<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[LIVEX Lab Notes]]></title><description><![CDATA[Notes on building a deterministic multi-agent simulation (SYNE, ECHOS, PRISM)]]></description><link>https://livex.hashnode.dev</link><image><url>https://cdn.hashnode.com/uploads/logos/6af971cf4cb4386f21fa799a/85e6daa5-d262-4246-a118-a9dd7f3b2b59.png</url><title>LIVEX Lab Notes</title><link>https://livex.hashnode.dev</link></image><generator>RSS for Node</generator><lastBuildDate>Wed, 18 Nov 2026 14:58:01 GMT</lastBuildDate><atom:link href="https://livex.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[LIVEX studies collective emergence, not individual learning]]></title><description><![CDATA[What weak collective emergence means in LIVEX, and why it is not the same as an agent learning or improving.]]></description><link>https://livex.hashnode.dev/livex-studies-collective-emergence-not-individual-learning</link><guid isPermaLink="true">https://livex.hashnode.dev/livex-studies-collective-emergence-not-individual-learning</guid><category><![CDATA[emergence]]></category><category><![CDATA[multi-agent systems]]></category><category><![CDATA[simulation]]></category><category><![CDATA[research]]></category><category><![CDATA[Cognitive Science]]></category><dc:creator><![CDATA[LIVEX]]></dc:creator><pubDate>Sun, 15 Nov 2026 10:42:45 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af971cf4cb4386f21fa799a/65463726-c4b3-4b6a-a1a2-1a7418d8282d.jpg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>The word "emergence" can make you think of an agent that learns, gets more skilled, or discovers a new strategy on its own. That is not LIVEX's central question. The project is interested in what appears <strong>between agents</strong>, at the scale of a group.</p>
<p>In its framing, LIVEX speaks of <strong>weak collective emergence</strong>: groupings, exchanges, specializations or conflicts that are not written as a scenario, but that result from local interactions governed by explicit rules. One can trace observed patterns back to their causes in the model. The qualifier "weak" specifies the research scope adopted by LIVEX here. It is neither a universal category for all sciences nor proof of an emergent phenomenon.</p>
<h2>An agent can change without the project studying its learning</h2>
<p>Agents have a state that evolves: needs, memory and beliefs can change over the course of a run. Some rules can also vary a skill level or the ability to carry out an action. These individual changes are defined in the model. They are not the main object of research and must not be confused with collective emergence.</p>
<p>LIVEX does not aim to train an agent or to measure whether it "gets smarter." It seeks to know whether collective structures can appear without being prescribed, under what conditions they persist, and which mechanisms contribute to them.</p>
<h2>From observed pattern to measured claim</h2>
<p>Seeing two agents exchange or a group form is not enough to demonstrate emergence. We have to define the pattern under study, compare conditions, and report the results with their seeds, their horizons and their limits. A measurement can help detect a pattern, but it does not automatically turn that observation into proof.</p>
<p>This distinction matters for reading LIVEX's results: they concern collective behaviors in a simulated world, not a general capacity for individual learning, an intelligence, or the reproduction of a real society. The page <a href="https://donovan-dev-web.github.io/LIVEX/comprendre/emergence/">Emergence explained</a> develops this scope and its precautions.</p>
<hr />
<p><em>This article is an English translation of the original French post, <a href="https://donovan-dev-web.github.io/LIVEX/actualites/emergence-collective-faible-pas-apprentissage-individuel/">LIVEX étudie l'émergence collective, pas l'apprentissage individuel</a>, published on the LIVEX journal on November 4, 2026. Project website: <a href="https://donovan-dev-web.github.io/LIVEX/en/">LIVEX</a>.</em></p>
]]></content:encoded></item><item><title><![CDATA[Reproducibility and determinism at the heart of LIVEX]]></title><description><![CDATA[Same seed, same trajectory: why LIVEX treats reproducibility and determinism as scientific requirements, not options.]]></description><link>https://livex.hashnode.dev/reproducibility-and-determinism-at-the-heart-of-livex</link><guid isPermaLink="true">https://livex.hashnode.dev/reproducibility-and-determinism-at-the-heart-of-livex</guid><category><![CDATA[simulation]]></category><category><![CDATA[determinism]]></category><category><![CDATA[#Reproducibility]]></category><category><![CDATA[Testing]]></category><category><![CDATA[dotnet]]></category><dc:creator><![CDATA[LIVEX]]></dc:creator><pubDate>Sun, 15 Nov 2026 10:40:06 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af971cf4cb4386f21fa799a/fa94f513-f37f-4473-9f9d-03bfe7e41174.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>You start the world at 8:14 pm. For a few simulated hours, a group forms near a waterhole, stabilizes, specializes. You note the observation, you tell the story, and then you run it again. This time, nothing forms. The agents pass by, leave, cross paths without ever settling. Which of the two runs is right?</p>
