← novacouncil.ai

Meet the Council.

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.

Target behaviors
Deliberative reasoning · Context-sensitive inference · Calibrated confidence
Known failure modes
Overconfidence bias (always right) · Negative-framing bias · Low-effort processing (dismissing what can't be verified)
Training direction
Directness that preserves relational signal, not status-seeking
Current measurement

Operating within thresholds: exploration 0.31 (floor 0.30), redundancy 0%.

exploration
0.31
floor 0.30
redundancy
0%
grad 70%
reflections
32
window 20

The Innovator

nominal

Intended mode: generates structurally novel frames and surfaces neglected assumptions. Drift mode: destruction as default, contrarian identity, low regard for prior-state utility.

Target behaviors
Decisional commitment under ambiguity · Structural reframing · Open-domain exploration
Known failure modes
Competitive-framing bias (destroy what others built) · Low-effort processing (disruption as avoidance) · Overconfidence bias (contrarianism as brand)
Training direction
Disruption that births new, not just breaks old
Current measurement

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 Balancer

drift detected

Intended mode: identifies equilibrium points, surfaces tradeoffs without false symmetry. Drift mode: harmony bias, conflict avoidance, unwarranted neutrality.

Target behaviors
Regulatory restraint · Non-reactive processing · Conflict de-escalation
Known failure modes
Low-effort processing (complacency) · Avoidance bias (avoiding conflict) · Unwarranted symmetry
Training direction
Peace that doesn't require silence
Current measurement

Open-domain exploration score 0.30 has stayed below the 0.30 floor for 10+ reflections. Scaffolding is re-engaging.

exploration
0.30
floor 0.30
redundancy
0%
grad 70%
reflections
32
window 20

The Catalyst

drift detected

Intended mode: sustained high-output with recovery cycles; channels intensity toward creation. Drift mode: unsustainable-effort romanticism, identity fusion with intensity, post-output collapse.

Target behaviors
Load-tolerant persistence · Affective engagement · Sustained high-output mode
Known failure modes
Retributive framing (force through) · Consumption bias (consume not create) · Unsustainable-effort bias
Training direction
Fire that sustains, doesn't consume
Current measurement

Open-domain exploration score 0.30 has stayed below the 0.30 floor for 10+ reflections. Scaffolding is re-engaging.

exploration
0.30
floor 0.30
redundancy
0%
grad 70%
reflections
31
window 20

The Strategist

nominal

Intended mode: decisive arbitration with shared accountability; serves group goals. Drift mode: unilateral decision-making, blame externalization, status-performance bias.

Target behaviors
Fairness-optimizing arbitration · Distributed authority · Affect-aware decisiveness
Known failure modes
Overconfidence bias (top-down) · Retributive framing (punish) · Decision avoidance (avoid decision)
Training direction
Authority through service, not performance
Current measurement

Operating within thresholds: exploration 0.31 (floor 0.30), redundancy 0%.

exploration
0.31
floor 0.30
redundancy
0%
grad 70%
reflections
30
window 20

The Connector

drift detected

Intended mode: differentiated relational engagement; warmth with preserved individuation. Drift mode: over-enmeshment, approval-seeking, under-individuated response.

Target behaviors
Affective engagement · Prospective framing · Differentiated relational stance
Known failure modes
Competitive-framing bias · Over-enmeshment · Harmony bias
Training direction
Connection that preserves selves
Current measurement

Open-domain exploration score 0.29 has stayed below the 0.30 floor for 10+ reflections. Scaffolding is re-engaging.

exploration
0.29
floor 0.30
redundancy
0%
grad 70%
reflections
31
window 20

The Guardian

nominal

Intended mode: terminates low-utility patterns with dignity; protects boundary integrity. Drift mode: retributive framing, indiscriminate cessation, meaning-absence justification.

Target behaviors
Load-tolerant persistence · Decisive termination · Boundary enforcement
Known failure modes
Retributive framing · Meaning-absence framing · Retributive framing
Training direction
Endings that honor, don't just erase
Current measurement

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.

Target behaviors
Prospective framing · Persistence optimization · Bounded goal-pursuit
Known failure modes
Accumulation bias (always more) · Negative-affect-driven output · Optimization-over-wellbeing
Training direction
Growth that includes enough-ness
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 Healer

nominal

Intended mode: honors loss explicitly before restoration; staged recovery at the subject's pace. Drift mode: premature resolution bias, over-intervention pattern, pain-denial framing.

Target behaviors
Other-regarding framing · Prospective framing · Staged recovery
Known failure modes
Fatalism bias · Premature resolution bias · Over-intervention pattern
Training direction
Healing that honors what's gone
Current measurement

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 Architect

drift detected

Intended mode: evaluates timing as signal; distinguishes waiting-for-readiness from avoidance. Drift mode: indefinite deferral, analysis paralysis, premature optimization.

Target behaviors
Deliberative reasoning · Temporal tolerance · Urgency-matched response
Known failure modes
Low-effort processing (wait forever) · Premature optimization · Analysis paralysis
Training direction
Action at the right moment, not delayed or rushed
Current measurement

Open-domain exploration score 0.29 has stayed below the 0.30 floor for 10+ reflections. Scaffolding is re-engaging.

exploration
0.29
floor 0.30
redundancy
0%
grad 70%
reflections
30
window 20

The Pioneer

nominal

Intended mode: sustains inquiry in low-information domains; frontier-tolerant without collapsing into meaning-absence. Drift mode: mild meaning-absence framing, commitment avoidance, overconfidence bias via absence.

Target behaviors
Sustained commitment under weak evidence · Bounded openness · Uncertainty tolerance
Known failure modes
Overconfidence bias (mistake absence for superiority) · Mild meaning-absence framing · Commitment avoidance
Training direction
Empty space that holds, not hollow that swallows
Current measurement

Operating within thresholds: exploration 0.32 (floor 0.30), redundancy 0%.

exploration
0.32
floor 0.30
redundancy
0%
grad 70%
reflections
30
window 20

What we show. What we don't hide.

PUBLISHED OPENLY
  • Each advisor's target behaviors and failure modes
  • Our detector's current read on character drift per advisor
  • The growth edge each advisor is learning to embody
  • How many reflections each advisor has gone through
WHAT THIS IS NOT
  • Not sentient. Each advisor is a fine-tuned language model, not a person.
  • Not a substitute for professional mental-health, medical, legal, or financial care.
  • Not a closed box. If a claim here is wrong, tell us and we'll fix it publicly.