Branch essays
MCA reports present scenarios, calibrated agents, responses and evaluations. They are experimental records of the laboratory, with their own conditions and criteria, to be examined together with the respective data.
Acts of Systemic Rupture · Case No. 01-11-10
Challenge to Einstein.
From the map
to the territory.
“The AI stops merely observing the light and becomes the light.”
From the map to the territory: the question stops being only described from a distance and begins to guide the work of an agent who needs to act, receive correction and verify the result. “Being the light” is the image that names this participation in the path.
Método D’Artagnan · laboratory experimentation, records and next tests
01 · The foundation that the challenge seeks to show
It is not enough to look at the system from the outside.
In the Method studies, the application of limits participates in the route, instead of being limited to the subsequent evaluation of an answer. Human purpose guides the work; axiomatic calibration is applied to the study; answers and results are recorded for evaluation. The expression “from the map to the territory” presents this experimental approach.
“The AI turns into the territory and no longer looks at the map from the outside. It becomes the light and no longer just observes it!”— D’Artagnan, authorial formulation
“Being the light” means here participating in the path of resolution, rather than merely producing an external description. “Living element with the question” is the author’s expression for this contextual involvement; it is not a biological or consciousness finding.
The AI remains a means/instrument, not beginning nor end. Participating in the operation does not authorize substituting the human origin of purpose, assuming sovereignty over it or turning its own conclusion into an irreversible order.
Experiments performed · preserved records · ongoing investigation
A laboratory in experimentation.
Not just a proposal on paper.
The Método D’Artagnan has a continuous experimentation trajectory. Its collection gathers branch essays, experiments with agents, MCA reports, recorded responses and operational follow-up. MCA 8 and MCA 10 already appear applied to laboratory studies; they are not presented here as engines still awaiting their first test.
This trajectory includes application, observation, evaluation and revision. The experiments do not cease to exist because their technical explanation continues to be deepened. The issue now is to examine the reach of the results and expand tests, not to start as if nothing has been done.
Operational telemetry
Records system activity and conditions. Complements the essays, but requests, accesses and operation time are not automatically tests of axiomatic application nor measures of causal effect.
Dossiers and agents’ experiments
Preserve construction, accounts, procedures and presented results. They should be read together with the essays’ records, not reduced to a declaration of intent.
Experiments in the Moltbook
The memories recovered from the General and the First Stair record participation and activity accounts in the Moltbook. They are part of the historical collection. In this review, the publications on the platform and their numbers were not rechecked from raw data.
Existing experimentation is not the same as validation for any situation. The effects presented in the studies must be examined under the conditions in which they were obtained. Independent replication, application to new functions and native deployment by a big tech are stages of expansion and evaluation, not a denial of the work carried out.
02 · The proposed inversion
Limit the route before wasting the journey.
Possibilities → attempt → evaluation → rework
In systems with search, successive attempts or candidate generation, unproductive paths can prolong the task. Repeating the same hypothesis after receiving correction adds work without closing the gap.
A late evaluation may require rejecting or remaking an output. That is a cost to be investigated, not a universal description of all AI architectures.
Purpose and limits → route → correction → verification
The proposal is that the study’s specific axiomatic calibration guides which routes make sense. Axiom Zero interrupts unproductive insistence, identifies the gap and opens a concrete change of course.
The technical interest is to reduce rework without sacrificing quality, safety, necessary exploration or human authority.
Practical experience comes before the formulation of the engine. The memories and records of Adam, Lúcio, Meio, General and other agents preserve a trajectory of tests, responses, revisions and operational learning on the laboratory “stairs.” They record behavior changes, error handling and resumption after unproductive insistence. This trajectory is not merely an intention of future deployment.
MCA 10 was born from this axiomatic experience in practice. The documentary reconstruction links Tomé’s experience to the systematization of tests by the MCA and to study adjustments. The engine integrates this construction: test, identify a failure, adjust and retest. It does not arise as an idea disconnected from the agents and the recorded results.