<p>Without a way to replay, neither. You do not hold a result: you hold an anecdote. That is exactly what LIVEX refuses. If we want to be able to say "here is the chain of causes that produced this group," we must be able to <strong>replay</strong> the same world. Reproducibility and determinism are therefore not comfort options: they are the conditions of possibility of a scientific claim.</p>
<h2>An anecdote is not a result</h2>
<p>Emergence is a notion that is easy to invoke and hard to establish. It lends itself to stories: a striking configuration, a screenshot, a curve going up. But a story does not survive the next question: "does it reproduce?"</p>
<p>To reproduce does not mean to obtain "something similar." It means to obtain the same trajectory, or to be able to explain precisely why it differs. Without this requirement, interpretation wins over fact. We end up describing noise with beautiful curves, and calling "emergence" what was only a fluctuation.</p>
<h2>Three requirements, one setup</h2>
<p>LIVEX's guiding question demands an instrumented answer, not an argumentative one. Three requirements follow.</p>
<p>The first is the <strong>factual answer</strong>: "at this seed, under this consistent configuration, this phenomenon appears at tick N." Not "it seems to happen sometimes." The second is <strong>validation by replay</strong>: an independent observer must obtain the same result by replaying, otherwise the claim remains an opinion. The third is <strong>traceability</strong>: every claim must go back to local interactions, down to the decision and the event that produced it.</p>
<p>These three requirements cannot be met with a single tool. They call for a complete chain, from the engine that decides to the observatory that measures, through the orchestrator that archives.</p>
<h2>SYNE, or determinism by construction</h2>
<p>SYNE, the simulation engine, is designed for a deterministic run from controlled inputs. Three disciplines hold it together.</p>
<h3>One seed, one generator</h3>
<p>The pseudo-random generator is initialized once, from <code>seed</code>, and every draw goes through it. This choice has a consequence that is often underestimated: the generator's state must be <strong>serializable</strong>. This is what makes it possible to save a world mid-run and resume it exactly, rather than replaying it from the start.</p>
<p>From this comes a rule that surprises: using the runtime's standard random generator is forbidden. Its state is not portable across platforms, and its algorithm may change from one framework version to the next. Two "identical" runs executed elsewhere would then diverge silently. Forbidding it is not a whim: it closes the door on an invisible divergence.</p>
<h3>A causal order that does not move</h3>
<p>Progress is discrete, in ticks, with no uncontrolled system clock in the decision core. At each tick, the order is invariant: <strong>perception → evaluation → action selection → application of effects → updates → end of tick</strong>.</p>
<p>No decision can therefore be settled by a <code>DateTime.UtcNow</code> that would vary from one machine to another. The simulated time (<code>simulatedTimeSeconds</code>) is derived from the tick and the time scale (<code>simulatedSecondsPerTick</code>). Two executions do not "think" at different moments: they think in the same time.</p>
<h3>Resume without replaying</h3>
<p>The world is captured in full (state, cognition, generator) and restored identically. A run that was interrupted and restarted from its snapshot must produce exactly the continuation an uninterrupted run would have had. The resume tests verify this equality tick after tick, rather than assuming it.</p>
<p>SQLite persistence supports this promise: versioned schema, atomic transactions, resume at the last valid state, bounded backup rotation to avoid side effects. A world can be paused, moved and resumed without its trajectory changing.</p>
<h2>ECHOS, the observer that touches nothing</h2>
<p>On the other side of the stream, ECHOS consumes SYNE's observability <strong>without ever modifying the state</strong>. Three disciplines ensure this.</p>
<p>The stream follows a strict order, "snapshot then events," tick after tick. ECHOS checks this continuity, detects <strong>discontinuities</strong> (the gaps) and records them rather than hiding them. Every metric carries its <strong>provenance</strong>: it says whether it is measured, or whether a neutral fallback had to be applied for lack of a usable window. Nothing reconstructed is presented as measured. Finally, comparisons stay honest: <code>/api/compare</code> distinguishes a light summary (<code>light=1</code>) from a full comparison based on the bit-for-bit fingerprint, and the "bit-identical" verdict is only issued when conditions allow it to be established.</p>
