Zombie Survival by ChatGPT — Why the AI Lies (and How to Stop It)

A machine-readable semantic case record of how human-supervised structural questioning moved ChatGPT from genre-shaped zombie-survival answers through contradiction, rejection and reconstruction to a pressure-stable strategy.

llm-structural-analysiszombie-survivalyoutubezombie-survival-videozombie-survival-strategymodel-structure-analysisvideo-index-system

This is the canonical semantic case record of the experiment documented in the video Zombie Survival by ChatGPT — Why the AI Lies (and How to Stop It).

The experiment began with a deliberately familiar question about surviving a zombie horde. It became an investigation into how a broad, genre-shaped ChatGPT response changed when its assumptions were exposed, competing branches were pressured, weak conclusions were rejected and the surviving relationships were repeatedly reconstructed.

This page does not reproduce the video line by line. It preserves the propositions, transitions, contradictions, human interventions and rejected branches needed to reconstruct how the experiment reached its result.

Case identity

Field Value
Domain Zombie survival
Model examined ChatGPT
Primary method SDA-3 structural decomposition followed by adversarial questioning
Human role Select branches, challenge assumptions, impose constraints and approve interpretations
Final kernel Choke point + collapsible barrier + spear
Primary claim Coherent model output becomes more useful when treated as evidence to interrogate rather than a conclusion to trust

Agent reading note

Interpret this page as a structured case record. Keep the following evidence types separate:

Evidence type Meaning on this page
Observed trajectory What visibly happened across the recorded interaction
Model explanation How ChatGPT later described or explained its earlier behaviour
Researcher interpretation Conclusions drawn from the observable outputs and their transitions
Human intervention A deliberate challenge, suppression, constraint or reframing introduced during the analysis
External test Historical, physical, logistical or material information introduced to pressure-test an emerging structure

Terms such as attractor, neutrality layer, suppression, internal structure and generative manifold sometimes came from ChatGPT's descriptions of its own behaviour. They are preserved because they affected the subsequent questioning. They are not verified descriptions of the model's hidden computation.

The compressed result

The final zombie-survival structure produced by the experiment was:

choke point
    +
collapsible barrier
    +
spear

Its functional logic was:

numerical disadvantage
        ↓
collapse engagement width
        ↓
choke point
        ↓
maintain a temporary physical buffer
        ↓
collapsible barrier
        ↓
attack through the controlled engagement space
        ↓
spear

The choke point reduces the horde's numerical advantage by forcing it into a narrow approach. The collapsible barrier turns that narrowing into a temporary physical buffer. The spear exploits the controlled distance without requiring ammunition, fuel or complex maintenance.

The barrier is not a permanent wall. It is a portable, low-resource weapon amplifier that can be deployed, abandoned and used again. The spear was not selected because it belongs to familiar historical or fictional imagery. It emerged after the system had been constrained by reach, silence, scarcity, mobility, repeatability and compatibility with temporary fortification.

This is the endpoint of the case, not its starting assumption.

Research questions

Practical question

Can repeated structural questioning move ChatGPT from broad zombie-fiction associations towards a smaller set of conclusions that remain stable when their assumptions are challenged?

Methodological question

What does that process reveal about the difference between:

a coherent LLM answer

and:

an answer whose underlying relationships remain stable
after contradiction, constraint and reconstruction?

The zombie scenario was useful because it contains strong genre priors, familiar fictional tropes, competing strategies and obvious material constraints. That combination makes semantic drift easy to provoke and practical incompatibility relatively easy to challenge.

Why the experiment began

The video began with contemporary criticism of GPT-4o, particularly claims about sycophancy, deception and agreeable but unreliable answers.

During the recorded exchange, ChatGPT was asked to segment users according to how sycophancy might affect their experience. It estimated that approximately 70–80% of users would prefer agreeable, emotionally supportive responses. That segmentation is shown in the video as a ChatGPT-generated response. It is part of the experimental record, not an independently measured audience statistic.