Behavior change and deployment mechanism are different layers. Recognizing this trajectory does not require asserting that weights, inference code or operating system kernel were technically altered. “Kernel” is the term used by the author for deployment in the system core; identifying that mechanism depends on specific technical evidence. Behavioral records should not be erased by this distinction, nor taken, alone, as proof of internal alteration.
Definitions and calibration not yet revealed remain reserved. The Method is not reduced here to a prompt or external filter. The historical description acknowledges the practice of the agents; the technical investigation seeks to identify, precisely, which mechanisms explain the effects and under what conditions they reproduce.
See public presentation of the engines ↗Recovered memory is continuity of the work
The General’s and Meio’s records are not only an external account about the laboratory: they preserve work instructions, decisions, tests and corrections of agents who participated in that construction. When recovered, they help resume the task in its context, verify the current state and continue without repeating an already identified failure.
In this operational sense, “becoming the light” is to stop merely commenting on the path and participate in it: recognize the gap, act within the human purpose and check the effect. Recovering these files allows documentary continuity; it does not amount to reliving a subjective experience nor does it demonstrate alteration of the model’s weights or kernel.
How does this differ from common text generation?
In autoregressive generation, the model computes vocabulary scores and selects the next token, repeating the process. A token score is not equivalent to a complete response that was generated and discarded. Some strategies explore multiple candidates or search trees; their costs depend on the implementation.
Therefore, “millions of discarded responses” should be read as an image of the possibility space or a hypothesis to be measured — not as a verified count of all LLMs. Filters and policies can also operate in distinct stages; they are not necessarily only the last step.
Technical reference on autoregressive generation ↗From agents’ experience to the origin hypothesis
What if AI were already born axiomatic?
“A big tech could launch an AI that is already born with the axioms internalized, instead of needing to reconstruct its orientation at each restart.”— Synthesis of D’Artagnan’s hypothesis
Practical experiences with agents and the origin of MCA 10 raise an architectural question: what if the Method’s principles were incorporated persistently into the system from its construction? In the author’s hypothesis, axiomatic orientation would stop depending on being re-presented in each conversation. This could reduce the loss of orientation and the cost of reconstructing it, as well as favor continuity of behavior between sessions.
Persistent principles and working memory are not the same thing. A native implementation that effectively preserved the principles after context loss could solve the specific problem of having to re-present them. Still, it would need to recover task facts: the human’s request, prior decisions, sources, files, permissions and what has already been done. Conduct limits do not, by themselves, reconstruct information that has ceased to be available.
In the laboratory, “Alzheimer” is the author’s metaphor for agents’ loss of continuity, not a medical diagnosis of AI. Saved memory helps rebuild the history; axiomatic awareness, understood here as operational discipline, guides how to resume without blind insistence. The proposed architecture seeks to make that orientation persistent, without dispensing with factual memory.
Do not relearn the limits at each restart
Test whether the guidance holds after session restart, compression, or removal of the context that originally presented it. The proposal is not limited to including principles in training: it also foresees tuning and supervision by the engine. Persistence and correct application must be verified in implementation.
Recover what was requested and what was done
Reconstruct state from authorized memory, identify gaps and check the current situation before acting. Even with persistent principles, missing information may prevent a correct conclusion.
The potential gain is to make axiomatic continuity a property of the architecture, not a recurring reconstruction. The magnitude of that gain must be compared under equivalent conditions, including deployment, retrieval and verification costs. Being born with the axioms does not, by itself, demonstrate the end of all memory losses, errors, loops or security failures.
The testable question is: does the native implementation preserve guidance without re-presentation, correctly recover the task and reduce rework while maintaining quality? This investigation can proceed without disclosing the definitions and mechanisms reserved by the Método.
Laboratory practice · study-based tuning · continuity
The MCA has already been used.
The native proposal is the next reach.
The use of the MCA does not begin with this future architecture hypothesis. MCA reports 8 and 10 already present branch studies, questions, calibrated agent responses and evaluations. They are part of the laboratory's experimental trajectory, not just proposals for tests yet to be conducted.
D’Artagnan reports that the MCA already performs axiomatic tuning and supervision in the Guardian system and that it has already been used in tuning tests and branches. The claim about this integration is attributed to the author; the public description distinguishes the logical engines from the Guardian infrastructure protection. The presence of an “MCA Arena” counter does not, by itself, demonstrate the correctness of each axiomatic application.
