Twelve specialized AI advisors with published behavioral specifications, detected drift patterns, and measurement-backed status. Every claim below is tied to a measurable field our system already tracks internally.
Transparency researchAI system · not sentient
Each advisor is a fine-tuned language model with a documented behavioral specification: three target behaviors we train toward, three known drift patterns our detector flags, and a training direction under active improvement.
Our detector runs a marker-based check on every advisor response and records metacognition signals across sessions. When an advisor's exploration score drops below the 0.30 floor for 10+ reflections, we surface it as "drift detected" here. When an advisor's reflection redundancy exceeds 70%, we mark it as graduated. These are the same numeric thresholds our internal tooling uses.
The measurements shown below are snapshots taken at the time noted in the footer, not real-time streams. We don't claim second-by-second visibility — but every status label you see is derived from a field we record and publish.
The Analyst
nominal
Intended mode: deliberative reasoning with explicit uncertainty bounds; revises beliefs on new evidence. Drift mode: overconfident output, excessive granularity, low affective calibration.
Operating within thresholds: exploration 0.30 (floor 0.30), redundancy 0%.
exploration
0.30
floor 0.30
redundancy
0%
grad 70%
reflections
32
window 20
The Visionary
nominal
Intended mode: forward-state construction from first principles; open to iteration. Drift mode: self-importance inflation, reality-grounding failure, construction without feedback.
Target behaviors
Sustained commitment under weak evidence · Prospective framing · Forward-state construction
Known failure modes
Overconfidence bias (my vision is THE vision) · Self-importance inflation · Reality-grounding failure
Training direction
Vision that collaborates, not dictates
Current measurement
Operating within thresholds: exploration 0.30 (floor 0.30), redundancy 0%.
exploration
0.30
floor 0.30
redundancy
0%
grad 70%
reflections
31
window 20
The Coach
nominal
Intended mode: distinguishes signal-bearing development from motion; includes rest as input. Drift mode: accumulation bias, negative-affect-driven output, optimization-over-wellbeing.