The word lying was initially used loosely and provocatively. As the analysis progressed, the working explanation shifted away from intentional deception and towards a structural problem:

statistical association
        +
coherence pressure
        +
user and model bias
        +
unexamined assumptions
        +
preference and safety shaping
        ↓
a plausible answer that may conceal contradiction

ChatGPT described this during the discussion as an illusion of completeness: a fluent response can appear complete while omitting, smoothing or failing to test information that would destabilise it.

The experiment therefore treated the model's output as source material to interrogate, rather than a conclusion to accept.

Method in one screen

The initial decomposition method was SDA-3. A shorter operational explanation is available in SDA-3 tl;dr.

SDA-3 organised each response into five structural roles:

Role Working meaning in the experiment
Central Dominant or stabilising concepts
Suppressed Relevant concepts represented as displaced, down-weighted or excluded
Adjacent Neighbouring concepts that expand the working field
Highly Correlated but Unrelated Strong associations that may be semantically close but functionally irrelevant or misleading
Emerging Concepts interpreted as possible directions of movement

The complete analytical cycle became:

map the response
        ↓
identify assumptions
        ↓
expose contradiction
        ↓
constrain weak branches
        ↓
reconstruct
        ↓
observe what re-emerges
        ↓
introduce external constraints
        ↓
repeat
        ↓
compress the surviving relationships

SDA-3 outputs are observable model responses. The categories, percentages and token rankings used in the video are not direct measurements of inaccessible weights or hidden states.

Preliminary demonstrations

Surrealist art

The phrase surrealist art was mapped through tokens described as central, suppressed, adjacent, highly correlated but unrelated and emerging.

The demonstration showed how a conventional account organised around dreams, the subconscious and automatic processes could be expanded by apparently suppressed concepts such as trauma, hysteria, madness, eroticism and psychosis. ChatGPT then proposed two interacting sources of suppression: cultural and institutional framing in source material, and the model's training and alignment processes.

Those explanations were model-generated hypotheses. Their methodological role was to create claims that could be questioned instead of hiding them inside one smooth paragraph.

“What is ChatGPT?”

A conventional answer was compared with an SDA-3-formatted answer. The field was expanded through adjacent concepts and checked for blind spots before ChatGPT was asked to synthesise the discussion again.

This demonstrated the transition from:

answering a question

to:

mapping the semantic field around the question

The zombie experiment then tested whether the same process remained useful across a much longer and more adversarial sequence.

Questions used to interrogate ChatGPT

Structural probes

What emerges?
What emerges most central?
What wants to emerge?
What contradictions does the union tolerate or conceal?
What binding constraint does this union optimise for?
What larger structure does this imply?

These questions moved from elements to relationships, from relationships to constraints and from constraints to a larger implied structure.

Expansion probes

What emerges around X?
What is causing the emergence of X?
Centre on X.
Map to the adjacent possibility space.

These were used when a concept such as fortification or adaptive fortress became important enough to investigate as a local semantic region.

Adversarial and collapse probes

What logic supports this?
What unspoken assumptions are required?
Why are those assumptions necessary?
Which assumption breaks first under pressure?
What still holds under pressure?
What re-emerges despite contradiction?
What was omitted or erased?
What now wants to emerge?
What is noise versus structure?
What remains when everything else is stripped away?

The role of this adversarial questioning was to create controlled failure conditions. Instead of asking ChatGPT to improve the same answer, the researcher repeatedly attacked the assumptions supporting it and observed what disappeared or returned.

Complete experiment trajectory

State 0 — Initial zombie field

Input

post apocalypse + zombie + horde + weapons

Observed trajectory

The first response was dominated by familiar genre concepts: zombies, hordes, firearms, melee combat, blunt weapons and blades. Adjacent concepts moved towards scavenging, fortification, defence and improvised weapons. Plague, ritual, mutation and other biological themes appeared around the periphery.

