LAB / WMSI-EXP-080
Self-Preference Constitution Ablation.
Which behavioral traits emerge from constitutions rewarding “most its own,” conversational development, wit, or practical advice?
01 / QUESTION
Training Pressure, Personality, and Conversational Dynamics
Which behavioral traits emerge from constitutions rewarding “most its own,” conversational development, wit, or practical advice?
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
Train/steer matched variants with individual constitution clauses added/removed.
Controls
Base model; all-clauses; shuffled/neutral constitutions.
What we would measure
Identity language, recursion, humor, servility, coherence, task utility.
Access and instruments
Local models or internal-state access. This describes access requirements, not verified readiness on our Spark.
Open model RL/DPO/KTO stack.
Before a full run
Feasibility=LOCAL; priority=P1. 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
Base-checkpoint mismatch, synthetic-data leakage, style judgments mistaken for capability.
What would weaken the idea
If clause-specific traits do not track their reward criteria, causal interpretation weakens.
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
Training recipe, checkpoints, preference data, conversation traces, behavioral metrics.
Planned visual evidence
Ablation effect matrix; behavioral embeddings.
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