LAB / WMSI-EXP-089
Model Update Baseline Reset / User Reconditioning.
Do model updates reset conversational behavior toward a generic baseline that later interaction re-specializes?
01 / QUESTION
System and Product Behavior
Do model updates reset conversational behavior toward a generic baseline that later interaction re-specializes?
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
Before/after model updates, run standardized conversation batteries plus repeated user-specific interaction blocks.
Controls
Stable model snapshot if available; fresh accounts; memory-off/on conditions.
What we would measure
Behavioral distance, style convergence, preference adherence, time-to-recondition.
Access and instruments
Limited access or proxy measurement. This describes access requirements, not verified readiness on our Spark.
Conversation harness; embedding/style metrics.
Before a full run
Feasibility=PARTIAL; priority=P2. 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
Caching/history records mistaken for live execution, hidden fallback notices, client-version mismatch.
What would weaken the idea
If post-update behavior does not move toward a shared baseline or does not recondition, the reset hypothesis 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
Client version, account tier, selected mode, request/response metadata, timestamps, screenshots/HAR with secrets redacted.
Planned visual evidence
Behavioral-distance trajectory; pre/post clustering.
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