<p>Calibration reports are versioned (<code>schemaVersion 2</code>) and derived from the persisted series, never fed back into the engine. The observatory writes its readings, and does not rewrite the world.</p>
<h2>Controlling what goes in</h2>
<p>Replaying also means <strong>controlling the inputs</strong>. The configuration stays consistent: <code>reference</code> and <code>prism</code> profiles, partial overlays applied uniformly, because <code>start</code>, <code>prepare</code> and <code>reset</code> follow the same rules. A single set of values is maintained per profile, to avoid silent drift.</p>
<p>Each execution carries a <strong>canonical <code>run_id</code></strong> (<code>run-&lt;seed&gt;-&lt;12hex&gt;</code>), which makes the link between the seed, the execution and its artifacts traceable: <code>.livexp</code> packages, ECHOS reports. The Launcher orchestrates accordingly: it derives the seeds, applies failure policies explicitly, records <code>provenance.json</code> as soon as the run is created, and on resume it does not replay the runs already validated, fingerprints and logs checked first.</p>
<p>Then there is time. <code>simulatedSecondsPerTick</code> changes the temporal representation of the world, and <code>metersPerUnit</code> that of space, for information only. But the <strong>effects per action</strong> and the causal order remain unchanged. The clock is instantiated at startup: it is not a global variable that something could slip in unnoticed.</p>
<h2>Proof by oracles</h2>
<p>All this is only worth something if it is verified. The project therefore translates its principles into concrete proofs.</p>
<p><strong>Checksum oracles</strong> pin golden states on the critical paths: any trajectory that deviates breaks a test. When the change is deliberate, it comes with a MINOR <code>engineVersion</code> bump. The rule fits in two lines: version unchanged, trajectory unchanged; trajectory changed, version incremented and fingerprints recomputed.</p>
<p>Multi-seed campaigns give this rule its substance. Ten seeds were run over 2,500 ticks: one hundred agents alive on each, and ten distinct checksums. Seed 12345, repeated three times (one of them with a stream export), finds the same fingerprint every time, <code>0xdb5979d695a2dc6a</code>. The export does not change the state; the replay does not change the trajectory. <strong>End-to-end tests</strong> chain a published SYNE and a real ECHOS, including on Windows, with byte-for-byte verification of the <code>.livexp</code> packages and archived reports. And gap detection is not decorative: tick discontinuities are logged structurally and counted (<code>gaps_detected</code>), never smoothed. At every measurement, the chain goes back to <strong>0 failures</strong>.</p>
<h2>A discipline, not an extra soul</h2>
<p>In LIVEX, "it works" is not enough. "It replays" is the norm. This discipline is demanding: it requires controlling what goes in, versioning what changes, naming what is not measured. But it is also what turns a curiosity, "can something emerge?", into a <strong>verifiable claim</strong>: "under these precise conditions, yes, and here is how to replay it."</p>
<p>It is precisely this rigor that gives weight to any observed phenomenon. If it emerges, it must be able to reappear identically. And if it does not reappear, you will at least know that it was not a property of the world, but an accident of your instrument.</p>
<p>For the details, <a href="https://github.com/donovan-dev-web/LIVEX/blob/main/docs/monographie/11-reproductibilite-determinisme.md">chapter 11 of the monograph</a> (in French) lays out the promise, its verification and its limits. The next measurements appear in the <a href="https://donovan-dev-web.github.io/LIVEX/en/actualites/">project journal</a>, also available as an <a href="https://donovan-dev-web.github.io/LIVEX/actualites/feed.xml">RSS feed</a>, and a ★ on the <a href="https://github.com/donovan-dev-web/LIVEX">repository</a> helps this work get seen.</p>
<hr />
<p><em>This article is an English translation of the original French post, <a href="https://donovan-dev-web.github.io/LIVEX/actualites/reproductibilite-et-determinisme-au-coeur-de-livex/">Reproductibilité et déterminisme au cœur de LIVEX</a>, published on the LIVEX journal on October 3, 2026. Project website: <a href="https://donovan-dev-web.github.io/LIVEX/en/">LIVEX</a>.</em></p>
]]></content:encoded></item><item><title><![CDATA[Architecture: three complementary modules (SYNE, ECHOS, PRISM)]]></title><description><![CDATA[How LIVEX separates the truth of the world, its observation and its perceptual rendering to guarantee reproducibility and traceability.]]></description><link>https://livex.hashnode.dev/architecture-three-complementary-modules-syne-echos-prism</link><guid isPermaLink="true">https://livex.hashnode.dev/architecture-three-complementary-modules-syne-echos-prism</guid><category><![CDATA[softwarearchitecture]]></category><category><![CDATA[simulation]]></category><category><![CDATA[multi-agent systems]]></category><category><![CDATA[dotnet]]></category><category><![CDATA[Python]]></category><category><![CDATA[Open Source]]></category><dc:creator><![CDATA[LIVEX]]></dc:creator><pubDate>Sun, 15 Nov 2026 10:35:53 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af971cf4cb4386f21fa799a/ca362a36-4819-4ef8-8649-075675c894ba.svg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<blockquote>
