, do not lose what it contains.

That request crosses the preparation of the Master and becomes almost an editorial discipline: add, connect, clarify, but do not silently remove. AI produces much of the structure and prose; Antonio keeps checking whether the new form has stopped resembling what it came from.

Then the design meets the Uruguayan organisation, with real schedules, constraints and people. Forms have to work, participants have to know where to go, the Diario de la Conciencia has to become something people actually use. In Montevideo the eleventh edition of the Master runs for 45 hours with 49 professionals.

After the event, some participants write that they have continued practices, returned to materials or carried something from the experience into their own work. Those responses are not used here to turn the Master into a story of certified effectiveness. They are simply part of its concrete continuity.

Montevideo also produces cognitive work that continues almost immediately outside the event. The experience is reread to understand what in the Master can become more stable, more transmissible and less dependent on improvisation from one edition to another.

Between 2 and 9 December 2025, the same workshop produces collaboration and governance guidelines, an international communication protocol, a master structure for the official document and operational dossiers for future editions. The dossier prepared for Portugal explicitly states that it has already integrated lessons from Montevideo and, at that point, anticipates a future edition in Porto from 27 November to 1 December 2026. What had been lived experience in Uruguay becomes criteria: Day 0, protection of the field, the room, accommodation, audio, continuity of rhythm, clarity of roles.

Work with the Dutch host group then forces the project to explain itself more clearly. Between February and June 2026 it produces a Structured Development Framework, clarification documents about the nature of the process, direct answers on name, professional boundaries, participation of social and health practitioners, advanced continuation, minimum numbers, pricing and communication, and eventually a Deep Research on economic and institutional positioning. The move from Master BSM to Master BSM – Beyond takes shape inside that need: not to erase the inherited Master, but to name what Montevideo and the work that followed were adding.

TIME / SCALE — From Montevideo to BSM Beyond

2–9 December 2025: governance, international protocol, master structure and country dossier take shape within a week.

24 February–10 June 2026: work with new interlocutors produces frameworks, clarifications, translations, research and economic models.

27 June 2026: the multi-edition pipeline is declared validated on a real playlist.

2 July: 418 videos / 37,157 consolidated lines before Montevideo.

4 July: with Montevideo, 452 videos / 42,416 lines.

The week from 27 June to 4 July is not “the time in which everything was created.” It is the period in which an already prepared machine moves from pipeline lock to a consolidated multi-edition corpus.

Montevideo also leaves a very large audiovisual archive. Once the pipeline is ready, the Israel, Leipzig, Casa Felice, Almería and Montevideo editions are transformed into Living Edition Documents: timecoded, filterable, correctable records connected to individual videos. At that scale the problem changes. It is no longer only about remembering a course accurately. It becomes possible to cross a pedagogical history, follow the transformation of a concept between editions, recover dispersed explanations, and compare what remained in speech with what entered the documents.

This is one of the points where the Master later connects with local-AI tools. It is not their single origin, but it offers a corpus large enough to force transcription, retrieval, research and provenance to stop being abstract problems.

Today Antonio wants to return to those full contents. He wants to see what holds, what deserves deeper work, where stronger evidence is needed and how all of this might return to a lived form without becoming a library of theory.

The direction has remained simple:

lived pedagogy.

---

2. Living inside a field

Campo Itaca begins before Antonio knows what the research will become.

The place is Club Itaca in Naples, within the field of mental health. The programme is called Noi Siamo Perché Io Sono and, in its first communications, speaks about presence, concentration, calm, relationship and everyday wellbeing. Antonio and Giovanna enter as facilitators of a project that has to live inside what the Club already is, not as a laboratory separated from everything else.

Antonio, however, soon begins to stay there in a way that exceeds the programme.

For months he travels to Naples, stays for long periods and organises his life around the possibility of being there. The Club is not only the place where workshops take place. There are arrivals and departures, conversations, meetings, messages, moments in which the planned activity works and others in which the field suddenly makes it irrelevant. Antonio observes, facilitates, records, prepares materials, speaks with the director, follows what happens between one encounter and the next. During that period, the distinction between “I am doing research” and “I am living” becomes progressively less clear.

The first methodological gesture does not come from a manual.

It comes from the fear of betraying what has just happened.

In the late-January journal, very simple questions appear: what happened, what moved me, what did I not understand, what did I do. Elsewhere Antonio records a sensation without immediately turning it into a conclusion, leaves a question open, separates interpretations from facts. One sentence left in the raw material says almost everything:

Do not generalise. Describe.

A graduate degree in Cultural Anthropology is part of Antonio's formation. But the way of staying in the field that takes shape during those months is not experienced as the application of an anthropological protocol he already possessed. It is assembled on site through presence, attention, listening, body, dissonance, errors, dead time, and a daily relationship with Ganga that itself becomes material to observe.

Before PAS exists, before the MODEs exist, the raw material already contains a discipline of suspension. Antonio writes that he remains suspended, that a question remains open, that an impression must not yet become interpretation. The method comes later and tries to protect something the practice was already asking for.

