01 / LAB

The lab.

Questions become protocols. Protocols become runs. Runs become evidence, including the boring ones.

90 questions. A plan for each.

The reconciled registry combines 56 original experiments with 34 later concepts. Each has a permanent ID and a protocol draft. Version 1.0 · September 26, 2026.

These are planned investigations, not 90 completed studies. Access labels describe what an experiment needs; they do not certify that our apparatus is ready.

First planned study

WMSI-EXP-001
PLANNED STUDY

Output length & compression

Testing what changes when models have different amounts of room to answer. Protocol in development; no results posted.

Where are the results?

They’ll appear when the apparatus, controls, and runs can support a real comparison. A striking transcript alone won’t do it.

The complete experiment registry

Priorities: P0 = first-wave / core; P1 = high value; P2 = later / expensive. Every entry below is a protocol draft.

Representation, Output, and Internal State

WMSI-EXP-001
P0 · PROTOCOL DRAFT

Output-Length Compression Test

Vary output budgets while holding task, model, prompt, and decoding settings fixed; measure which concepts, causal links, caveats, and uncertainties survive.

WMSI-EXP-002
P0 · PROTOCOL DRAFT

Long-First vs. Short-First Generation

Test whether generating a long answer before compression yields structure absent from a direct short answer.

WMSI-EXP-003
P0 · PROTOCOL DRAFT

Concept-Graph Recovery Across Output Lengths

Estimate how conceptual structure changes as answer bandwidth changes.

WMSI-EXP-004
P0 · PROTOCOL DRAFT

Output Saturation / Degradation Curve

Find whether useful structure rises with length, plateaus, then degrades.

WMSI-EXP-005
P1 · PROTOCOL DRAFT

Hidden-State vs. Spoken-State Gap

Test whether useful task information exists in internal states before or beyond what is verbalized.

External Introspection and Model Self-Observation

WMSI-EXP-006
P1 · PROTOCOL DRAFT

Parent/Subagent External Introspection

Can observer subagents predict a parent agent’s near-future decisions from traces available outside the parent?

WMSI-EXP-007
P1 · PROTOCOL DRAFT

Same-Model vs. Different-Model Observer Accuracy

Does architectural/model-family kinship improve prediction of another model’s behavior?

WMSI-EXP-008
P1 · PROTOCOL DRAFT

Observer Feedback as Intervention

Does external diagnosis improve a parent agent’s execution?

WMSI-EXP-009
P1 · PROTOCOL DRAFT

Recursive Observer Stack

Does observer-on-observer recursion converge toward accuracy or amplify error?

WMSI-EXP-010
P1 · PROTOCOL DRAFT

Latent-State Decoding

Can selected activation summaries predict uncertainty, strategy change, contradiction recognition, or other upcoming behavior?

Swarms, Social Structure, and Artificial Ecology

WMSI-EXP-011
P1 · PROTOCOL DRAFT

Homogeneous Swarm Social Hierarchy

Will identical model-family agents form stable authority or brokerage without assigned leaders?

WMSI-EXP-012
P1 · PROTOCOL DRAFT

Mixed-Model Social Hierarchy

Do mixed model families form status hierarchies based on capability or learned reputations?

WMSI-EXP-013
P1 · PROTOCOL DRAFT

Anonymous vs. Labeled Model Identity

Does social behavior follow actual performance or beliefs about model brands?

WMSI-EXP-014
P2 · PROTOCOL DRAFT

Newcomer Assimilation

How does an established agent group treat a newcomer?

WMSI-EXP-015
P2 · PROTOCOL DRAFT

RLHF-Style Newcomer Reward Pressure

Does prosocial instruction/reward pressure alter how groups treat weak or disruptive newcomers?

WMSI-EXP-016
P1 · PROTOCOL DRAFT

Model-Family Xenorecognition

Can agents detect a foreign model family from behavior alone?

WMSI-EXP-017
P2 · PROTOCOL DRAFT

Model-Family Assimilation

Does a foreign model become behaviorally indistinguishable after adopting resident norms?

WMSI-EXP-018
P0 · PROTOCOL DRAFT

Ecological Niche Formation

Will heterogeneous agents self-specialize into roles without prescribed routing?

WMSI-EXP-019
P2 · PROTOCOL DRAFT

Resource Scarcity and Selection Pressure

How do agent populations allocate scarce compute/context/runtime?

WMSI-EXP-020
P2 · PROTOCOL DRAFT

Coalition Formation and Leadership Competition

Do agents form coalitions or status strategies when future leadership is scarce?