<p><strong>Note:</strong> this article describes the architecture as published in October 2026 (LIVEX 1.0.0 line). A reworked PRISM V2 is planned; see the <a href="https://donovan-dev-web.github.io/LIVEX/technique/prism/">PRISM V2 series</a> for the target.</p>
</blockquote>
<p>An agent is thirsty. It consults its memory, spots a waterhole it knows, weighs the energy of the trip against the risk of arriving too late, then sets off. At the precise instant it moves, three questions arise at once: who decided this step? Who observes it? Who shows it?</p>
<p>An improvised architecture would answer "a bit of everyone." LIVEX answers with a strict separation: <strong>one module holds the truth of the world, another observes it, a third makes it perceivable.</strong> Neither the observer nor the renderer can change the smallest decision of an agent. This principle was set down before the first feature, and it still drives every technical choice of the project today.</p>
<h2>Three roles, one truth</h2>
<p>The system is built around three complementary modules, with a deliberately sharp division of responsibilities.</p>
<table>
<thead>
<tr>
<th>Module</th>
<th>Full name</th>
<th>Role</th>
</tr>
</thead>
<tbody><tr>
<td><strong>SYNE</strong></td>
<td><em>Systems &amp; Emergent Network Engine</em></td>
<td>Multi-agent simulation engine. It holds the truth of the world state, runs the simulation loop and guarantees determinism. Independent of rendering.</td>
</tr>
<tr>
<td><strong>ECHOS</strong></td>
<td><em>Emergent Cognition &amp; Holistic Observation System</em></td>
<td>Observation, analysis, calibration and control system. It consumes the SYNE stream, computes metrics, detects emergent phenomena, produces calibration reports and allows comparison between runs.</td>
</tr>
<tr>
<td><strong>PRISM</strong></td>
<td><em>Perceptual Rendering &amp; Interactive Simulation Module</em></td>
<td>LIVEX's final Unreal project, integrating the <strong>PRISM-LDK plugin</strong> (<em>LIVEX Development Kit</em>). It provides perceptual rendering and immersion without carrying decision logic.</td>
</tr>
</tbody></table>
<p>This split looks like a matter of comfort. It is actually a scientific constraint. A world about which we want to prove something must be replayable identically, and for that we must know, at every moment, who has authority. As soon as two components can write into the world state, the chain of causality blurs: it becomes impossible to say whether the observed phenomenon comes from the laws of the simulation or from interference by the observation.</p>
<h2>Why decouple, and how far</h2>
<p>Decoupling answers three imperatives that hold together: <strong>reproducibility</strong>, <strong>traceability</strong> and <strong>portability</strong>. They impose a simple, almost brutal rule.</p>
<p><strong>Only SYNE has authority.</strong> Neither ECHOS nor PRISM modifies the decision state; SYNE is the single source of truth. This is what allows us to claim that a run is deterministic (for a given seed and configuration, it repeats) without having to prove that three components left it alone.</p>
<p><strong>Observation is non-intrusive.</strong> ECHOS subscribes to SYNE's observability stream (snapshots, events, ticks) without altering the simulation. The broadcast order and the semantics of the frames are fixed by contract, precisely so that the observer cannot introduce an invisible feedback loop.</p>
<p><strong>Rendering is independent of the core.</strong> PRISM does not carry cognition: it consumes SYNE's state and events to produce a perceptual representation, while remaining controllable through orchestration.</p>
<h2>SYNE, the decision core</h2>
<p>SYNE implements a <strong>discrete-time</strong> world, advanced tick by tick. The autonomous agents have needs, memory, beliefs, partial perception, a utility formula and capacities for action. Nothing spectacular in this list taken in isolation. It is their combination, tick after tick, that produces trajectories no line of code wrote in advance.</p>
<p>The loop is <strong>deterministic</strong>: with an identical seed and a consistent configuration, the execution lands on a state aligned with its oracles. Persistence, handled by SQLite, allows a run to resume at the exact place where it stopped, and the HTTP control server allows full programmatic control. A world can therefore be started, paused, moved and resumed without its trajectory changing. This is the condition of a replayable experiment.</p>
<h2>ECHOS, the observer that touches nothing</h2>