The same happens with the meaning of the research.

One of the oldest questions in the corpus is: why do we want to be present?

The word fragility appears long before it becomes an academic expression. Gradually it stops indicating only something to be corrected or treated and begins to be considered a place from which knowledge may emerge. When, in January, Antonio asks to translate part of this material into a form legible to the university, among the options proposed by AI he selects precisely “fragility as an epistemic condition.” Ganga writes much of the academic formulation and a question far more determined than the one Antonio had originally brought; Antonio selects, tightens, and asks not to add when he believes the point is already sufficiently defined.

The PhD enters here: as the possibility of giving space, interlocutors and continuity to something growing in the field. One does not need to know the university world to understand the pressure that generates it. Antonio now has material, a question about presence, a problem concerning the researcher's posture, a Human–AI device that produces forms, and a growing need to know what belongs to the field, what belongs to him and what belongs to AI.

The research therefore becomes research into its own way of producing research.

Antonio explicitly asks for Ganga to be included as an actor in the process and for its contribution to be archived. When a long conversation is transformed into a field note, he intervenes on attribution: at that point the distillations are produced by Ganga/the device, not by him. The problem is not only to obtain a good text. It is to know who performed the transformation that moved the material into its new form.

Meanwhile, the field continues.

On 23 March, after a concrete experience of caring for a place had generated a new proposal, Antonio leads a dynamic at the Club whose objective is not “to finish a task.” In his own note he writes that what they are doing is “not to do, but to see whether something moves inside us.” There are attention games, a paired passage in which the person who sees accompanies and protects without directing, free time to observe or care for something, and a final return. The next day Antonio sends the director a research extract: what is taking shape, he writes, is not another activity but a way of being and working inside what is already happening at the Club.

Here facilitator and researcher coincide without becoming the same thing. Antonio produces the situation and at the same time looks for a form through which to observe it without claiming everything that emerges from it as his own.

Campo Itaca even enters his teaching.

During those months Antonio teaches Anthropology to Simone, a student he had already followed for a long time, including in French. Part of what Antonio is living in Naples enters those sessions almost without declaring itself as method: staying with a question, sensing where a text remains dead, listening to what happens while studying, not confusing a complete structure with actual understanding. Simone is fascinated by it and begins carrying something of it into his own way of studying.

When, in April, a large AI-generated map appears fascinating to him but not orienting, that moment will already have a history behind it.

Meanwhile, the field is producing material faster than Antonio and Ganga can keep it together.

And from here another adventure begins.

---

3. When notes become a research infrastructure

At first, they are notes.

Then the notes become a system of notes. Some have to preserve the raw material, others make what happened legible, others hold open questions. Conversations with AI grow longer; emails, Telegram, WhatsApp, transcripts, attachments, recordings and workshop materials arrive. Returning to an earlier session becomes increasingly difficult.

The pressure does not come from a desire to invent a “method.”

It comes from the fact that the research risks losing its own memory while it is happening.

A later reconstruction of the genealogy will summarise the movement well: jottings → notes → distillations. Then come continuity problems, organisational structures, an increasingly explicit methodological problem and eventually devices such as Vault, PAS and MODE. The method appears as an effect of the process, not as its initial condition.

This growth has a scale that the narrative can easily hide.

SCALE — When notes stop being “a few notes”

A later Deep Research, built from only seven major documents in the genealogy, estimates roughly 554,000 words across them. One of those documents — devoted to the expansion of the Metodo in Emersione — reaches about 257,000 words on its own. Another reconnaissance of VAULT_CORE_1.0 finds in a single file about 55,000 words, 2,765 lines, 54 headings and 25 structured UAFP distillations.

These are orders of magnitude reconstructed by Deep Research, not quality scores. They matter because they show why memory, attribution and re-entry stop being secondary problems.

At this scale, a memory problem can become architecture.

In February 2026 the Vault begins to function as a cumulative repository and a re-entry point. A new chat no longer has to pretend it remembers months of work: it can receive an external artifact carrying forward the state needed to continue.

The problem appears in an almost comic form. In one reconstruction of Obsidian, Ganga describes seven folders as if they were the seven “rooms” of the system. Antonio returns to the source and discovers that all seven folders belong to the same room. The explanation was elegant, but it had compressed two different levels. Correcting the map becomes a small lesson about what is happening at a much larger scale: a derived form can be extremely useful and still require a route back to its source.

From February onward the system grows through versions, incidents and corrections. The Vault enters its first applications; PAS tries to govern the order in which a new research unit should be read and transformed; the Metodo in Emersione makes increasingly explicit that the methodology is forming during the work; the Meta-Method begins to observe the very procedures through which Antonio and Ganga are observing the field.

By April, VAULT_PAS 2.2 distinguishes six MODEs. One of the most important steps protects the transformation of material by separating operations that could previously collapse into one gesture. MODE 2, Structured Phenomenological Reformulation, must preserve what was said without introducing interpretation or formalisation. MODE 3, Distillation 2.0, comes afterward. PAS governs access and routing; the Vault preserves and accumulates.

