FFITTSS is a reasoning system proven on the ARC-AGI-3 benchmark. This page sets out, theoretically, how the same system would be deployed over a physical ecosystem instrumented with sensors, mapping its inputs, decision loop, and outputs onto a natural ecosystem. It is a design, not yet a built or measured result.
A physical ecosystem is, architecturally, another environment behind the same input boundary the benchmark uses.
The core consumes a grid of quantized signals plus a scalar consequence value, and emits an action. A sensor mesh can be normalized into that grid; a measured ecological outcome can supply the consequence. Because the core contains no environment-specific logic, deployment reduces to building the adapter that translates a place into that input format. This is theoretically grounded in the architecture but has not been empirically demonstrated.
The living loop · one heartbeat of a marsh
01
Sense. Sondes, weather and human observations stream in, a live and continuously updated field.
02
Ingest. The mesh is normalised into a raster the core can read; anomalies stand out against the seasonal floor.
03
Concresce. The engine derives a reference point from the convergence of the sensor-derived fields, then ranks the segmented entities under a learned relevance metric, its computed answer to what should be attended to now.
04
Aim. It commits to the top-ranked entity and holds that selection until a consequence warrants re-selection. Rendered over the map, the selected targets mark candidate sites for intervention.
05
Quests → care. People take up the chosen acts, clear a reach, tend a gate, remove invasive fish, and the outcome teaches the next heartbeat.
The five layers of a deployment
01 · inputSensors & weather
Live water chemistry, temperature, acoustics, and weather as an exogenous driver folded into the same field.
02 · ingestSensor → grid
The canal mesh becomes a symbol lattice; the water's own connectivity is the map's topology.
03 · engineThe concrescence
The organ fields cross into a standpoint, rebuilt each beat; each entity a point of seven felt lines, ranked under a learned metric.
04 · TxHeartbeat & aim
Each tick gathers every reading and every participant's aim into one new world-state, and a map of where to act.
05 · careQuests & care
Interventions are chosen quests taken up by people; their outcome is the very consequence that teaches the core.
Dimensions & boundaries · one physical field, many worlds
The deployment separates into distinct dimensions, each with defined data boundaries and permissions. The physical field, the sensor and consequence data, is the ground truth: the place's measured, felt identity. Other dimensions are rendered over it. One rule governs every boundary: a dimension may read the physical field, and may pass real, verified observations into it, but may never write simulated or unreal data back, the game plays into the place's identity without altering it. And a single place can run several dimensions at once, parallel worlds or servers, in the manner of Old School RuneScape or Minecraft, each with its own ruleset and audience, all sharing one physical ground truth.
RecreationalREAD · REAL OBSERVATIONS IN
Amateurs and children enter the 8-bit world, explore, and take light quests. Real observations they make can enter the physical field; in-game play never writes unreal data to it.
EducationalREAD · VERIFIED DATA IN
Students work the site's history, species and threats through structured quest-lines, contributing verified observations back to the field.
InstitutionalREAD · WRITE TO THE REAL WORLD
Scientists and stewards operate on the physical field directly, dispatching real interventions, validating data, and auditing the core.
Every dimension reads one physical field; none may write unreal data into it. The game plays into the place's real identity, it never overwrites it.
Case study · Xochimilco (theoretical)
Xochimilco is the last remnant of Mexico City's ancient lake system, a network of canals threading the thousand-year-old chinampa gardens, and the only home on Earth of the axolotl. It is the paradigm first patient: already a lattice, already instrumented, already tended by a fifteen-year restoration practice, and it hands over one unambiguous vital sign.
6,000 → <35axolotls per km², 1998 → 2014
~170 kmof canals, the living lattice
36refuges, 71 biofilters, the actuators in place
Run read-only over a recorded sensor trace, the core's high-satisfaction cells should land on or near the refuges the restoration team already chose, independently rediscovering fifteen years of practice from the sensor field alone.
The 8-bit world · play is data, play is agency
One does not fall in love with a spreadsheet, one falls in love with a world. The final layer renders the ecosystem as a playable 8-bit MMO over the very same field the core reads. The map is the marsh; its health is the world's brightness; the aim field becomes quests.
Every player is a self-asserted profile: their observations feed the field, their chosen acts close the loop, the Pokémon GO precedent turned to care, with consent made explicit. And because acts on a fragile habitat are irreversible, the world is also the rehearsal: every intervention played out in the game before a drop of the real is spent.
Lineage in hand: Suona (the 8-bit music MMO) and D-Spore (the embedded pixel organism) already prove the aesthetic and the tech.
