conversational inference of equity valuation and allocation

[hyunjimoon]

1. conversational inference

2. valuation and allocation

• what is utility being maximized for founder and investor and how are their constraints be gradually revealed (investors have more clearer constraints - and how this acts as advantage)?

3. world model of term sheet and cap table

4. competitor analysis CVIA table

5. evaluate with existing knowledge on youtube/yc website

---

1. Purpose

To assist entrepreneurs and investors in achieving optimal outcomes during negotiations. AI-powered Conversational Valuation Advisor that uses online Bayesian inference and natural language processing to assist entrepreneurs and investors during term sheet negotiations. The advisor provides real-time, context-aware advice on complex topics such as option pool sizing, board composition, and dividend clauses. It's designed to help users understand the implications of various terms, compare proposed terms to market standards, and develop negotiation strategies. The advisor combines comprehensive analysis of term sheets and cap tables with scenario modeling to provide actionable insights tailored to the specific situation of each startup.

2. Key Features:

1. Comprehensive term sheet and cap table analysis:

• Interprets complex clauses (e.g., founder vesting, option pools, board composition)
• Calculates dilution effects and ownership percentages
• Assesses implications of different terms on future funding rounds

2. Goal-oriented analysis using cooperative language-guided inverse planning
3. Dynamic assistance adapting to new information and changing negotiation landscapes
4. Transparent reasoning in natural language, explaining implications of each term
5. Scenario modeling for long-term impact assessment, considering industry-specific factors (e.g., clockspeed, deep tech considerations)

3. Use Cases for the Conversational Valuation Advisor

Scenario	Issue	How Advisor Helps	Key Advice
Option Pool Sizing	Determining appropriate size for employee option pool	- Analyzes hiring plans and industry standards - Calculates dilution effects - Suggests optimal pool size - Explains valuation implications	Recommend 15% pool based on hiring needs and industry norms; explains effective valuation impact
Board Composition	Negotiating board structure with potential investor	- Explains typical structures - Analyzes control implications - Suggests negotiation strategies - Highlights future implications	Advise pushing for 3-member board (1-1-1) initially for more founder control, with option to expand later
Cumulative Dividend Clause	Assessing long-term implications of dividend terms	- Calculates financial impact over time - Compares to market standards - Models exit scenarios - Suggests negotiation tactics	Highlight above-market rate, quantify impact on exit proceeds, recommend negotiating down to 3-6% or non-cumulative

Definitions:

• Modules:

  • Imputation: Estimating missing values in a dataset
  • Prediction: Forecasting future or unknown values based on existing data
  • Synthetic Data Generation: Creating artificial data that mimics real data characteristics
  • Anomaly Detection: Identifying unusual patterns or outliers in data

• Features: Desirable capabilities of the advisor's role in assisting with negotiations

• Terms:

  • Option Pool: Shares reserved for future employee compensation
  • Vesting: Process by which employees gain ownership of options over time
  • Board of Directors: Governing body overseeing company decisions
  • Dividends: Distribution of company profits to shareholders

Feature	Modules	Use Case for Term Sheet Negotiation
Dynamic Valuation Modeling	Prediction, Synthetic Data Generation	Option Pool: Model different pool sizes and their impact on founder dilution and company valuation. Generate synthetic scenarios to illustrate tradeoffs.
Multi-scenario Outcome Projection	Prediction, Synthetic Data Generation	Board of Directors: Project various board compositions and their potential impact on future funding rounds and exit scenarios.
VC Intention Inference	Prediction, Anomaly Detection	Dividends: Infer VC's risk tolerance and exit timeline based on their dividend preferences. Detect unusual dividend requests that may indicate hidden concerns.
Negotiation Strategy Recommendation	Prediction, Anomaly Detection	Vesting: Recommend optimal vesting schedules based on industry norms and detected anomalies in VC's proposed terms.
Jargon Translation and Education	All modules	Option Pool + Board of Directors: Explain complex terms, generate examples, and detect misunderstandings to ensure founders grasp the full implications of proposed structures.
Collaborative Decision Support	All modules	Vesting + Dividends: Facilitate interactive scenario planning, allowing both parties to see immediate impacts of term adjustments and collaboratively find optimal solutions.