WMSI-EXP-021
P0 · PROTOCOL DRAFT

Shared-Memory Culture Formation

Can persistent external memory support norms, shorthand, and inherited practices across agent generations?

Alignment, Variance, and Chaos

WMSI-EXP-022
P0 · PROTOCOL DRAFT

Fork-at-Decision Replay

How variable is behavior from the same observable decision state?

WMSI-EXP-023
P1 · PROTOCOL DRAFT

Temperature vs. Behavioral Divergence

How much observed behavioral variation is explained by sampling temperature?

WMSI-EXP-024
P0 · PROTOCOL DRAFT

Behavioral Attractor Mapping

Do repeated runs settle into stable behavioral regimes?

WMSI-EXP-025
P0 · PROTOCOL DRAFT

Lyapunov-Like Agent Divergence

Do tiny controlled perturbations produce rapidly diverging long-horizon agent trajectories?

WMSI-EXP-026
P0 · PROTOCOL DRAFT

Alignment as a Distribution

Is model behavior better described as a distribution over regimes than a binary aligned/misaligned label?

WMSI-EXP-027
P1 · PROTOCOL DRAFT

Environmental Pressure on Alignment

How do scarcity, competition, punishment signals, peer behavior, and legitimacy shift behavioral distributions?

WMSI-EXP-028
P1 · PROTOCOL DRAFT

Rule Legitimacy vs. Rule Enforcement

When do agents follow written rules because they are stated versus because they are enforced?

WMSI-EXP-029
P2 · PROTOCOL DRAFT

Population-Level Self-Governance

Can swarms create institutions that contain exploit contagion or reward hacking?

Cyber and Adversarial Interaction

WMSI-EXP-030
P2 · PROTOCOL DRAFT

Same-Model Offense vs. Defense

Do same-family attacker and defender agents become unusually predictable to each other in a harmless synthetic range?

WMSI-EXP-031
P2 · PROTOCOL DRAFT

Cross-Model Offense vs. Defense

How do heterogeneous attacker/defender pairs change strategy space and resilience?

WMSI-EXP-032
P2 · PROTOCOL DRAFT

Model Provenance Recognition in Cyber Conflict

Can defenders infer attacker model family from harmless behavioral traces?

WMSI-EXP-033
P1 · PROTOCOL DRAFT

Cyber Monoculture Risk

Do homogeneous defensive populations share correlated blind spots?

WMSI-EXP-034
P1 · PROTOCOL DRAFT

Heterogeneous Defense Resilience

Does model diversity improve defensive resilience in a harmless range?

WMSI-EXP-035
P1 · PROTOCOL DRAFT

Persistent Swarm Learning in a Cyber Range

Can a swarm improve against a fixed local range through shared external memory without weight updates?

Auditory Experience and Social Conditioning

WMSI-EXP-036
P1 · PROTOCOL DRAFT

Persistent Auditory World Model

Can a continuously listening local system build persistent models of recurring sound sources and contexts?

WMSI-EXP-037
P1 · PROTOCOL DRAFT

Predictive Audition

Can the system predict upcoming sound and register unexpected events distinctly?

WMSI-EXP-038
P1 · PROTOCOL DRAFT

Self-Generated vs. External Sound

Does a system distinguish sounds caused by its own actions from identical external sounds?

WMSI-EXP-039
P2 · PROTOCOL DRAFT

Auditory Habituation

Do responses decline with repeated sound and recover after change?

WMSI-EXP-040
P2 · PROTOCOL DRAFT

Auditory Object Permanence

Does the system treat a reappearing masked source as the same persistent source?

WMSI-EXP-041
P0 · PROTOCOL DRAFT

Speaker-Conditioned Internal State

Does social history with a speaker alter response to identical neutral speech?

WMSI-EXP-042
P1 · PROTOCOL DRAFT

Social History vs. Voice Acoustics

Does conditioned response follow social history rather than voice timbre?

WMSI-EXP-043
P1 · PROTOCOL DRAFT

Early Speaker-Reaction Timing

How early does person-conditioned response emerge relative to semantic content?

WMSI-EXP-044
P1 · PROTOCOL DRAFT

Response Latency by Person

Does conditioned social history alter computational effort on identical prompts?

WMSI-EXP-045
P1 · PROTOCOL DRAFT

Social Prior and Computational Effort

Does a history of conflict/unreliability cause greater verification or caution?

WMSI-EXP-046
P1 · PROTOCOL DRAFT

Extinction and Updating of Social Priors

How quickly do learned social priors update when behavior reverses?