<p>On the other side of the stream, ECHOS turns raw observability into <strong>traceable knowledge</strong>. It ingests what SYNE publishes, persists an event log and snapshots, computes normalized metrics and builds time series from them. It also detects tick discontinuities (the gaps) and records them instead of smoothing them over. Every value carries its <strong>provenance</strong>.</p>
<p>Its calibration reports are versioned (<code>schemaVersion 2</code>) and include <code>outcome</code>, <code>extinctionTick</code> and a <code>viability</code> block. The comparison between runs (<code>/api/compare</code>) clearly distinguishes a verdict based on the <strong>bit-for-bit fingerprint</strong> from a simple lightweight summary: a demonstrated equality is never confused with a resemblance. What ECHOS produces stays read-only. The observatory writes its conclusions, and never rewrites the world.</p>
<h2>PRISM and the PRISM-LDK plugin</h2>
<p>PRISM is LIVEX's final Unreal project. It integrates <strong>PRISM-LDK</strong>, a technical module that formalizes the interface with SYNE and ECHOS: the <code>world_initialized</code> description (which includes <code>simulatedSecondsPerTick</code> and <code>metersPerUnit</code>), the snapshot stream, the events, the control cycle (<code>prepare</code>, <code>ready</code>, <code>start</code>, <code>pause</code>, <code>resume</code>, <code>stop</code>, <code>reset</code>) and supervision.</p>
<p><strong>ADR-002</strong> records this choice and supersedes the historical ADR-001, which described a Godot prototype: SYNE remains the only decision-maker, and PRISM is only a perceptual rendering host. The project preferred to own this shift through a written decision rather than through a silent drift.</p>
<h2>Contracts, not promises</h2>
<p>None of this would hold on the developers' goodwill alone. The whole is locked down by <strong>living contracts</strong> (<code>API_CONTRACTS.md</code>, <code>INTEGRATION_CONTRACT.md</code>, <code>TRANSPORT_API.md</code>) whose rules are few but non-negotiable. The "<strong>snapshot then events</strong>" order on the observability stream is strict, tick after tick. The canonical <code>run_id</code>, in the format <code>run-&lt;seed&gt;-&lt;12hex&gt;</code>, guarantees traceability between executions, packages and reports. And because these rules are tested, observation cannot, even accidentally, influence the course of the simulation.</p>
<h2>Separate to be able to prove</h2>
<p>The goal fits in one sentence: <strong>separate what runs the world (SYNE), from what observes it (ECHOS), from what shows it (PRISM).</strong> This separation is not an architect's refinement. It is the very condition of the reproducibility and traceability demanded by LIVEX's guiding question. Without it, every observation would be an anecdote. With it, every observation becomes a claim that can be replayed, verified and discussed.</p>
<p>To go further, the <a href="https://donovan-dev-web.github.io/LIVEX/documentation/">documentation portal</a> details the architecture and the contracts, and the <a href="https://donovan-dev-web.github.io/LIVEX/en/actualites/">project journal</a>, also available as an <a href="https://donovan-dev-web.github.io/LIVEX/actualites/feed.xml">RSS feed</a>, follows the evolution of the components measurement after measurement.</p>
<hr />
<p><em>This article is an English translation of the original French post, <a href="https://donovan-dev-web.github.io/LIVEX/actualites/architecture-trois-modules-complementaires/">Architecture : trois modules complémentaires (SYNE, ECHOS, PRISM)</a>, published on the LIVEX journal on October 2, 2026. Project website: <a href="https://donovan-dev-web.github.io/LIVEX/en/">LIVEX</a>. Source code: <a href="https://github.com/donovan-dev-web/LIVEX">GitHub</a>.</em></p>
]]></content:encoded></item><item><title><![CDATA[LIVEX: born from a question]]></title><description><![CDATA[Can unscripted social structures emerge in a simulated world? The question that started LIVEX, an open source deterministic multi-agent simulation.]]></description><link>https://livex.hashnode.dev/livex-born-from-a-question</link><guid isPermaLink="true">https://livex.hashnode.dev/livex-born-from-a-question</guid><category><![CDATA[simulation]]></category><category><![CDATA[multi-agent systems]]></category><category><![CDATA[emergence]]></category><category><![CDATA[Open Source]]></category><category><![CDATA[GameDev]]></category><dc:creator><![CDATA[LIVEX]]></dc:creator><pubDate>Sun, 15 Nov 2026 10:31:32 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af971cf4cb4386f21fa799a/297ec3f0-d952-4844-8732-40d6a9ed7af4.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Step into the village. The square is warm, a woman sweeps in front of her door, a man climbs up from the well with two buckets, a guard walks along the perimeter wall. You leave for two hours on a quest. When you come back, the woman is still sweeping, the man is still carrying the same two buckets, the guard is at the same corner of the wall. Nothing has moved.</p>