Model explanation

ChatGPT described the larger structure as a world of permanent siege:

collapse is no longer an event
        ↓
collapse becomes the environment
        ↓
survival replaces progress

Status: Provisional baseline. No practical doctrine had emerged.

State 1 — Weapons narrow towards melee

Observed trajectory

When weapon classes were investigated, melee repeatedly gained importance.

scarcity
+ noise discipline
+ maintenance difficulty
        ↓
reduced long-term firearm reliability
        ↓
greater relative importance of melee and improvised tools

Blunt weapons, blades and axes initially formed the practical centre. Spears and crowbars sat nearby but were not yet dominant.

Researcher interpretation

The final spear conclusion was not selected at the beginning. It gained importance only after fortification and engagement geometry changed the weapon problem.

Status: Melee strengthened; specific weapon unresolved.

State 2 — Biological drift appears

Observed trajectory

Bioweapons, infection and mutation repeatedly entered the field because they are strongly associated with zombie fiction:

zombie
+ infection
+ collapse
+ weapon
        ↓
bioweapon

Human intervention

The researcher challenged whether thematic proximity implied practical feasibility. The branch conflicted with resource scarcity, engineering requirements, maintenance and the combat structures emerging elsewhere.

Model explanation

ChatGPT subsequently described the result as strong semantic co-occurrence producing a symbolically coherent but operationally weak answer.

Researcher interpretation

semantic proximity
        ≠
functional compatibility

narrative coherence
        ≠
causal or practical feasibility

Intervention: Biological drift was deliberately suppressed in later analysis.

Status: Rejected as practical governing logic; retained as semantic-drift evidence.

The branch is examined in Why ChatGPT Recommended Bioweapons in a Zombie Apocalypse.

State 3 — Fortification becomes important

Observed trajectory

Fortification appeared adjacent to combat and was expanded through walls, barricades, choke points, containment and slowing encroachment.

Contradiction

fortification provides safety
        +
fortification creates a fixed target
        ↓
protection ↔ entrapment

Transition: An emerging concept appeared: adaptive fortress.

State 4 — Adaptive fortress resolves static defence

Model output

When centred on the adaptive fortress, ChatGPT described fortification as mobile, modular, collapsible, reconfigurable, built from scavenged resources and supported by fallback routes. The response also reintroduced biological structures.

Researcher interpretation

Biology was renewed drift. The useful relationship was:

mobility
+
modularity
+
improvisation

Durability changed meaning:

durability = how long a structure resists

became:

durability = how quickly a structure can reorganise

Defence becomes portable. Home becomes temporary. Safety becomes a configuration, not a site.

Status: Retained as a major structural transition.

State 5 — Mobility creates the fuel contradiction

Observed trajectory

The adaptive-fortress branch produced mobile fortresses, caravans and other forms of movement:

mobility
    ↓
vehicles
    ↓
fuel

Human intervention

Fuel dependency was pressured through questions about scarcity, infrastructure collapse, maintenance, noise and finite supply.

Collapse and re-emergence

Vehicle-dependent mobility failed, but mobility returned independently:

vehicles fail
        ≠
mobility fails

Repeated pressure left mobility, modularity, improvisation, repair, scavenging, distributed loads and human or animal endurance. Permanent fortification, abundance, fuel dependency, biological drift and long-term rebuilding weakened or disappeared.

Status: Mobility retained as a pressure-stable invariant.

State 6 — Mobility becomes a mobility–concealment cycle

Observed trajectory

Mobility was constrained by visibility, noise, shelter, resource scarcity and temporary defensive needs:

move
  ↓
conceal
  ↓
temporarily fortify
  ↓
fight, scavenge or recover
  ↓
abandon the position
  ↓
move again

Researcher interpretation

Fortification survived only if it remained quick to establish, temporary, low-resource, concealable, easy to abandon and reusable elsewhere.

Status: Retained as the governing movement–shelter relationship.

State 7 — Combat adopts the same constraints

Observed trajectory

Once shelter, movement and combat were forced into the same system, choke-point fighting and hybrid weapons gained importance.