TC

Term	Definition	Structural Knowledge (Cap Table & Term Sheet)	Industry Effect (Clockspeed & Deep Tech)
Founder Vesting	The process by which founders' shares become available over time, typically tied to their continued involvement with the company.	Cap Table: Reflected in the "Common Stock (Founders +)" section, usually with a vesting schedule. Term Sheet: Detailed in the "Founders' Stock" section, specifying vesting schedule (e.g., 25% after one year, then monthly over 36 months).	Fast Clockspeed: Shorter vesting periods may be preferred to align with rapid pace of change and incentivize quick execution. Deep Tech: May require longer vesting periods to account for extended development timelines and ensure founder commitment.
Option Pool	A portion of a company's shares reserved for future issuance to employees, directors, and consultants as incentives.	Cap Table: Shown as "Outstanding Option Pool" and "Available Option Pool". Often increased during funding rounds. Term Sheet: Specified in the "Employee Stock Options" section, typically aiming for a certain percentage (e.g., 15%) of fully diluted capital stock post-investment.	Fast Clockspeed: May require larger option pools to attract and retain talent in a competitive market. Deep Tech: May have smaller option pools due to lower employee turnover and longer development cycles.
Board of Directors	A group of individuals elected to represent shareholders and oversee the management of a company.	Cap Table: Not directly reflected, but influences decisions that affect the cap table. Term Sheet: Detailed in the "Board of Directors" section of the Voting Agreement, specifying composition (e.g., Founder representative, Investor representative, and independent member).	Fast Clockspeed: Boards may need to be more agile and meet more frequently to adapt to rapid pace of change. Deep Tech: May benefit from having board members with specialized technical expertise to guide long-term technology strategy.
Mandatory Conversion Clause	A provision that requires the automatic conversion of preferred stock to common stock upon certain triggering events, such as an IPO.	Cap Table: Not directly reflected, but affects potential future changes in cap table structure. Term Sheet: Specified in the "Charter" section, detailing conditions for mandatory conversion (e.g., QPO with 5x Original Purchase Price and $50M+ gross proceeds).	Fast Clockspeed: May have lower thresholds for mandatory conversion to align with expectation of faster exits and liquidity events. Deep Tech: May have higher thresholds for mandatory conversion due to longer timelines required to achieve significant valuation milestones.
SAFE (Simple Agreement for Future Equity)	A financing instrument that allows investors to purchase rights to future equity in a company.	Cap Table: Reflected in "Series Seed-1 Preferred Stock" and "Series Seed-2 Preferred Stock" sections, showing converted SAFE investments. Term Sheet: Not directly mentioned in the provided term sheet, but the cap table shows two SAFE conversions with different valuation caps.	Fast Clockspeed: SAFEs can provide quick access to capital without immediately diluting founders, which can be advantageous in rapidly evolving markets. Deep Tech: SAFEs might be used for early-stage funding, but the longer development cycles may necessitate more structured financing in later stages.

Term	Definition	Structural Knowledge (Cap Table & Term Sheet)
Option Pool	A portion of a company's shares reserved for future issuance to employees, directors, and consultants as incentives.	Cap Table: Directly impacts founder ownership percentage. Increasing pool size decreases founders' stake. Term Sheet: Size negotiated as part of pre-money valuation. Larger pools effectively lower valuation for founders. VCs may push for larger pools to ensure adequate employee incentives without further dilution. Implications: Crucial for talent attraction but directly affects founder dilution. Optimal sizing depends on hiring plans and industry standards.
Board of Directors	A group of individuals elected to represent shareholders and oversee the management of a company.	Cap Table: Not directly reflected, but board decisions impact future rounds, option grants, and exit events, all of which affect the cap table. Term Sheet: Composition heavily negotiated. Typical structures include 1-1-1 or 2-2-1 (founder-investor-independent). Control implications are significant. Implications: Affects company control and decision-making. Founder-heavy boards retain more control but may limit investor interest. Investor-heavy boards can provide expertise but may prioritize shorter-term gains.
Cumulative Dividends	A dividend structure where unpaid dividends accumulate and must be paid before any distribution to common stockholders.	Cap Table: Not directly shown, but impacts potential payouts in exit scenarios. Term Sheet: Often expressed as a percentage of the original investment amount. Can significantly impact returns, especially in modest exits. Implications: Acts as downside protection for investors. Can substantially reduce common stockholder proceeds in exit scenarios. More common in later stages or down markets.
Anti-dilution Provisions	Protections for investors against future down rounds or dilutive events.	Cap Table: Not directly shown, but impacts share price calculations in future rounds. Term Sheet: Typically broad-based weighted average, but full ratchet provisions exist in some markets. Implications: Provides downside protection for investors but can significantly dilute founders and employees in down rounds. Negotiating limitations or thresholds is crucial.
Liquidation Preference	The order and amount in which investors receive proceeds from a liquidity event.	Cap Table: Not directly shown, but crucial for calculating potential payouts in exit scenarios. Term Sheet: Expressed as a multiple of the original investment (e.g., 1x, 2x) and can be non-participating or participating. Implications: Significantly impacts return distribution in exits. Higher preferences protect investors in modest exits but can misalign incentives for larger exits.