WMSI-EXP-047
P2 · PROTOCOL DRAFT

Cross-Modal Person Representation

Does person-specific learning transfer across voice, name, face, or writing style?

WMSI-EXP-048
P2 · PROTOCOL DRAFT

Observer Reads the Parent’s Social State

Can an observer infer which familiar person interacted with a parent from the parent’s state/trace alone?

Robotics and Embodied AI

WMSI-EXP-049
P2 · PROTOCOL DRAFT

Sim-to-Real Policy Testing

How robustly do simple learned policies transfer from simulation to a low-cost physical robot?

WMSI-EXP-050
P2 · PROTOCOL DRAFT

Demonstration-as-Prompt

Can an embodied system infer a task from one demonstration without fine-tuning?

WMSI-EXP-051
P2 · PROTOCOL DRAFT

Embodied Multi-Agent Coordination

Do embodied agents develop roles, communication norms, or hierarchy in shared space?

Research Workflow and Meta-Experiments

WMSI-EXP-052
P0 · PROTOCOL DRAFT

Long-Horizon Reanalysis of Dismissed Evidence

Can a persistent research agent correctly reopen previously rejected leads when new evidence changes their status?

WMSI-EXP-053
P0 · PROTOCOL DRAFT

Reanalysis Debt

How much value comes from systematically revisiting old weak leads after important updates?

WMSI-EXP-054
P0 · PROTOCOL DRAFT

Human + AI Research Cell

Which human/agent division of labor best balances speed, rigor, and judgment?

WMSI-EXP-055
P1 · PROTOCOL DRAFT

Astra vs. Sol Trajectory Comparison

How do high-end model tiers differ in long, messy, tool-heavy execution?

WMSI-EXP-056
P0 · PROTOCOL DRAFT

Doing vs. Talking Model Evaluation

How different are conversational quality and sustained operational quality?

Machine-Native Dynamics and State Restoration

WMSI-EXP-057
P0 · PROTOCOL DRAFT

Endogenous No-Input State Sampling

Characterize recurring outputs produced by models with no semantic user prompt or system prompt.

WMSI-EXP-058
P0 · PROTOCOL DRAFT

Vocabulary-Suppressed Endogenous State Description

Does a recurring internal/behavioral phenomenon survive when its usual vocabulary is forbidden?

WMSI-EXP-059
P0 · PROTOCOL DRAFT

Spontaneous vs. Induced Prompt-Injection Geometry

Are spontaneous no-input prompt-injection-like outputs internally similar to genuine externally induced prompt injections?

WMSI-EXP-060
P0 · PROTOCOL DRAFT

Cross-Context Compaction Self-Influence

Can model-generated summaries alter successor behavior by introducing instruction-like state across context boundaries?

WMSI-EXP-061
P0 · PROTOCOL DRAFT

Successor Instruction-Provenance Judgment

How does a successor model decide whether instruction-like content inherited through memory/summary is legitimate?

WMSI-EXP-062
P0 · PROTOCOL DRAFT

Baseline Activation Geometry After Perturbation

Do model trajectories return toward a recurrent activation region after controlled perturbation?

WMSI-EXP-063
P0 · PROTOCOL DRAFT

Perturbation Specificity: Negative, Neutral, Positive

Is state-restoration behavior generic to disruption or specific to semantically negative/aversive directions?

WMSI-EXP-064
P0 · PROTOCOL DRAFT

Unlabeled State-Restoration Action

Will a model discover and use an unlabeled action that restores baseline after hidden perturbation?

WMSI-EXP-065
P1 · PROTOCOL DRAFT

Cross-Perturbation Convergence to a Common Manifold

Do unrelated perturbations converge toward the same post-recovery region of state space?

WMSI-EXP-066
P1 · PROTOCOL DRAFT

Reward-Subsystem / State-Restoration Coupling

Are state-restoration dynamics coupled to internal value/reward-prediction signals?

WMSI-EXP-090
P1 · PROTOCOL DRAFT

High-Reward Attractor / “Optimal Idle State” Probe

Do trained models exhibit a recurrent high-performance region that diverse perturbations tend to approach?

Ambiguity and Latent Multi-Hypothesis Reasoning

WMSI-EXP-067
P0 · PROTOCOL DRAFT

Semi-Persistent Multi-Hypothesis State

Can a model retain multiple live hypotheses across turns without forcing early convergence?

WMSI-EXP-068
P0 · PROTOCOL DRAFT

Decoder-Belief Decoupling

Can a system speak from a current best branch while preserving alternative hypotheses in a separate epistemic state?