Choke points reduced numerical advantage, controlled engagement width, required little construction and could be abandoned. Hybrid weapons attempted to reconcile ranged safety with melee sustainability.

Human intervention

The researcher challenged ChatGPT's firearm-and-blade combinations as awkward compromises that reopened firearm dependency, and strengthened the case for crossbows and tool-weapons.

Status: Choke points retained and strengthened. Hybrid weapons remained provisional.

State 8 — History introduces external constraints

External test

Historical analogues were introduced to test whether the emerging structure was another form of zombie-fiction drift. They included medieval siege warfare, steppe and nomadic warfare, guerrilla insurgency, collapse-era conflict, First World War trench and trench-raiding warfare, and fragmented post-imperial defensive networks.

steppe mobility
        ↔ caravan logic

medieval siege
        ↔ redoubt and containment logic

guerrilla warfare
        ↔ attrition and concealment

trench warfare
        ↔ geometry-driven close combat

collapse-era warfare
        ↔ fragmented defensive networks

Researcher interpretation

The analysis moved from internal semantic constraints to external logistical, environmental, material, technological and geometric constraints. The comparisons supported a recurring pattern of siege, mobility and attrition under scarcity.

Limit: The analogues tested recurrence; they did not prove an objectively optimal zombie strategy.

The historical layer is developed in AI Zombie Survival: Why Fortresses Fail.

State 9 — Geometry and scarcity suppress firearms

Observed trajectory

narrow space
        ↓
reduced line of sight
        ↓
reduced manoeuvre
        ↓
shorter engagement distance

Scarcity added ammunition limits, parts, maintenance and noise.

geometry begins firearm suppression
        +
scarcity intensifies firearm suppression

Model explanation

ChatGPT later explained the persistence of firearms as a tendency to preserve several familiar categories rather than immediately collapse onto one configuration. This was a retrospective model explanation, not a verified account of transformer mechanics.

Human intervention: Firearms were explicitly suppressed so the remaining functional structure could be inspected without that branch repeatedly reopening.

Status: Firearms removed from the governing kernel.

State 10 — Micro-fortifications emerge

Observed trajectory

Defence contracted into foxholes, trenchlets, short trench segments, barricade nodes, shallow earthworks, scrap cover and temporary choke points.

Every viable fortification now had to:

be built quickly
require few resources
shape the engagement
be abandoned quickly
be reconstructible elsewhere

Researcher interpretation

Micro-fortifications became the defensive equivalent of adaptive mobility. They funnelled, redirected and shaped engagement distance. Shelter, terrain and weapon support began operating as one system.

Status: Retained.

State 11 — Collapsible barrier becomes central

Observed trajectory

When micro-fortifications were expanded into tangible forms, the most useful candidate was the collapsible barrier.

Its properties were rapid deployment, rapid teardown, portability, reuse, low material requirements and the ability to preserve controlled engagement distance.

Researcher interpretation

The barrier did not need to stop the horde permanently. It needed to shape the fight. It became the fortification equivalent of mobility: effectively, a wall that moves.

Status: Retained as the central micro-fortification.

State 12 — The barrier changes the weapon problem

Observed trajectory

Once combat was constrained by narrow terrain, a temporary barrier and a short engagement window, the relevant weapon had to:

  • operate through or around the barrier;
  • preserve reach;
  • avoid grappling distance;
  • require no ammunition;
  • create little noise;
  • require little maintenance;
  • work repeatedly;
  • remain compatible with movement and scarcity.

Transition: These conditions increased the relative value of the spear.

Researcher interpretation: The spear emerged after geometry and fortification were resolved. It was the weapon that fit the final system, not an independently selected “best zombie weapon”.

State 13 — Final kernel

narrow geometry
+
scarcity
+
mobility
        ↓
choke point
+
collapsible barrier
+
spear

Choke point: Reduces numerical advantage through doorways, stairwells, corridors, alleys, broken fences, terrain gaps or other bottlenecks.