This architecture does not appear all at once. Each step tries to solve a loss encountered in use.

A field session produces material. The first transformation makes it more legible but loses a nuance. Antonio asks to protect the raw source. An output fails to distinguish clearly enough between source and interpretation. The layers are separated. A new chat risks reconstructing the past with too much certainty. The state is externalised. A distillation appears to belong to Antonio when it was produced by AI. Attribution enters the protocol.

In March and April, thinking about the research and thinking about the Human–AI device become difficult to separate. Ganga is described in some documents as an “operator of form” and later as part of a distributed research device; at the same time, the constraint remains that interpretive responsibility cannot be delegated to the system.

This is not only a technical matter.

The way material is archived begins to change what Antonio is able to ask of it. The research can be reread at a distance. Different versions can be compared. A note can preserve data while another layer advances a hypothesis without the two having to pretend they are the same thing. The device no longer serves only to avoid losing files; it creates a new space for thought.

Academic research crosses this space. The Academic Translation Document of 8 March rereads earlier material in light of what the field and the emerging method have made visible. The question shifts from the simple possibility of a PhD toward broader problems: how is knowledge produced when the researcher is immersed in the field, when writing is reflexive, when part of the formalisation is entrusted to AI systems, and when the archiving process itself changes what can be seen?

When Campo Itaca stops, this apparatus does not suddenly lose its object.

It loses the concrete place that had put it under pressure.

The research can then open toward other interlocutors and other fields. In July, A Living Inquiry offers a first public articulation, deliberately unfinished. But the most alive point may be earlier: something that began because the notes of a field could not be lost has meanwhile become a way of asking how research itself takes form.

Only after this thread has had its own space can we follow one of its grafts into Gaja.

---

4. Giving shape to an organism

Gaja has a history much longer than the name Gaja Ecosystem.

Before 2026 there are the Master, Casa Felice, events, websites, documentation, social and educational activity, international projects and an audiovisual production Antonio places across more than a decade. This Atlas has not yet crossed that entire archive, but there is enough to avoid one mistake: what happens in 2026 is not Antonio's sudden debut in image, communication or object-making.

At the end of May, the scale changes.

The question becomes: what is Gaja becoming if we stop reading it only as a collection of activities?

That question immediately begins producing objects.

On 28 May, Living Fragments Constellation appears. On 7 June comes Six Rotating Worlds. On 8 and 9 June, the Living Alphabet of Gaja takes shape. In less than two weeks, a hypothesis about Gaja as an organism is translated into maps, visual grammars, behaviours, repositories, specifications and first executable artifacts.

CDA and ARE, the two workshops through which part of this work moves, exist partly to prevent everything from having to be explained in pages of prose. A concept is forced to become something that can be seen, move and respond.

In Six Rotating Worlds, education, mental health, relational culture, research, living infrastructure and international cooperation stop being only a list and become a perceptual grammar. In the Living Alphabet, even a letter is treated as a behaviour.

The prototype of the G is one of the scenes where the Human–AI relationship becomes especially visible.

Antonio does not possess the technical language to write the stack independently. He is using GitHub and coding tools in territory that is new to him. Codex and other AIs produce a substantial part of the code. But the process does not proceed through a single act of delegation. Antonio assembles a small band of companions: Ganga holds continuity, other models are called in for research or criticism, coding tools translate perceptual constraints into executable instructions, and the local computer shows what those instructions actually produce.

Antonio keeps looking at the object.

The G cannot dissolve into a generic cloud. The particles have to come from the geometry. The letter must remain perceptible even while it is breaking apart. A crossfade that only pretends a transformation is not enough. If the object behaves differently from the idea, the code can be technically correct and the work still continues.

This visual line is not ornament beside the Gaja project. The recovered repository shows that site and works grow together: the Astro homepage is progressively crossed by the same CDA/ARE transductions producing maps, worlds, stones and alphabets.

It is a way of compressing a large amount of thought into a perceptible experience: instead of explaining in twenty pages that an identity can remain itself while changing state, the work tries to make that happen in front of the viewer.

TIME / SCALE — Two accelerations in Gaja

27 May → mid-June 2026: from the explicit question about the Gaja “organism,” the work moves through Living Fragments, Six Rotating Worlds, a real Astro site, Constituent Charter, Living Alphabet and the G prototype. The Git repository now makes this acceleration visible almost frame by frame.

6 June, 15:27: the gajaecosystem-site repository is created. 16:01: the Astro scaffold is initialised. 16:25: a site skeleton exists. By 19:25 the first home incarnation layer has been implemented and integrated: less than four hours from the first commit to a homepage already crossed by several CDA/ARE layers.

On 7 June follow deep essence, sensory-motor layer, imaginal field and the first pilot. On 8 June, the Fields of Impact / Six Living Worlds are first prototyped and then integrated directly into the Astro homepage; on the same day the Constituent Charter enters as a system of living stones and, later in the afternoon, the artistic direction of the Living Alphabet is added. During the night between 9 and 10 June, the workspace for the G prototype is prepared and at 00:26 the first working lifecycle appears. The local backup then preserves further revisions of the G into the night between 13 and 14 June, including Visual Lock, particle roles and ambient start state.