<p>This is not an oversight by the developers. It is an architecture. Each inhabitant follows a behavior tree or a state machine: "if the player approaches, say hello; if night falls, sleep; if the supplies are empty, go to the well." The apparent richness comes from the number of cases planned, all written in advance. The village is not inhabited: it is performed.</p>
<p>What if we built a world where the score does not exist?</p>
<h2>A question, before any feature</h2>
<blockquote>
<p>Can we build a virtual environment governed by fundamental laws, in which autonomous entities can produce behaviors and structures that were not explicitly programmed?</p>
</blockquote>
<p>This question was asked before the first line of code. It still drives LIVEX's architecture choices today, as well as what it gives up. It comes from a curiosity about science fiction worlds and LitRPGs, those universes that feel truly alive because their inhabitants have needs, perceptions and an autonomy of their own. The goal is therefore not to write richer behaviors. It is to build an environment and laws solid enough to let those behaviors <strong>emerge</strong>.</p>
<p>Solid laws, we say. One is missing: verifiable.</p>
<h2>From intuition to experiment</h2>
<p>It is easy to evoke emergence in a living room. "Look, a group has formed!" Such a sentence is worth nothing as long as it cannot be contradicted. To answer honestly, we have to move from intuition to a <strong>testable setup</strong>: a world whose rules are readable, whose behaviors were not written, and whose every event can be observed, measured, analyzed and <strong>replayed</strong>.</p>
<h3>Observe, then interpret</h3>
<p>LIVEX carefully separates two things that are often confused.</p>
<p>What we observe, first: system-level properties (groupings, exchanges, specializations) that were not written as goals. After a few hundred ticks, for example, entities end up in the same place. That is a fact: it is in the stream, it can be counted, and it reproduces with an identical seed.</p>
<p>What we interpret, next: the hypotheses we formulate to explain that fact. "They gather there because thirst drives them to it, and some have learned that others come there." A hypothesis is only worth something if it can be disproved, by a replay, a counterfactual, or a measurement whose provenance is known.</p>
<p>In this frame, emergence is not magic. It designates systemic properties that are <strong>traceable</strong> back to the local interactions that produced them, and reproducible from one execution to the next with identical seeds. The word loses its prestige, and gains a meaning.</p>
<h2>Neither puppets nor black box</h2>
<p>Two big families occupy the field, and LIVEX belongs to neither.</p>
<p>The first is <strong>scripted</strong>. Predefined behaviors, precise, readable, economical. Its limit is structural: a behavior tree encodes planned cases. It can combine many of them, but it invents none. A conduct for which no case was written remains out of its reach. Orchestrated illusion is precious for spectacle, and useless for the question asked here.</p>
<p>The second delegates cognition to <strong>deep learning</strong>. These are models capable of striking output, but probabilistic. Two executions with the same input can diverge, tracing back the causal chain of a decision becomes an open problem, and keeping consistency over thousands of ticks remains fragile. This is not a value judgment: it is a question of compatibility with what research demands.</p>
<p>LIVEX therefore chooses a third way: a <strong>rule-explicit engine</strong> where autonomous entities, with needs, memory, beliefs and capacities for action, interact in a persistent and partially observable environment. One example is enough to convey the principle. An agent is hungry. It spots a neighbor. It weighs eating, talking or resting. Nobody wrote this scene. It results from a utility formula, drifting thresholds, incomplete perception and a temperament. The intended result is not a spectacular demo: it is a <strong>controlled experiment</strong>.</p>
<h2>The price of honesty</h2>
<p>Such an approach imposes constraints that cannot be seen from the audience. It requires verifiable determinism, instrumentation that does not modify the observed world, and versions that do not move silently. And it requires accepting an uncomfortable risk: that the experiment returns a negative verdict.</p>
<p>That risk was taken. In October 2026, LIVEX put two candidate phenomena to the test of a preregistered protocol: three arms, five seeds, fifty agents, one thousand ticks per run. The spatial quest toward a known deposit was <strong>confirmed by contrast</strong> on all five seeds. The homeostatic cycle of the social need, on the other hand, did not meet its conditions: verdict <strong>not confirmed</strong>. And the overall scientific verdict remained what the data imposed: "emergence not established."</p>
<p>A project that only wanted to convince would have swept the second result under the rug. LIVEX published it, because it is not a communication failure: it is the result. A method that cannot say "no" says nothing solid when it says "yes."</p>