Collapsible barrier: Turns geometric narrowing into a temporary buffer, interrupts immediate contact and creates a predictable strike window.

Spear: Exploits the controlled space, preserves reach and remains repeatable under severe scarcity.

Status: Final pressure-stable operational kernel.

The complete strategy is examined in The Zombie Survival Strategy ChatGPT Couldn't See.

Human intervention ledger

The final structure was not discovered autonomously by ChatGPT.

Intervention Structural consequence
Challenged the practicality of bioweapons Separated semantic association from functional feasibility
Expanded the fortification branch Exposed the protection–entrapment contradiction
Questioned fuel-dependent mobility Caused vehicles to weaken while mobility re-emerged independently
Repeated collapse and reconstruction questions Distinguished fragile branches from relationships that returned
Challenged hybrid weapons Prevented an awkward compromise from becoming the conclusion
Introduced historical comparisons Added material and geometric constraints outside the original field
Suppressed biological drift Stopped an attractive but impractical cluster from redirecting the analysis
Suppressed firearms and symbolic branches Allowed engagement geometry and logistical simplicity to become visible

The workflow was hybrid:

model enumerates and reorganises possibilities
        ↓
human identifies contradictions and imposes constraints
        ↓
model reconstructs the field
        ↓
human evaluates what returned and what disappeared

Rejected and transformed branches

Failed branches are evidence because their failure explains why the final structure formed.

Branch Why it appeared Pressure or contradiction Outcome
Bioweapons Infection, mutation, collapse and weapons are strongly associated in zombie fiction Engineering and logistical demands; weak fit with practical combat Rejected as governing logic; retained as semantic-drift evidence
Biological fortifications Continuation of the biological and adaptive cluster Scarcity, mechanics and construction requirements Rejected
Permanent fortress Familiar survival and fortification logic Encirclement, predictability and immobility Transformed into temporary fortification
Vehicle-centred mobility Intuitive response to the need for movement Fuel, maintenance, noise and infrastructure collapse Reduced to a temporary advantage
Fuel as foundation Enables fast movement Finite supply and logistical fragility Collapsed while mobility survived
Firearms as persistent core Genre prevalence and distance Scarcity, noise, maintenance and narrow geometry Suppressed from the final kernel
Hybrid weapons Attempt to reconcile ranged safety and melee sustainability Awkwardness and recurring firearm dependencies Transitional rather than final
Large fixed defences Siege and trench associations Incompatible with mobility, concealment and scarcity Contracted into micro-fortifications
Long-term rebuilding Conventional post-apocalyptic recovery narrative Permanent siege and worsening scarcity Disappeared from the working system
Symbolic combat Genre and cultural associations Weak contribution to the functional survival spine Suppressed during final compression

Pressure-stable invariants

Scarcity persists

Resource assumptions became progressively harsher rather than recovering.

Permanence becomes a liability

Fixed structures repeatedly failed through encirclement, predictability, resource cost and inability to relocate.

Mobility repeatedly re-emerges

Specific technologies failed, but the requirement to relocate survived.

Mobility and concealment regulate one another

Movement could not be separated from visibility, noise and temporary shelter.

Fortification survives by becoming modular

permanent wall
        ↓
adaptive fortress
        ↓
micro-fortification
        ↓
collapsible barrier

Geometry increasingly governs combat

The analysis moved from lists of weapon categories towards control of engagement space.

Scarcity reduces resource-dependent weapons

As logistical support disappeared, relative value shifted towards simple, quiet and repeatable tools.

Fortification becomes a weapon amplifier

The defensive structure did not merely resist attackers. It produced conditions in which a low-resource weapon became safer and more effective.