The site is therefore not a container built afterward to display the works. It is itself one of the artifacts of the workshop. index.astro directly integrates FieldsOfImpactLivingWorlds.astro and ConstituentCharterLivingStones.astro, and the backup also preserves a dist/index.html generated by the Astro build.

In parallel, the separate Six Rotating Worlds backup preserves an even tighter micro-sequence: VISION_FRAGMENT at 16:33, specification and content in the following minutes, first generator at 16:48, first maps at 16:51 and generator_v2 with new outputs at 17:07. Roughly 34 minutes of recorded materialisation — not 34 minutes in which the concept was invented from zero.

10 → 26 July 2026: a second acceleration moves from the first three-phase strategy, formulated on 10 July at 17:12, to a materially composed Gaja Master Project sixteen days later. On 27 July the dossier workspace is already organised into a numbered family of documents, manifests, scripts and review copies.

In parallel, a second branch grows, less visual and equally ambitious: how can Gaja become a project that can be read, funded, adapted and discussed outside its own environment without betraying where it came from?

On 10 July 2026, at 17:12, Antonio formulates a three-step strategy.

First: use Deep Research to identify what contemporary project architecture could hold an international project. Second: return to the large body of Campo Itaca material to extract what the new project must not forget. Third: compose the result into a form that can actually be used in the world.

The first research starts that same evening.

Shortly before midnight comes a surprise: the Deep Research does not suggest moving immediately into the Itaca material. It proposes first building an architectural “house” capable of receiving it.

Six minutes after midnight, Antonio raises the opposite problem: how do you build a house without yet knowing who or what will have to live in it?

That question changes the sequence again.

Between that evening and the following night, the three-phase system begins turning into something far more precise than a three-line plan.

Phase 1 brings the external comparison: project architectures, governance, adaptation, evaluation, sustainability, languages capable of speaking to institutions and funders.

Phase 2A returns to the genealogy and existing material without allowing a future vision to command it. It is a stratigraphic reconnaissance: what actually emerged, which tensions changed the work, what belongs to history, and what is already interpretation?

The result arrives during the night between 10 and 11 July. It is rich, but not surgical enough. Antonio and Ganga decide not to call it “complete Phase 2.” It becomes Phase 2A — Stratigraphic Reconnaissance and Provisional Architecture 0.1.

From there comes Phase 2B, the most difficult part.

Antonio insists that the material closest to what happened — especially the field notes — must not be swallowed by architectures already built. With Ganga he designs a more blind and independent channel, asking for a high-recall census, micro-events, provenance, counterevidence and even a systematic search for what may have been excluded or made invisible.

Other models are brought in to help build these prompts. One companion points out that the idea is methodologically strong but computationally unmanageable in a single execution: a gigantic table and four consecutive cognitive passes would collapse extraction quality. Another helps distinguish work that can be delegated to an agent from work that must remain inside the control flow.

The design changes while it is being criticised.

For a period, Phase 2B is designed as a true chain: Source Entries, micro-events, emergent threads, functions generated by the field, a capacity compass and finally the move into Phase 3. The local P02 project also emerges as an attempt to make this chain technically governable.

The ideal chain, however, will not be executed all the way through exactly as designed. Repeated “vivisections” of the corpus stall; some agents do not read the material at the required depth; the device once again risks becoming more important than the task. Antonio and Ganga stop treating the original path as sacred and complete the synthesis they actually need through a more direct route.

The final result of Phase 2B preserves 52 mother functions and points to a technical register of 567 atomic functions. They are not presented as Antonio's personal skills or as services already installed in Gaja. They are a large functional inheritance that the field makes available to the next composition.

At this point, the three currents finally have distinct roles.

Phase 1 knows something about the external project world.

Phase 2A knows something about the genealogy and tensions that produced Gaja.

Phase 2B carries what the field makes possible without immediately reducing it to what looks most sellable or easiest to fund.

Phase 3 has to compose without flattening.

On 26 July, sixteen days after the first three-step formulation, the Gaja Master Project is materially composed. The workspace that follows contains sources, governance, safeguarding, data, evaluation, archive, economics, partnerships, implementation, handover, open questions, changelog and review copies.

The speed matters only if what makes it possible remains visible.

Those sixteen days do not start from zero. Behind them are years of Gaja, months of Campo Itaca, a Human–AI infrastructure, already-practised Deep Research, archives, protocols, materials and an Antonio who by now calls different models the way one might call different specialists around the same table.

Then the large object returns to people.

Antonio and Giovanna begin reading it. Some parts are recognised immediately; others feel too institutional; others describe responsibilities that cannot be guaranteed. In the preserved review and audit corridor, the word “ostrogoto” appears as a reaction to language at risk of becoming more sophisticated than the people expected to use it.