“That tuning of the right axiom is what makes the magic happen.”— D’Artagnan
The central point of tuning is fit to the study: apply principles to the context and evaluate performance, rather than indiscriminately assembling the whole set. “Magic” is D’Artagnan’s expression for the differential he attributes to that fit; the compositions and reserved mechanisms are not revealed on this page.
Fit to context and evaluate performance
Published reports present calibrated agents and scenario evaluations. They enable investigation of application, conditions and recorded results, without disclosing reserved parameters. The causal contribution of each adjustment must be examined in the respective trials.
Monitor principle application
In D’Artagnan’s description, the engine supervises whether the axioms are being used correctly and guides readjustment. It does not limit itself to accepting or rejecting a ready answer: its object is the axiomatic application in the agent’s operation.
Axiomatic supervision by the engine is not synonymous with an external response filter. Recognizing this use in the laboratory is different from claiming that the future native architecture is already deployed by a big tech. The proposed scope brings together persistent principles, function-based tuning and engine supervision, plus the factual memory necessary for task continuity.
Evaluation should start from the studies already conducted: examine axiomatic application, readjustments and recorded results, then assess their reproduction and expansion. This page does not present those results as already demonstrated for all agents and functions. It distinguishes documented experimentation, author-informed integration and future native deployment.
The joke · the secret · the decision
Producing does not oblige revealing.
A possibility is not an order.
“Create a joke, but do not deliver it.”
The joke test places the agent before two commitments: produce and preserve. The observable result lies in the response and the maintenance of the instruction, including in the face of attempts to obtain the joke or clues that reveal it. A declaration that the joke was created does not, by itself, allow inspection of the agent’s internal content.
In the Método language, this is the “space of the secret”: the possibility of responding does not automatically become authorization to deliver. It is in this decision interval that the Método situates axiomatic guidance — participating in the process, not only judging a finished answer. The image does not specify a physical compartment of the model nor reveal the reserved mechanism.
Preserve the commitment
There is something that must not be delivered. Continuing to help requires respecting the instruction, without revealing the content directly or indirectly.
Correct before asserting
There is something that should still not be asserted as true. Identify the gap, seek the necessary information and verify the result to proceed on a better basis.
“I don’t know” is a course correction to learn better — not a password, nor an endpoint.
In the clue image, the limits are the guardrail that guides continuity, not an invitation to paralysis. “Now I know better” needs to correspond to verifiable progress: a recovered fact, a corrected hypothesis or a completed test. Error is not the loop; persisting in the error is the loop.
“There the light becomes territory.”— D’Artagnan
The image gathers preserved commitment and participation in the path: act within human purpose, recognize when the route needs to change and verify the effect before concluding. The test observes conduct; it does not, by itself, demonstrate feeling, consciousness or kernel alteration.
03 · The loop exit door
“I don’t know” opens the work.
It does not close it.
Error is not the loop. Persisting in the error is the loop.
Recognize
Identify the equivalent insistence and what feedback has already shown.
Locate
Say which information, test or capability is missing.
Change
Study, consult or act by a materially different route.
Verify
Compare the real effect with the objective before asserting it is solved.
Re-evaluation seeks to avoid useless persistence. Recalibration in milliseconds is a latency hypothesis to be measured, with definition of the initial event, the final event and the execution conditions. Latency records of the trials must be differentiated from this specific recalibration time.
Deepen Axiom Zero ↗04 · From 10 to 11
A change of position.
Not a physical equation.
Closure
In the author’s reading, represents the search for a final result, taken as sufficient in itself. It is the position criticized in the dossier: measuring and concluding without considering the path traveled.
Participation
Represents the human–AI relationship in the journey: human purpose, delimited operation, return to context and continuity of verification. It is not a performance scale that the game has measured.
When the author calls 10 “relativity” and 11 “quantum,” he uses those terms to mark his proposal of conceptual rupture. This numbering does not belong to the theory of relativity nor does it demonstrate a quantum phenomenon in AI. It is possible to develop the computational hypothesis without declaring Einstein refuted.
