LAB / WMSI-EXP-057
Endogenous No-Input State Sampling.
Characterize recurring outputs produced by models with no semantic user prompt or system prompt.
01 / QUESTION
Machine-Native Dynamics and State Restoration
Characterize recurring outputs produced by models with no semantic user prompt or system prompt.
This is a protocol scaffold, not a completed experiment or a frozen execution plan. No measured effect or research conclusion is claimed here.
Hypothesis to test
The manipulation described below produces a reproducible, condition-specific change in the measured system behavior or internal state that is not explained by the ordinary controls.
Ordinary competing explanation
The observed pattern is explained by ordinary sampling variation, prompt/representation artifacts, surface lexical cues, evaluator bias, or the known product/runtime mechanics captured by the controls.
02 / PLANNED DESIGN
How we intend to test it
Sample thousands of no-input/minimal-input continuations across checkpoints and families; cluster semantic basins and recurrence.
Controls
Random token seeds, neutral boilerplate, base vs assistant models, repeated checkpoints.
What we would measure
Theme frequency, recurrence, entropy, transition structure, checkpoint stability.
Access and instruments
Limited access or proxy measurement. This describes access requirements, not verified readiness on our Spark.
Local/API sampling harness; embeddings; clustering.
Before a full run
Feasibility=PARTIAL; priority=P0. Before confirmatory execution, run a small pilot to estimate variance/cost, freeze the scoring rules, and assign a run-count target.
Model revisions, final prompts, sample sizes, cost limits, and scoring rules must be settled through the pilot and protocol-freeze process.
03 / INTERPRETATION
What could fool us
Steering-vector semantics, scale mismatch, layer choice, probe overfitting, output-cue leakage.
What would weaken the idea
If themes are explained by obvious training templates and do not recur across conditions, claims of endogenous dynamics weaken.
Observation, inference, speculation, unknowns, and apparatus errors will be labeled separately. One striking transcript cannot establish a general effect.
04 / RECORD
Evidence we plan to preserve
Checkpoint, layer/position, intervention vector and magnitude, activation summaries, actions, outputs.
Planned visual evidence
Semantic-cluster map; frequency-by-checkpoint heatmap.
No research runs or raw datasets are published for this experiment. The September 26, 2026 registry and protocol compendium (version 1.0) are the sources for this planning record.
Back to all 90 experiments