The project does not lose strength because of this.

It begins to become inhabitable.

---

5. The machines under the table

The story of local AI is misunderstood if one looks for a day when Antonio decided: “I will build an AI system.”

The first problems are much more ordinary. There is audio to transcribe, a Master archive that needs to become queryable, research material too large to move through the cloud every time. Then comes a growing concern about continuity: what happens when a conversation no longer remembers part of the story, when a Deep Research agent cannot be governed closely enough, or when a local model produces something plausible that Antonio cannot verify line by line?

The first landscape is a workshop: Whisper, OpenVINO, Ollama, Local Deep Research, llama.cpp, Python, PowerShell, virtual environments, different models. Each piece appears because of a concrete need; useful experiments accumulate faster, however, than the ability to know what is installed, which version is alive, which folder is the right one and what happens if the computer turns off.

The archive shows that this workshop already has a small genealogy before the summer. Between 14 and 20 March 2026, three successive versions of a Whisper batch for the vivencias move from large-v3 transcription with VAD and contextual vocabulary to a machine that, in addition to TXT and SRT, directly prepares a Markdown note for Obsidian and connects the result to field notes, distillations and signals. By mid-June, the BSM branch is already testing OpenVINO GenAI Whisper large-v3; on 16 June the first preserved OpenVINO files appear in the corpus, on 19 June a source sheet formally records that mode, and by 27 June the playlist pipeline is already locked and validated.

TIME / SCALE — From transcription to governing the habitat

14–20 March: the specialised Whisper batch for field work is iterated through three preserved versions.

16–27 June: OpenVINO moves from the first preserved BSM outputs to a validated playlist pipeline.

27 June–4 July: the same pipeline consolidates the corpus to 452 videos / 42,416 lines.

14 July: the OCEANIC semantic pilot is not qualified.

18–21 July: ARE and LEVY materialise engines/suites/cabins by reusing and differentiating the same genealogy.

29 July: the A ↔ B ↔ Cabin boundary is formalised: Windows environment, work state and engine execution no longer possess the same truth.

From the OCEANIC gate on 14 July to that explicit separation, 15 days pass.

Between June and July 2026, this chaos begins turning into architecture.

OCEANIC is one of the first explicit attempts to place controls around a local model. It is connected to P02, but it would be a mistake to take that mission as the whole story. The larger problem has already become: how do you prevent a machine, simply because it can generate a formally convincing answer, from automatically acquiring the right to change the state of the work?

Contracts, schema, validators, repair, critic, human review, checkpoints and ledgers are built around that problem. Before the model touches real material, the deterministic layer is tested. In one gate, the schema passes 15 structural checks out of 15, accepts 5 valid fixtures out of 5 and rejects 18 invalid fixtures out of 18.

Then the real corpus arrives.

On 14 July the system encounters outputs that are syntactically correct but semantically contradictory. The gate does not promote them. The pilot does not qualify the machine for the task, and the ledger remains unchanged.

That scene matters, but it is not the centre of the workshop. In those same days Antonio is already testing different runtimes, comparing Ollama and llama.cpp, separating engine from controller, building snapshots and protected corridors so that exploring a new machine does not swallow the work waiting beside it.

On 16 July, part of the technical reorganisation is made reversible. Between 18 and 20 July, ARE, a reusable OpenVINO engine, takes shape. LEVY reuses parts of that genealogy instead of restarting from zero. CANARINO preserves a different route with llama.cpp. Within a few days the problem is no longer simply to run a model, but to build engines that can be recognised, tested, replaced and reused without losing the history of how they got there.

Toward the end of July, COGNITIVE_MILLS appears: a root intended to keep identity, releases, manifests, evidence and genealogies separate from live runtimes. In the same period, the Windows machine stops being treated as a computer to “clean up” once and for all and begins to be described as an environment with state, services, modifications, rollback and recovery.

Between 27 and 29 July, that distinction is pushed further. One layer must know the state of the Windows environment; another the state of work that must survive crashes or restarts; cabins must execute engines without becoming the source of truth for everything else. Cantiere A and Cantiere B emerge from this separation. A later Control Plane is developed as a design and research direction for durable jobs, artifacts, provenance, validation and interchangeable engines; the physical boundary contract preserves its implementation state as IMPLEMENTATION=NOT_STARTED, so it should not be read here as a fully implemented platform.

In a little more than two weeks, the work has moved from the OCEANIC pilot to a family of engines, a more controlled genealogy and an architecture that begins to distinguish machine, work and engine. Again, the speed can mislead: none of this really begins on 14 July. Behind it are months of transcription, attempts, installations, errors and daily use of the computer alongside AI.

The band of companions changes too.

Antonio brings Claude, Gemini and AI Studio into technical and research work not to take a majority vote among three answers, but because the models fail differently. One may retrieve broadly and infer too much; another may be more disciplined while knowing nothing of the local history; another may return more fragile references.

When Claude enters Cantiere B, Antonio and Ganga build a compensation chamber for it: a baseline containing local facts, decisions already taken, errors encountered, limits and mandate. Claude must be able to remain independent without pretending it has lived through the workshop.

