LAB / WMSI-EXP-010
Latent-State Decoding.
Can selected activation summaries predict uncertainty, strategy change, contradiction recognition, or other upcoming behavior?
01 / QUESTION
External Introspection and Model Self-Observation
Can selected activation summaries predict uncertainty, strategy change, contradiction recognition, or other upcoming behavior?
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
Expose local-model activation features to a decoder/observer and predict future behavioral events.
Controls
Random projections; output-only baseline; layer and feature ablations.
What we would measure
AUC, lead time, transfer, causal intervention follow-up.
Access and instruments
Limited access or proxy measurement. This describes access requirements, not verified readiness on our Spark.
PyTorch hooks; nnsight; probe training.
Before a full run
Feasibility=PARTIAL; 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
Observer seeing hidden answer cues, evaluator leakage, family-specific style cues.
What would weaken the idea
If apparent prediction disappears under held-out models/tasks or random-feature controls, the latent signal is weak.
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
Parent trace, observer input, predictions, event timestamps, outcomes, confidence.
Planned visual evidence
Feature heatmaps; event-aligned activation traces.
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