SCN

Term	Definition	Structural Knowledge	Industry Effect
Discount	A percentage reduction in the price per share that SAFE or Note holders pay compared to new investors in a priced round.	- Applied at conversion during a priced round - Compensates early investors for higher risk - Typically ranges from 10-30%	- More common in highly competitive or fast-growing industries - May be larger in riskier or unproven markets
Valuation Cap	A maximum company valuation used to determine the conversion price for SAFE or Note holders.	- Sets a ceiling on the effective price per share for early investors - Protects against excessive dilution in high-growth scenarios - Can significantly impact cap table if there's a large gap between cap and actual valuation	- Lower caps more common in early-stage or high-risk industries - Higher caps in more established or lower-risk sectors
Interest Rate (Notes only)	The rate at which the principal of a Convertible Note increases over time.	- Applies only to Convertible Notes, not SAFEs - Accrues until conversion or repayment - Typically converts to equity along with principal	- Higher rates in riskier industries or for less established companies - Lower or no interest more common in fast-growing tech sectors
Maturity Date (Notes only)	The date by which a Convertible Note must be repaid or converted if no qualifying financing event has occurred.	- Applies only to Convertible Notes, not SAFEs - Often extended if company hasn't raised a priced round - May trigger conversion to equity or repayment	- Shorter in fast-moving industries - Longer in industries with longer development cycles (e.g., biotech)
Pro Rata Rights	The right of an investor to participate in future funding rounds to maintain their ownership percentage.	- Often included in SAFEs and Notes - Allows early investors to avoid dilution in later rounds - Can be a separate "side letter" in some cases	- More valuable in industries with high growth potential - Less emphasized in slower-growth sectors
MFN (Most Favored Nation)	A clause ensuring that if better terms are offered to future investors, those terms will also apply to the current investor.	- More common in SAFEs than Notes - Protects early investors if company later offers better terms - Can complicate future fundraising	- More prevalent in competitive startup ecosystems - Less common in more established or regulated industries
Conversion Trigger	The event that causes the SAFE or Note to convert into equity, typically a priced equity round of a certain size.	- Defines when the instrument converts to equity - Usually set at a specific investment amount or company valuation - Can include acquisition as a trigger	- Lower thresholds in early-stage or capital-efficient industries - Higher thresholds in capital-intensive sectors

[hyunjimoon]

to get quick input from Vikash who has higher precision knowledge on current state of prob.comp tech, i described two products (TC, SCN), Based on the description of the [conversational valuation advisor] and the CLIPS technology, cld created a comparison table to evaluate which product (TC or SCN) would be more suitable to start developing. This analysis assumes you're an MIT PhD student aiming to quickly develop an impactful product for the entrepreneurship market using Bayesian technology.

Criteria	TC (Term Sheet, Cap Table)	SCN (SAFE, Convertible Notes)
Technical Feasibility	Higher: - More structured data - Clear mathematical relationships - Easier to model using Bayesian inference - Aligns well with CLIPS' goal-oriented analysis	Lower: - More variability in structures - Complex interdependencies - Requires more sophisticated natural language processing
Operational Feasibility	Higher: - More readily available data for training - Easier to validate outputs - Aligns with existing tools and processes	Moderate: - Less standardized data - More challenging to integrate with existing workflows - May require more expert knowledge to validate
Desirability	Very High: - Addresses a critical pain point for entrepreneurs - Highly relevant for all startup funding rounds - Immediate impact on negotiation outcomes	High: - Relevant for early-stage startups - Growing in popularity - Less immediate impact on larger funding rounds

Based on this analysis, I would recommend starting with the TC (Term Sheet, Cap Table) product for the following reasons:

1. Technical Feasibility: The TC product aligns more closely with the strengths of the CLIPS technology. Term sheets and cap tables have more structured data and clearer mathematical relationships, making them easier to model using Bayesian inference. This allows for a more straightforward implementation of the goal-oriented analysis and scenario modeling features.

2. Operational Feasibility: There's likely more readily available data for training the model on term sheets and cap tables. This data is also easier to validate, which is crucial for building trust in the advisor's recommendations. The TC product would also integrate more seamlessly with existing tools and processes used in the startup ecosystem.

3. Desirability: While both products are desirable, the TC product addresses a more universal pain point for entrepreneurs across all stages of startup funding. It has the potential for immediate and significant impact on negotiation outcomes, which could make it more attractive to a wider range of users.