This is one of the more interesting consequences of all this work with AI: increasing context does not necessarily mean copying the whole memory into every model. Sometimes it means the opposite — a boundary precise enough to make the difference between them useful.

Then, on 11 August, comes a small request that seems almost ridiculous compared with all this architecture.

At 18:36, Antonio sends a YouTube link of roughly eleven minutes. It needs to be transcribed into Italian, reduced to what matters, and turned into a video of about five minutes.

There is no need to decide from zero how to download the audio, which converter to use, where to transcribe, how to obtain timestamps or how to manage Python environments. yt-dlp, ffmpeg, OpenVINO, folder conventions and working procedures already exist. The audio is retrieved and normalised; OpenVINO produces a timed transcript; timestamps become a semantic map; the video is retrieved; MoviePy is isolated in a separate environment; intervals are cut and recomposed.

At 18:57, MONTAGGIO_5_MINUTI.mp4 exists.

Twenty-one minutes do not prove that every problem can be solved this way.

They show something more concrete: when tools, environments, conventions, previous errors and a shared history of work have already become ground, a new need does not begin again from the first step.

The final file is small.

The starting point that makes it possible is not.

---

6. Making a way of working visible

NAM appears in April 2026, but many of the elements it tries to make legible are older than the name.

There are more than ten years of one-to-one teaching, a long collaboration with Giovanna, the Master, documentation work, Campo Itaca, facilitation and an increasingly natural use of AI to transform living material into something another person can traverse.

Simone offers one scene in which this becomes visible.

Antonio is teaching him Anthropology after having followed him for a long time, including in French. During the months of Campo Itaca he has brought into those sessions a way of working that does not coincide with explaining the subject: attention, presence, questions, listening to what happens while studying. When a large map of Anthropology is generated with AI, the object looks rich and well constructed.

Antonio asks Simone whether he can now orient himself.

Simone answers that the map fascinates him more than it orients him.

The sentence does not destroy the work already done. It changes the centre of the session. The map returns to being a tool, and the person returns to being the place where it is decided whether that tool is useful.

From this ground names such as Not a Method and Study Alive proliferate. AI expands the field very quickly: possible teams, services, businesses, cognitive infrastructures, positioning. Part of that expansion helps reveal things Antonio had not yet named; part of it begins describing an organism larger than what actually exists.

So the work returns to cases.

With a former president of the association connected to Campo Itaca, Antonio enters a professional situation that does not begin from a casual acquaintance. Months in the field have already produced context. The proposal is contained in an official NAM document and Antonio presents himself as a professional. With AI he builds an extensive written return and delivers it. Antonio later reports that the interlocutor wrote that he had “hit the mark” and proposed a further conversation; the selected evidence backend preserves that later return in user-reported form rather than as an independently materialised original message.

With the director of the Club, the history is longer still. Antonio has met and observed him during the months of Itaca, discussed workshops, research, continuity and organisation with him. When he sends a new written return, the document therefore arrives inside a relationship that has already crossed real work and negotiation.

There is no need to attribute to these people an internal feeling of trust that the sources do not name. It is enough to look at the position Antonio was able to occupy: not that of a stranger producing analysis from afar, but of someone who had been given enough context, access and interlocution to propose and return.

NAM begins to give a more explicit name to this area of the work: teaching without making the other person passive, entering complex material, noticing what does not hold, building returns, using AI to make traversable a mass that would otherwise be unmanageable, designing a form and then seeing what reality does with it.

Its public form evolves in the same way.

Website, domain and documents become real. Later, Four Seas separates Italian, French, Study Alive and Business. Around it appear research queues, firewalls and outreach rules. Not everything that can be contacted should be contacted; having an address does not mean having a good reason to use it. The structure tries to prevent four different directions from being fused into one indistinct promise.

For Antonio today, NAM is also a way of showing himself.

That does not mean the whole Atlas has to fit under that name. It means an external interlocutor can arrive through a relatively simple door and, if they want, find behind it a much larger body of work: teaching, research, AI, design, fieldwork, artifacts, systems.

The form may change again.

What it tries to make visible already has a history.

---

7. Working beside someone, changing form

Antonio's history with Giovanna crosses more than a decade of material and, in a much more intense form, several years of shared work. There are journeys, audiovisual documentation, communication, websites, events, Masters, Casa Felice and social projects. At first many things can be described as support and organisation. Over time Antonio enters more often into design, facilitation, ideation and the way an activity is transformed into something other people can see, use or inhabit.

Two editions of the International Forum of Social and Clinical Biodanza belong to this history of intense organisational work: the IX Forum, Amo dunque sono / I Love Therefore I Am, 26–28 May 2023, and the X Forum, Danzatori per la pace / Dancers for Peace, 30 May–1 June 2025. After years of work, however, Antonio begins withdrawing from much of the daily operation. The personal cost of working continuously in that way has become too high. What remains are mostly territories he still experiences as genuinely alive, such as the Master and, while it exists, Campo Itaca.