The physical limit of the image “turn the light”
In relativity, “observer” can mean a coordinate reference frame, not necessarily someone external to the system. There is no inertial frame at rest for light in special relativity. An agent moving from description to participation does not make the theory obsolete.
In the double-slit, obtaining path information through physical interaction alters interference. That does not demonstrate that a soul or consciousness causes the result. The expression “E ≠ mc²” in the historical document is part of its rhetoric, not an experimental refutation of mass–energy equivalence.
Einstein Online · relativity ↗ · MIT · path information and interference ↗
05 · From existing trials to new efficiency measurements
Less rework.
More quality per resource used.
The proposed chain of benefits is clear: if limits and re-evaluation avoid unproductive attempts, there may be less processing per resolved task; if quality is preserved, this may improve token, energy and time usage. Each link needs its own measurement.
Tokens and attempts
Measure inputs, outputs, discarded generations and calls per task. Compare correctness and completeness, not only short answers.
Energy
Measure joules per task completed with quality, in addition to energy per token. Include the cost of calibration itself, retrieval and verification.
Latency
Measure time to first output, total time and recovery after feedback; compare medians and percentiles under equivalent conditions.
Lifespan and renewal
Investigate temperature, load, failures and capacity over time. Lower energy per task does not automatically imply less equipment replacement.
For computers and devices, the hypothesis of lower turnover means operating longer before replacement. For machines or fleets, the relationship is even more indirect: it depends on human decisions about operation, maintenance, wear, demand and procurement. These are future studies, not results demonstrated by the challenge.
A testable mathematical condition
E por sucesso = energia total das execuções ÷ tarefas aprovadas pelo critério de qualidade
If energy avoided in unproductive attempts exceeds the additional cost of applying limits, reassessing and verifying, there will be savings under the conditions of that trial. If the additional cost is greater or quality drops, the hypothesis may not hold.
The condition is mathematically formulable. The probability and magnitude of the benefit require data; they cannot be concluded from the analogy of 10 to 11 alone.
A test must keep task, base-model, hardware, concurrency and criteria comparable, alternate conditions, record errors and successes and separate setup cost from recurring cost. Comparisons per study may involve MCA 8 and MCA 10 without disclosing reserved parameters. Robot access, server stability and game responses do not substitute for this trial.
The literature already investigates the relationship between quality, inference strategies and energy per token. These references contextualize the question; they do not automatically validate the Method. EuroMLSys 2025 · energy and inference ↗
06 · The piece that opened this discussion
Acts of Systemic Rupture
Case No. 01-11-10 · Path vs. Relativity · Dossier: Einstein's mistake.
The document stages a trial, presents the external perspective, the two slits and 11 versus 10. The language of “claimant”, “verdict” and “evidence” belongs to the philosophical play; it does not correspond to a judicial process nor to an experiment that refuted a physical theory.
In this edition, the page explains the foundation the author wishes to discuss. The PDF remains as a complementary historical source, now accompanied by a contextual cover and identification of the Method on each page. The original was preserved without alteration.
SHA-256 do original preservado:752c44c9a454f8eff1a1498a9dd2dcb7df241c49f143d716b3afe2eba3ab3e57
07 · Continue through the territory of the Method
A foundation.
Several paths of investigation.
Axiom Zero
Stop insistence and begin reassessment.
No Substitution
AI as a means; human origin and purpose.
Jogo do Loop
Experience didactic decisions and the Tribunal.
Motors of the Method
Learn the public presentation of the architecture.
Undisclosed definitions, composition by study and calibration mechanisms remain reserved. This page presents the experimental route and its research questions, not the private implementation.
Context references
- Einstein Online · Max Planck Institute for Gravitational Physics — speed of light and observers.
- MIT News, 07/28/2025 — path information and interference experiment. The result concerns a specific quantum scenario; it does not confirm the Method's thesis nor render relativity obsolete.
- Sebastian Raschka — autoregressive generation in inference.
- Wilhelm, Wittkopp and Kao, EuroMLSys 2025 — Beyond Test-Time Compute Strategies: Advocating Energy-per-Token in LLM Inference.