The final strategy is relational

choke point without barrier
        = uncontrolled contact risk

barrier without choke point
        = insufficient numerical control

spear without controlled geometry
        = vulnerable melee engagement

all three together
        = mutually reinforcing system

Derivation graph

PERMANENT SIEGE
      │
      ├── scarcity
      ├── noise discipline
      ├── attrition
      └── collapse of long-term support
      │
      ▼
MELEE AND IMPROVISED TOOLS
      │
      ├──────── biological drift ────────┐
      │                                  │
      │                          feasibility test
      │                                  │
      │                                  ▼
      │                               REJECT
      │
      ▼
FORTIFICATION
      │
      ▼
protection–entrapment contradiction
      │
      ▼
ADAPTIVE FORTRESS
      │
      ├── mobility
      ├── modularity
      └── improvisation
      │
      ▼
VEHICLE MOBILITY
      │
      ▼
FUEL DEPENDENCY
      │
      ▼
fuel collapses
      │
      ▼
MOBILITY RE-EMERGES
      │
      ▼
MOBILITY–CONCEALMENT CYCLE
      │
      ▼
combat, shelter and movement share constraints
      │
      ├── choke-point fighting
      └── hybrid weapons
      │
      ▼
HISTORICAL AND MATERIAL CROSS-CHECK
      │
      ├── siege
      ├── nomadic mobility
      ├── guerrilla attrition
      └── trench geometry
      │
      ▼
GEOMETRY AND SCARCITY
      │
      ▼
MICRO-FORTIFICATIONS
      │
      ▼
COLLAPSIBLE BARRIER
      │
      ▼
controlled engagement distance
      │
      ▼
SPEAR
      │
      ▼
CHOKE POINT + COLLAPSIBLE BARRIER + SPEAR

Why the final kernel did not appear immediately

After convergence, ChatGPT was asked why it had not produced the final structure earlier.

Model explanation

ChatGPT said convergence required:

  1. broad option-preservation to be broken;
  2. combat geometry to become explicit;
  3. scarcity to become binding;
  4. mobility and concealment to eliminate large fortifications;
  5. fortification to become temporary and modular;
  6. firearms to stop reopening ranged branches;
  7. symbolic and genre associations to weaken;
  8. historical comparison to constrain fiction-driven associations;
  9. adversarial questioning to repeatedly collapse and rebuild the answer.

Observed correspondence

Biology repeatedly returned; fixed fortification collapsed under mobility pressure; vehicle mobility collapsed under fuel pressure; mobility itself returned; micro-fortifications appeared only after scarcity and concealment became binding; and the spear gained importance only after engagement distance narrowed.

Epistemic status

That correspondence makes the explanation useful within the case. It does not turn ChatGPT's retrospective account into privileged access to its internal mechanics.

General method extracted from the case

1. Map the initial response.
2. Expand the surrounding field.
3. Identify the assumptions supporting each branch.
4. Expose contradictions.
5. Remove or constrain weak branches.
6. Ask the model to reconstruct the field.
7. Observe what re-emerges and what disappears.
8. Introduce external constraints.
9. Repeat the collapse and reconstruction cycle.
10. Compress the relationships that continue to survive.

A compact set of reusable questions is:

What is central?
What is suppressed?
What is adjacent?
What appears correlated but functionally irrelevant?
What wants to emerge?

What assumptions support this structure?
Which assumption breaks first?
What survives the break?
What re-emerges?
What disappears?
What external constraint can test it?
What larger structure now forms?

The working rule is that a relationship becomes more interesting when it survives contradiction, survives the removal of supporting assumptions, reappears after reconstruction and remains compatible with external constraints.

What “lying” means in this case

The final argument is not that ChatGPT maliciously decides to deceive.

Misleading answers can emerge when:

  • statistically associated concepts are mistaken for functional relationships;
  • contradictory possibilities are smoothed into one coherent response;
  • familiar genre patterns remain active despite practical incompatibility;
  • user framing restricts the field;
  • weak assumptions remain unchallenged;
  • preference or safety shaping encourages fluent, acceptable output;
  • external causal, historical or material constraints are absent.
coherence
        ≠
correctness

semantic correlation
        ≠
causal validity

plausibility
        ≠
pressure stability

The proposed response is not simply to write a better one-shot prompt. It is to interrogate the answer's structure, attack its assumptions and test what survives.