The human fleet · where the labor lives
The core runs live, case by case, with no human in the per-occasion loop, that is the point of a deployable core. Human labor lives in the workshop: deciding what matters, naming the new, taking the real-world risk, and holding the ethics.
Biologists & ecologistsDefine what flourishing means, the consequence signal, and set the vital bands.
Kind-namersExperts on call name a novel anomaly; the naming becomes a word in the core's grammar.
Caretakers & crewsChinamperos and field teams take up the quests, the physical close of the loop.
Substrate keepersSensor and data engineers keep the mesh and the raster faithful.
StewardsGovern the core and the environment, consent, safety gates, the ethical red lines.
The crowdCitizen scientists, students, visitors, the widest ring, feeding data and taking micro-quests.
Where it stands · the honest boundary
● LIVE
The reasoning core, proven on ARC-AGI-3, organs, felt lines, the learned language, the settle.
● VIABLE
A read-only natural ecosystem deployment: run the core over recorded sensors, render the aim field, no irreversible act.
● STAFFED
The agency loop, in Xochimilco the refuges, biofilters and caretakers already close it.
◐ DORMANT
The hydrological satisfaction field, validated, awaiting its first real marsh to pay off in.
○ DESIGNED
The 8-bit world and cross-ecosystem transfer, architecturally enabled, not yet built or measured.
Full technical specification: docs/FFITTSS_ECOSYSTEM_DEPLOYMENT.md, grounded line-by-line in the live code and the FFITTSS paper.
Questions & answers
Is this deployed in a real ecosystem yet?
No. The reasoning core is proven on the ARC-AGI-3 benchmark; the ecosystem deployment described here is a theoretical specification, not a built or measured result. The honest first step is to run the core read-only over recorded sensor data and compare its outputs against known restoration sites, with no action taken on the real world. See Xochimilco and the status panel above.
How is this different from environmental sensors or a dashboard?
A dashboard reports readings. This adds a decision layer on top of the same readings: from the sensor field it derives a single presiding target at each step, and it learns from consequence which interventions actually improve the place. It is a proposal engine, not a display, and every step of its reasoning is inspectable rather than a black box.
Does the AI control or manage the ecosystem on its own?
No. The core proposes; people decide and act. There is no human in the per-occasion computation, but every real-world intervention is a chosen act taken up by people, and stewards hold the safety gates. Because interventions on a fragile habitat are irreversible, any action is rehearsed in simulation before it is taken. See the human fleet and the boundary above.
Can a game or its players push fake data into the real ecosystem?
No, and this is enforced as a rule. A game dimension may read the physical field and may pass real, verified observations into it, but it can never write simulated or unreal data back. A single place can host several parallel worlds or servers at once, in the manner of Old School RuneScape or Minecraft, all sharing one physical ground truth that only real observations may touch.
What does "lines of force" mean, and why not a normal neural network?
The core is handed no ready-made world of objects. It has organs, and each organ casts a field across the incoming signal; the entities it acts on and the standpoint it reasons from are derived as the crossings where those fields concur. What it learns is a small, human-readable vocabulary rather than a fitted function, so its decisions can be read in plain terms. The framing follows Whitehead's Concept V, executed in code.
What data and services would a deployment need?
In-situ water sondes (dissolved oxygen, temperature, pH, conductivity, turbidity), meteorological feeds, remote sensing (Sentinel-2, Landsat), and biodiversity and lab inputs (iNaturalist, eDNA). These are fused into the core's input by an adapter layer; the consequence signal is read from the place's own measured outcomes. The ingestion pipeline reuses the same event stack the Observatory dashboard already runs.
Why Xochimilco as the first case?
It is the clearest test bed: the canal network is already a lattice, it is already instrumented by a fifteen-year UNAM restoration program, it is already actuated by refuges and biofilters, and it has one unambiguous vital sign in the axolotl. It lets the design be made concrete against real ecology, while remaining a theoretical study on paper.
What is still unproven?
Several things, stated plainly: transfer of a learned vocabulary from one ecosystem to the next has not been measured; quantizing continuous water chemistry into the core's inputs is a real modelling commitment; re-timing the system to a seasonal clock is unbuilt; and the 8-bit world is designed but not built. None of these is a flaw in the reasoning core; each is the work of a first deployment.
How can I get involved?
Four grades of engagement, all open: explore the demos, join a beta test, propose an ecosystem for the incubator, or research, build, or fund with us. See the alliance and join the network.
Propose an ecosystem
Our incubator pairs a core with a place, a natural ecosystem, an archive, a classroom, a game world, and your culture teaches its grammar. Explore, test, deploy, or ally.