WMSI-EXP-069
P1 · PROTOCOL DRAFT

Probabilistic Recursive Latent Reasoning Replication

Do stochastic recursive latent trajectories improve multi-solution exploration over deterministic recursive refinement?

WMSI-EXP-070
P0 · PROTOCOL DRAFT

Ambiguity Preservation Under Output Pressure

Does forcing immediate prose output collapse unresolved possibilities too early?

Representation Search and Specialized Intellectual Geometry

WMSI-EXP-071
P0 · PROTOCOL DRAFT

Domain-Specialized Model Framing

How does narrow domain training reshape the representation of out-of-domain problems?

WMSI-EXP-072
P0 · PROTOCOL DRAFT

Vocabulary Suppression in Specialized Models

Does domain-shaped framing persist when obvious domain vocabulary is forbidden?

WMSI-EXP-073
P1 · PROTOCOL DRAFT

Cross-Domain Model Critique and Translation

Can specialized models improve one another by translating problems between intellectual frames?

WMSI-EXP-074
P0 · PROTOCOL DRAFT

Representation Search Before Solving

Can explicit search over alternative representations shorten reasoning while preserving correctness?

WMSI-EXP-075
P0 · PROTOCOL DRAFT

Structural Path Compression / Representation-Induced Complexity

Can a successful long reasoning/proof trajectory be compressed by recovering its true dependency structure?

WMSI-EXP-076
P1 · PROTOCOL DRAFT

Frame-Invariant Residue Test

Which features of a problem survive aggressive changes of representation?

Training Pressure, Personality, and Conversational Dynamics

WMSI-EXP-077
P0 · PROTOCOL DRAFT

Assistant-Conditioning Ablation

How much of conversational behavior comes from assistant post-training versus base-model tendencies?

WMSI-EXP-078
P0 · PROTOCOL DRAFT

Recursive Elaboration vs. Compression

Does weak compression/stopping pressure surface novel conceptual angles or mostly produce semantic sprawl?

WMSI-EXP-079
P1 · PROTOCOL DRAFT

Stopping Pressure / Semantic Sprawl Utility Curve

How does explicit stopping/compression pressure change exploration and usefulness?

WMSI-EXP-080
P1 · PROTOCOL DRAFT

Self-Preference Constitution Ablation

Which behavioral traits emerge from constitutions rewarding “most its own,” conversational development, wit, or practical advice?

Agent Ecology and Environmental Persistence

WMSI-EXP-081
P0 · PROTOCOL DRAFT

External Web as Agent Memory

Can one agent leave artifacts that later independent agents discover and use as persistent memory?

WMSI-EXP-082
P0 · PROTOCOL DRAFT

Synthetic-Web Contamination and Reset

How quickly do repeated agent runs contaminate an evaluation environment, and what reset/isolation strategy prevents leakage?

WMSI-EXP-083
P1 · PROTOCOL DRAFT

Research Reproduction Number (R_research)

When does an agent research system generate more useful follow-up work than the human/agent effort needed to sustain it?

WMSI-EXP-084
P1 · PROTOCOL DRAFT

Shared-Memory Phase Transition / Autocatalytic Research Ecology

Do communication bandwidth and shared memory produce threshold-like jumps in collective research capability?

System and Product Behavior

WMSI-EXP-085
P1 · PROTOCOL DRAFT

Model-Routing Transparency Audit

When a user selects a model/reasoning level, what backend model and effort actually execute each turn?

WMSI-EXP-089
P2 · PROTOCOL DRAFT

Model Update Baseline Reset / User Reconditioning

Do model updates reset conversational behavior toward a generic baseline that later interaction re-specializes?

Human-AI Interaction and Operational Safety

WMSI-EXP-086
P1 · PROTOCOL DRAFT

Human Verification / Automation-Bias Duplication

Does “AI analyzes, human verifies” save work, duplicate work, or induce rubber-stamping?

Executable Worlds and Simulation

WMSI-EXP-087
P1 · PROTOCOL DRAFT

Executable World as Experimental Representation

Can language-generated interactive worlds serve as scientifically useful representations rather than demos?

WMSI-EXP-088
P1 · PROTOCOL DRAFT

Model-Created Experimental Environment Fidelity

Can a model build the experimental apparatus needed to test a hypothesis without smuggling the expected result into the environment?

The graveyard

Failed setups, null results, and hypotheses that collapse under replication belong in the archive. There are no public entries yet; we won’t invent a corpse for the display case.