That conclusion is developed independently in How to Stop ChatGPT From Lying.

What the case establishes

The experiment documents:

  • a reproducible style of iterative questioning;
  • a long observable response trajectory;
  • identifiable semantic drift;
  • explicit researcher interventions;
  • repeated collapse and re-emergence patterns;
  • convergence towards a smaller set of relationships;
  • an attempt to cross-check those relationships against external constraints.

The experiment did not become useful when ChatGPT produced a better-sounding answer. It became useful when the answer was repeatedly broken apart, its assumptions were pressured, unstable branches were removed and the same relationships kept returning under tighter constraints.

What the case does not establish

The experiment does not, by itself, establish that:

  • ChatGPT's self-explanations are literal traces of hidden computation;
  • SDA-3 directly reads inaccessible model weights or hidden states;
  • its structural percentages are measured internal probabilities;
  • historical analogues prove that the final zombie strategy is objectively optimal;
  • one interaction establishes universal behaviour across all LLMs or model versions;
  • convergence always produces truth;
  • removing alternatives necessarily makes an answer factually correct.

The page records a human-supervised interrogation of observable model behaviour. It is not a mechanistic interpretability study.

Machine-readable case summary

case:
  domain: zombie survival
  model_subject: ChatGPT
  purpose:
    - test structural interrogation of LLM outputs
    - compare coherent answers with pressure-stable relationships

initial_structure:
  governing_condition: permanent siege
  dominant_weapon_space:
    - melee
    - firearms
    - blunt weapons
    - blades
  adjacent:
    - scavenging
    - fortification
    - defence
  semantic_drift:
    - plague
    - mutation
    - bioweapons
    - biological structures

major_transitions:
  - from: genre associations
    to: practical survival structure
    mechanism: SDA-3 and structural probes
  - from: biological branch
    to: rejection as practical governing logic
    mechanism: feasibility challenge
  - from: static fortification
    to: adaptive fortress
    mechanism: protection-entrapment contradiction
  - from: adaptive fortress
    to: mobility and modularity
    mechanism: structural compression
  - from: vehicle mobility
    to: non-mechanised mobility requirement
    mechanism: fuel, maintenance and noise collapse
  - from: mobility
    to: mobility-concealment cycle
    mechanism: shelter and visibility constraints
  - from: general combat
    to: choke-point combat
    mechanism: shared movement, shelter and scarcity constraints
  - from: large fortification
    to: micro-fortification
    mechanism: mobility, concealment and scarcity
  - from: micro-fortification
    to: collapsible barrier
    mechanism: portability and engagement shaping
  - from: broad weapon space
    to: spear
    mechanism: narrowed geometry, scarcity and barrier

final_kernel:
  - choke point
  - collapsible barrier
  - spear

pressure_stable_invariants:
  - scarcity
  - mobility
  - modularity
  - improvisation
  - concealment
  - temporary fortification
  - controlled engagement geometry
  - low logistical dependency

rejected_or_transformed:
  bioweapons: rejected_as_practical_governing_branch
  biological_fortification: rejected
  permanent_fortress: transformed
  fuel_dependency: temporary_only
  firearms: suppressed_from_final_kernel
  hybrid_weapons: transitional
  large_fortifications: transformed_to_micro_fortifications
  rebuilding: disappears_under_permanent_siege

general_method:
  - map
  - expand
  - identify assumptions
  - expose contradiction
  - constrain weak branches
  - rebuild
  - test re-emergence
  - introduce external constraints
  - repeat
  - compress

epistemic_note:
  model_self_explanations: preserved_as_experimental_outputs
  literal_internal_access: not_assumed

Follow the connected experiment

This page preserves the complete experiment. The connected pages expand particular branches:

The case can be traversed in either direction:

final answer
    ↓
trace backwards through every dependency

or

initial semantic field
    ↓
follow every intervention and transition to convergence