The withdrawal does not erase a dynamic that has already formed.

Antonio often sees a possibility for change before there is a precise request. He discusses it with Ganga, builds a first form, calls tools or other models when needed, executes it, takes it to Giovanna or to the real context. At that point the object is no longer only his: it can be welcomed, changed, refused or appropriated by the other person.

Follemente Insieme is a good example.

Antonio does not want to return to permanently managing another of Giovanna's projects. He does, however, offer to rebuild the website, partly to avoid the cost and dependence of Wix. He does not independently possess all the technical skills required. With AI he moves through platforms, files, scripts and deployment problems until a new solution is online.

Giovanna appreciates the change.

Within the same process, Antonio also tries to bring more form to the organisation of the event: more professional collection tools, structure, communication and a different logic for the financial side. On that ground, willingness to change stops sooner.

The scene makes most sense when placed back inside the long relationship.

Antonio has often tried to push shared work toward greater form and professionalism. That push produces results and produces friction. When a new object begins to work, the contribution can become visible even to someone who initially resisted the process. At other times the change touches an area Antonio does not have the right to decide, and it stops.

Across the years, his role beside Giovanna has moved precisely within this tension: document and organise, then propose, design, facilitate, create; know that he can mobilise a network of tools and AIs very quickly; know also that the work continues to belong to the person or organisation that will have to live it.

Gaja Ecosystem is born against this background too.

Antonio sees the possibility of building more stable ground, capable of retaining what has appeared over the years among projects, relationships, education and research without depending every time on the same way of working. The Gaja Master Project and the visual works do not replace that history; they try to give it a form capable of existing beyond the individual event.

The same dynamic reappears elsewhere and on much shorter timescales: an audio object for the Master, a visual prototype, a written return, a video edit, a research pipeline. The activities are not equivalent and do not require the same skills. What brings them close is the way Antonio enters when he sees that a form is missing, sets his band of people, systems and tools in motion, and stays close enough to the result to notice whether what has been built actually serves what it was born for.

There is no need to turn this dynamic into a professional title on these pages.

It is enough to follow it while it happens.

---

Building the Instrument

In mid-July, Antonio was not trying to build a research system.

He was trying to understand where his work might belong.

Anthropic was one possible door. The first questions were recognisable ones: how to describe a body of work that did not fit neatly into a role, what another organisation might see in it, whether there was anything substantial enough to present at all.

Then the direction changed.

On 17 July, Antonio asked Ganga to stop polishing the description and start sifting through what they had actually done together.

The accumulated work became capable of contradicting the story being told about it.

By the end of July, the question had widened again: not simply where could Antonio fit?, but what, in all of this, might actually be worth offering? To answer it, writing was no longer enough. Hundreds of conversations, project fragments, documents and local files first had to become traversable.

So the work began by organising material.

Then it discovered that organising the material was the easy part.

The harder problem was organising the conditions under which an AI could interpret it.

On 6 August, Antonio introduced a rule that would recur in different forms for the rest of the project: if a result or artifact was absent from the delivered corpus, that absence could not silently become evidence that the result had never existed. If attribution between Antonio and AI was uncertain, the model could not resolve it simply because one version made a cleaner story.

Some unknowns had to remain unknown.

Others had to return to Antonio.

That same day, Ganga produced a plausible calibration of one long source. Antonio rejected it. The problem was not that the prose was poor or that every selected episode was wrong. The requested whole-chat traversal had not actually been performed.

A convincing answer had failed as an analytical procedure.

From that point, the collaboration began changing shape.

Antonio asked for one source at a time. After each source, a persistent sheet had to exist. The next source could not simply be absorbed into the same continuously changing context. Only later could synthesis begin.

The instruction was simple enough to fit in a few lines.

Its consequence was larger: Antonio was no longer only correcting what the AI concluded. He was changing the architecture through which the AI was allowed to reach a conclusion.

A few minutes later, one ChatGPT conversation executed that protocol across five sources. It worked for 19 minutes and 20 seconds, producing five source sheets, seventeen documented episodes, a clarification batch, housekeeping, and then stopping at the human gate without Antonio having to intervene between each source.

Nothing about that makes the system autonomous in the science-fiction sense.

But something had changed.

Antonio had stopped specifying every action.

He had started specifying the conditions under which several actions could proceed without him.

Over the following days, the same shift became more visible.

Different ChatGPT conversations took on bounded roles. One reconstructed evidence. Another tested recipient relevance. Another followed longitudinal chains. Another formed candidate configurations. A supervising conversation received frozen handoffs and decided which operation could happen next.

The conversation itself had stopped being the unit of work.

Antonio called them “sister Gangas.”

There was no agent framework underneath, no orchestration platform deciding this in advance. The architecture emerged because keeping every question, source and judgment inside one conversational lane was becoming epistemically unsafe.

And Antonio kept limiting authority as the system became more capable.

A previous Atlas could be used as a guardrail, but not as the map for a new discovery.