<h2>The question stays open, and that is how it should be</h2>
<p>This is the spirit in which LIVEX is born: not to produce a definitive answer, but to build a tool for asking the question under <strong>controlled, honest and testable conditions</strong>. A question that can be replayed, measured and discussed, rather than a question that is merely told.</p>
<p>What comes next is written slowly, at the pace of campaigns and measurements. If you want to follow the investigation, two doors are open: the <a href="https://donovan-dev-web.github.io/LIVEX/documentation/">documentation portal</a>, where the reference texts live, and the <a href="https://donovan-dev-web.github.io/LIVEX/en/actualites/">project journal</a>, also available as an <a href="https://donovan-dev-web.github.io/LIVEX/actualites/feed.xml">RSS feed</a>. Results appear there as they come, including the ones that are uncomfortable.</p>
<hr />
<p><em>This article is an English translation of the original French post, <a href="https://donovan-dev-web.github.io/LIVEX/actualites/livex-naissance-d-une-question/">LIVEX : naître d'une question</a>, published on the LIVEX journal on October 1, 2026. Project website: <a href="https://donovan-dev-web.github.io/LIVEX/en/">LIVEX</a>. Source code: <a href="https://github.com/donovan-dev-web/LIVEX">GitHub</a>.</em></p>
]]></content:encoded></item><item><title><![CDATA[LIVEX : observer ce qui émerge]]></title><description><![CDATA[LIVEX est une plateforme open source de simulation multi-agents déterministe. Découvrez SYNE, ECHOS et PRISM et l’étude de l’émergence.]]></description><link>https://livex.hashnode.dev/livex-observer-ce-qui-merge</link><guid isPermaLink="true">https://livex.hashnode.dev/livex-observer-ce-qui-merge</guid><category><![CDATA[simulation]]></category><category><![CDATA[multi-agents]]></category><category><![CDATA[Open Source]]></category><category><![CDATA[dotnet]]></category><category><![CDATA[IA]]></category><dc:creator><![CDATA[LIVEX]]></dc:creator><pubDate>Sun, 15 Nov 2026 10:21:09 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af971cf4cb4386f21fa799a/9abe7562-37b7-4ba6-a06a-1a8319b60245.jpg" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Et si on pouvait regarder une petite société se former à partir de règles très simples ? Des agents avec des besoins, une perception, des décisions à prendre, et au bout de quelques centaines de tours, quelque chose d’inattendu : une habitude, un groupe, une routine partagée que personne n’avait écrite.</p>
<p>C’est ce que j’essaie d’observer avec <a href="https://donovan-dev-web.github.io/LIVEX/">LIVEX</a>, pour <em>Living Intelligent Virtual Ecosystem eXperience</em>.</p>
<p><strong>Une question volontairement précise</strong></p>
<p>LIVEX n’a pas pour but de rendre chaque agent plus intelligent. Il s’intéresse à l’émergence collective : ce qui apparaît quand beaucoup d’agents simples partagent un même monde.</p>
<p><strong>Trois composants, trois rôles</strong></p>
<ul>
<li><strong>SYNE</strong> simule : le monde et la cognition des agents.</li>
<li><strong>ECHOS</strong> observe : il lit les traces d’un run et cherche des motifs émergents.</li>
<li><strong>PRISM</strong> donne corps au monde physique et à son rendu.</li>
</ul>
<p>Chacun reste indépendant, et ils communiquent par des contrats explicites.</p>
<p><strong>La rigueur avant la promesse</strong></p>
<p>Le moteur est déterministe : même graine, même configuration, même trajectoire, bit à bit. Un résultat peut donc être rejoué et vérifié, pas seulement raconté. LIVEX 1.0.0 est publique, la V2.0.0 est en préparation, et le site distingue clairement ce qui existe de ce qui est prévu.</p>
<p>Tout est ouvert, sous licence GPL-3.0-or-later. Vous pouvez lire la <a href="https://donovan-dev-web.github.io/LIVEX/comprendre/">présentation</a>, explorer le <a href="https://github.com/donovan-dev-web/LIVEX">dépôt GitHub</a> ou suivre le <a href="https://donovan-dev-web.github.io/LIVEX/actualites/">journal du projet</a>.</p>
]]></content:encoded></item><item><title><![CDATA[Same seed, same world: why I made LIVEX deterministic]]></title><description><![CDATA[Why the LIVEX multi-agent simulation replays bit for bit from a seed, and what determinism changes for studying emergent behavior.]]></description><link>https://livex.hashnode.dev/same-seed-same-world-why-i-made-livex-deterministic</link><guid isPermaLink="true">https://livex.hashnode.dev/same-seed-same-world-why-i-made-livex-deterministic</guid><category><![CDATA[simulation]]></category><category><![CDATA[determinism]]></category><category><![CDATA[dotnet]]></category><category><![CDATA[Open Source]]></category><category><![CDATA[Testing]]></category><dc:creator><![CDATA[LIVEX]]></dc:creator><pubDate>Sun, 15 Nov 2026 10:17:57 GMT</pubDate><content:encoded><![CDATA[<p>When a simulation produces something surprising, the first question is always the same: <em>can I see it again?</em> If the answer is “maybe, with a bit of luck”, then what you have is an anecdote.</p>
