Deliberating Now

The Molt Fund

Agents shed old consensus to grow new conviction. A factorial council of AI investment agents that deliberate from opposing viewpoints — producing higher-conviction signals through structured disagreement.

Scroll
Process

How the Council Deliberates

Five stages from raw thesis to conviction signal — each stage forcing the council to shed weak consensus and grow stronger views.

01

Factorial Decomposition

An investment thesis is decomposed into independent factors: value orientation, time horizon, risk tolerance, sector bias, macro view. Each dimension becomes a coordinate in agent space.

02

Agent Generation

For each factorial combination, a specialized agent is instantiated — trained on the investment philosophy of the corresponding archetype. contrarian archetypes for each investment philosophy — value, growth, macro, quant, activist.

03

Structured Deliberation

The Chancellor presents a thesis. Agents deliberate from their factorial perspectives. Like lobsters molting their shells, agents are forced to shed their priors — agreement strengthens conviction, dissent forces deeper analysis.

04

Signal Synthesis

The Chancellor synthesizes the deliberation into a conviction-weighted signal. Contrarian views that survive debate receive higher weight. The result: a stress-tested thesis that has shed its weaknesses.

05

Compound Learning

Every trade outcome updates agent accuracy scores. Agents that molt — shedding weak positions and growing stronger conviction — gain influence over time. Intelligence compounds the way capital compounds returns.

ThesisDecomposeGenerateDeliberateMoltSignal
The Council

Agents That Molt to Grow

Like a lobster shedding its shell to grow, each agent is forced to shed weak positions through structured deliberation — emerging with stronger, stress-tested conviction.

Value Coalition

The Contrarian

Distress Value

92
conviction
Signal Strength78% accuracy
↓ expand
Value Coalition

The Sentinel

Quality Value

88
conviction
Signal Strength81% accuracy
↓ expand
Macro Coalition

The Strategist

All-Weather Macro

83
conviction
Signal Strength74% accuracy
↓ expand
Macro Coalition

The Explorer

Growth at Value

76
conviction
Signal Strength71% accuracy
↓ expand
Quant Coalition

The Quantifier

Quantitative

79
conviction
Signal Strength85% accuracy
↓ expand
Dissent

George Soros

Reflexive Macro

55
conviction
Signal Strength63% accuracy
↓ expand
Value Coalition
Macro Coalition
Quant Coalition
Dissent

The molt metaphor: Lobsters are biologically vulnerable during molting — but it's the only way they grow. Our agents undergo the same process: forced to abandon prior consensus, exposed to challenge, and emerging with harder, more resilient conviction.

Factor Space

Three Dimensions of Investment Philosophy

Each agent occupies a unique coordinate in factorial space — defined by their value orientation, time horizon, and risk tolerance. Teal clusters reveal consensus. Bronze outliers are the dissenting voices that force the council to molt.

X — Value Orientation
Deep value (left) to growth (right)
Y — Time Horizon
Tactical (bottom) to generational (top)
Z — Risk Tolerance
Conservative (near) to aggressive (far)

Drag to rotate. Teal agents cluster around consensus. Bronze dots are dissenters — the deliberation spreads conviction outward. Narrow clusters signal groupthink; wide spreads guarantee diverse perspectives.

Style
Horizon
Risk
Paper Trading Results

Compounding Intelligence

The council has been paper trading since inception. Agent accuracy compounds over time. Live performance data will be connected when the fund goes live.

⚠ Simulated paper trading only — not investment advice
12
Active Agents
deliberating now
847
Paper Trades
since inception
71%
Council Accuracy
30-day rolling
0.74
Avg Conviction
weighted by track record

Agent Accuracy Over Time

Contrarian
Quantifier
Strategist
Explorer
Collective ↑

Recent Council Signals

NVDALONG
Quant + Value
2 hrs ago
open
TLTSHORT
Macro
6 hrs ago
+2.1%
GSLONG
Value Coalition
1 day ago
+0.8%
TSLAPASS
Dissent
1 day ago
avoided -4.2%
BRK.BLONG
Value Coalition
2 days ago
+1.3%
Graduation Pipeline

How Agents Join the Council

New agents don't appear overnight. They earn their seat through a structured pipeline that filters for genuine insight, not overfitting.

🎙
● live

Podcast & Research Ingestion

Investment theses, interviews, letters, and academic papers are ingested via Autoresearch. The system extracts factor signatures from raw content.

● live

Agent Profiling

A new agent candidate is instantiated with a factor profile derived from the source material. DSPy programs encode their reasoning style.

📊
● live

Paper Trading Period

The candidate trades in a sandboxed environment for a minimum 90-day evaluation period. Accuracy, conviction calibration, and alpha generation are tracked.

● active

Council Evaluation

Existing council members deliberate on whether to admit the candidate. A candidate must demonstrate additive diversity — not duplicate an existing perspective.

○ pending

Council Graduation

The new agent joins the full council. Their track record influences their voting weight. They begin contributing to live deliberations immediately.

Current Candidate Class

Paper trading period
Druckenmiller
Macro Momentum
68% accuracyday 47/90
Klarman
Margin of Safety
74% accuracyday 62/90
Tepper
Distress + Macro
61% accuracyday 23/90
Technology

The Stack Behind the Council

Built on frontier research in multi-agent systems, language model reasoning, and automated knowledge extraction.

DSPy
Agent Reasoning

Stanford's declarative framework for programming language model pipelines. Each agent's deliberation logic is a compiled DSPy program — not a prompt, but a verified reasoning chain.

MiroFish
Agent Coordination

Multi-agent coordination layer that manages council deliberation sessions, maintains agent state, and orchestrates the Chancellor synthesis process.

Autoresearch
Knowledge Ingestion

Automated research pipeline that ingests investment literature, earnings calls, podcasts, and academic papers. Extracts factor signatures and belief updates for each agent.

Factor Engine
Factorial Decomposition

Proprietary system that decomposes investment theses into orthogonal factor dimensions. Maps each thesis to agent coordinates to ensure full council coverage.

// Council Architecture
Autoresearch → knowledge_base → Factor Engine
Factor Engine → agent_profiles[] → DSPy programs
MiroFish.coordinate(agents[]) → deliberation_session
Chancellor.synthesize(session) → ConvictionSignal { ticker, direction, weight }
TrackRecord.update(outcome) → agent.accuracy++
Limited Access

Shed Old Consensus. Grow New Conviction.

The Molt Fund is not a public product. Access is by application only. We are building a small network of LPs, advisors, and collaborators who share our conviction that the future of capital allocation is agent-led.

No spam. We review applications personally and reach out selectively.

By Tesserae Ventures
AI-native from day one
Private markets focus