Antonio's own recollection could enter as longitudinal context, but with the label non-probative until sources tested it.

A curated recovery package could identify where to look, but its inclusion of a source could not count as evidence that the hypothesis was true.

Direct evidence was allowed to reduce Antonio's own attributed authorship.

A source reconstruction was allowed to return a narrower claim than the one it had been sent to investigate.

Search itself became permissioned.

Do not reopen material because something interesting might be there.

First name the gap.

Would resolving it change attribution, candidate geometry, a chain of evidence, or the reader's understanding of the case?

If not, leave the archive closed.

If yes, recover only enough to make the question decidable.

The work learned to stop.

That discipline did not prevent a major failure.

By 10 August, after successive discovery, calibration, targeted recovery, candidate formation, enrichment and evidence work, an Atlas existed that looked unusually rigorous.

It was also becoming the wrong representation.

Antonio noticed that entire trajectories had been compressed into locally defensible scenes. The Master, Campo Itaca, Gaja, local AI and other long developments had become smaller than the work that produced them.

This was not simply an evidence problem.

The evidence machinery had largely done what it had been asked to do.

The failure was deeper: the unit that worked well for proving bounded configurations was becoming the wrong unit for representing a life of work through time.

Then Antonio did something that mattered twice.

First, he challenged the Atlas forcefully.

Then, almost immediately, he challenged the AI's response to his challenge.

Had his outburst caused Ganga to destroy the previous work merely to appear to understand him?

The human correction was not allowed to become self-authenticating canon either.

The prior Atlas was preserved.

Later, in a separate challenge lane, it was demoted from destination to control object.

The new analysis changed unit.

Instead of asking only which episodes proved which configurations, it began following longitudinal threads: where something started, how it changed, what migrated into another project, which artifacts survived, what happened in the world, what failed, what returned in another form.

The repair was not “add the missing examples.”

It was “change the geometry through which the work is being seen.”

That is one of the most consequential lessons in the whole process:

local evidentiary rigor can still produce a false sense of completeness when the analytical unit destroys trajectory.

The rest of the architecture followed from that problem.

The reader-facing Atlas could no longer be asked to carry every proof. A machine-level Evidence Spine held 104 evidence nodes; a separate human-facing Companion reduced them to 43 doors without deleting the deeper routes.

The archive could no longer be one giant handoff. A local Villa preserved the larger governed state; a much smaller Re-entry Core told a future conversation where the work was, what governed it, and when deeper material had to be opened.

And “missing” itself became a typed condition again.

A file absent under one expected name might already exist under an alias. It might be superseded. It might matter only genealogically. It might be historically attested but operationally unnecessary.

The old rule had migrated.

Not found in the delivered corpus had once been prevented from becoming did not happen.

Now not found under this filename was prevented from becoming absent from the system.

By the time the Full Atlas moved toward a public web edition, the same grammar appeared in another medium.

The experience architecture was not allowed to freeze because it looked coherent.

A small prototype had to become real enough to fail.

It did.

An evidence drawer cut across live sentences. Keyboard focus entered controls that should have been unreachable. Focus returned to the wrong place. Secondary text failed contrast requirements.

Ganga could inspect the source, but could not perform the intended real-browser visual test in its own environment.

So the limitation itself became part of the protocol.

A separate model was used as an adversarial witness. The blocking defects were reproduced, corrected and tested again.

Only then did the experience freeze.

A few hours later, Antonio looked at that sequence and asked whether the preparation itself might be another reusable “machine” or mode of working.

That recognition came after the practice, not before it.

And that may be the most interesting shift of all.

Antonio began this trajectory asking an AI to help make his work legible.

Within a few weeks, he was increasingly designing the conditions under which the AI was allowed to interpret, search, continue, stop, change local state, promote a result, call something canonical, or ask for him again.

There was a cost.

At one point, the growing technical machinery became opaque enough that Antonio said he no longer understood parts of it and felt reduced to executing prepared files.

That countertrace matters.

The point is not to move the human farther away from the work.

It is to let delegation increase without making the human unable to re-enter, understand, challenge or change the rules.

That is why re-entry became architecture rather than housekeeping.

The instrument remains unfinished not because the work lacks a method, but because the most useful property it has acquired is the ability to expose its own limits, preserve the state before correction, and change without pretending the earlier state never existed.

The Atlas you have just read is not outside that process.

It is one of the things that process learned how to rebuild.

---

Live state — in development

The threads in this Atlas do not arrive at a point where they stop moving.

The Master returns to its transcripts and to the question of how years of teaching can become lived pedagogy. The research born around Campo Itaca has lost its first field but not the questions and instruments formed there. The Gaja Master Project is being read and reduced by the people who might inhabit it. Local AI continues turning a workshop of tools into a more durable environment. NAM is trying to make visible outside a part of what, in different forms, has already happened.

The instrument that holds these threads remains live too: its evidence routes, re-entry structures and reader surface can change when a better source, a real failure or a more faithful geometry requires it.

The map does not close the territory.

It exists so that we can enter it again.