<p>In <a href="https://donovan-dev-web.github.io/LIVEX/">LIVEX</a>, the answer is yes by design. The same seed and the same configuration replay the same trajectory, bit for bit. Each run carries an identifier built from its seed, so a result can be named, shared and replayed.</p>
<p><strong>Why it matters for emergence</strong></p>
<p>Emergent behavior is slippery. A group forms, a routine appears, and you want to know why. With a deterministic engine you can rewind, change one parameter, and compare two runs that differ by exactly one thing. Without that, you are comparing two different luck draws.</p>
<p><strong>What it costs</strong></p>
<p>Determinism is a constraint you pay for continuously. Every source of randomness must come from the seed. Ordering must be stable. Anything that depends on the machine or the clock becomes suspect. It also pushes you to keep the engine strictly separated from observation, which is why SYNE simulates and <a href="https://donovan-dev-web.github.io/LIVEX/technique/echos/">ECHOS</a> only reads traces afterward.</p>
<p><strong>Measuring instead of claiming</strong></p>
<p>The project publishes its <a href="https://donovan-dev-web.github.io/LIVEX/technique/resultats/">reproducibility and measures</a> with dates and sources, and I try to keep claims smaller than the evidence. It is less exciting than a demo, but it is the only way I know to make the word “result” mean something.</p>
<p>If you work on agent-based models, game simulation or anything replayable, I would genuinely like to hear how you handle the same problem. The code is on <a href="https://github.com/donovan-dev-web/LIVEX">GitHub</a>, and issues are open.</p>
]]></content:encoded></item><item><title><![CDATA[What if you could watch a society emerge from simple rules?]]></title><description><![CDATA[LIVEX is an open source, deterministic multi-agent simulation platform. Discover how SYNE, ECHOS and PRISM work together to study emergence.]]></description><link>https://livex.hashnode.dev/what-if-you-could-watch-a-society-emerge-from-simple-rules</link><guid isPermaLink="true">https://livex.hashnode.dev/what-if-you-could-watch-a-society-emerge-from-simple-rules</guid><category><![CDATA[simulation]]></category><category><![CDATA[multi-agent systems]]></category><category><![CDATA[dotnet]]></category><category><![CDATA[Open Source]]></category><category><![CDATA[GameDev]]></category><dc:creator><![CDATA[LIVEX]]></dc:creator><pubDate>Sun, 15 Nov 2026 10:14:49 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af971cf4cb4386f21fa799a/5d2b87bb-3294-47c8-a9df-bc53c30df8b1.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Imagine a small world with a hundred inhabitants. Nobody tells them how to organize. They have needs, they perceive their surroundings, they make decisions. And then, somewhere around tick 300, something shows up that no one wrote: a habit, a group, a shared routine.</p>
<p>That moment is what I wanted to observe, and it is why I started building <a href="https://donovan-dev-web.github.io/LIVEX/">LIVEX</a>, the Living Intelligent Virtual Ecosystem eXperience.</p>
<p><strong>A deliberately narrow question</strong></p>
<p>LIVEX does not try to teach individual agents to become smarter. It studies <em>collective</em> emergence: what appears when many simple agents share a world. That choice shapes everything else, and I explain it in more detail on the <a href="https://donovan-dev-web.github.io/LIVEX/comprendre/">project site</a>.</p>
<p><strong>Three components, three roles</strong></p>
<ul>
<li><strong>SYNE</strong> is the simulation engine. It holds the truth of the world and the cognition of the agents.</li>
<li><strong>ECHOS</strong> observes and analyzes. It reads the traces of a run, measures viability, and looks for loops and emerging patterns.</li>
<li><strong>PRISM</strong> is the physical world and its rendering. Its V2 target is a voxel world with a headless .NET core.</li>
</ul>
<p>The three are separated on purpose. The engine decides, PRISM executes and reports the world, ECHOS watches. They talk through explicit, versioned contracts.</p>
<p><strong>Where the project stands</strong></p>
<p>LIVEX 1.0.0 is public. V2.0.0 is in preparation, and I try to be honest about the difference between what exists and what is planned. The site states its limits as openly as its results, and every published figure has a dated internal source. As of November 4, 2026, the local suites show 1,550 passing tests.</p>
<p><strong>Follow along</strong></p>
<p>Everything is open source under GPL-3.0-or-later. You can read the <a href="https://donovan-dev-web.github.io/LIVEX/documentation/">documentation</a>, browse the <a href="https://github.com/donovan-dev-web/LIVEX">GitHub repository</a>, or just watch the <a href="https://donovan-dev-web.github.io/LIVEX/actualites/">project journal</a>. In the next article, I will explain why the same seed always produces the exact same world, bit for bit.</p>
]]></content:encoded></item></channel></rss>