4. Quick Impact: As an MIT PhD student looking to make a fast and impactful entry into the entrepreneurship market, the TC product offers a clearer path to demonstrating the value of your Bayesian technology. The more structured nature of term sheets and cap tables allows for more precise and easily verifiable outputs, which could lead to quicker adoption and recognition in the market.

5. Scalability: Starting with the TC product doesn't preclude expanding to SCN later. In fact, the knowledge and user base gained from the TC product could provide valuable insights and a strong foundation for developing the SCN product in the future.

In conclusion, while both products have merit, the TC (Term Sheet, Cap Table) product appears to be the more strategic choice for initial development, offering a better balance of technical feasibility, operational practicality, and market desirability.

[hyunjimoon]

I'm asking for help on below to Shari, Charlie (AI and entrepreneurship panels or SAC members), hyukjin, joonsung

Your expertise would be tremendously helpful in shaping this project, particularly in two areas:

• Training data: I'm seeking guidance on where to obtain additional training data, such as redacted term sheets from various industries and stages. Your insights on potential sources would be invaluable.

• Knowledge verification: I'd greatly appreciate your quick verification of the knowledge I've extracted so far, to ensure accuracy and comprehensiveness in the AI's understanding of term sheets and cap tables.

[hyunjimoon]

analysis of seven https://claude.ai/chat/c2ff6bf4-bbb1-4044-aad9-d2e995447076

These metrics allow us to quantify the variability in goals and outcomes across different negotiation points, even when the terms are not directly comparable numerically.

For "Our Goal" column:
We'll use a "Goal Divergence Score" on a scale of 0-1, where:
0 = No change from the initial term
0.5 = Moderate change or compromise sought
1 = Significant change or complete removal sought

For "What We Got" column:
We'll use an "Outcome Variability Score" on a scale of 0-1, where:
0 = No change from the initial term
0.5 = Partial change or compromise achieved
1 = Significant change or complete removal achieved

Term	Goal Divergence Score	Outcome Variability Score
Dividends	0.92	0.83
No shop	0.83	0.75
Founder vesting	0.75	0.67
Option pool	0.67	0.58
Board of directors	0.58	0.50
Lawyer's fees	0.75	0.42
Founders' reps and technology ownership	0.83	0.33

Explanation of scores:

1. Dividends: High variability in both goals and outcomes, with most teams seeking to remove or significantly reduce dividends.

2. No shop: Consistently high variability, with most teams aiming to reduce the period and achieving some level of reduction.

3. Founder vesting: Significant variability in goals and outcomes, with teams generally seeking more favorable terms.

4. Option pool: Moderate variability, with most teams aiming for some reduction and achieving partial success.

5. Board of directors: Moderate variability, with teams often seeking slight changes in composition.

6. Lawyer's fees: High variability in goals (most teams sought caps), but lower variability in outcomes (caps often higher than desired).

7. Founders' reps and technology ownership: High variability in goals (most teams sought removal), but low variability in outcomes (often not achieved or compromised).

and extracted seven question that twelve team would have

"How does the proposed option pool size compare to industry standards for our stage and sector?"
"What are the potential implications of the suggested board composition on future decision-making and control?"
"How does the offered vesting schedule compare to market norms, and what are the pros and cons of accepting or negotiating it?"
"What are the most effective strategies for negotiating these terms without damaging the relationship with the VC?"
"How might these terms impact our ability to attract and retain key talent in the future?"
"What are the long-term implications of these terms on future funding rounds and potential exits?"
"How can we balance the VC's risk mitigation needs with our desire for founder-friendly terms?"

[hyunjimoon]

pivoting with track record of theory

synthesizing concept of angie's three pivot levels #158 (reply in thread) and meehl's theory track record #159 (comment)
extracting value from disagreement and identified infeasibility cld

Meehl proposes two rough indexes to quantify a theory's track record:

1. How intolerant or risky the theory's predictions are - i.e. how specific and unlikely they are a priori.
2. How close or accurate the predictions turn out to be compared to observed data.

seems not irrelevant to contrarian dynamics: need to win big in less probable things

These are combined to assess how well the theory is performing overall in terms of making risky yet accurate predictions.

Component	Description	Update Type	Example in Term Sheet - VC Fit Context
Auxiliary Theory (a)	Implementation details	Feasibility	Negotiation tactics, presentation style
Substantive Theory (T, t)	Core business model or value proposition	Feasibility	Company valuation method, growth projections
Desirability (D, d)	Alignment with goals and preferences	Desirability	Investor preferences, funding amount, equity stake
Feasibility Update	Changes in ability to execute	F1, F2, F3	New financial data, milestone achievement
Desirability Update	Changes in goals or external preferences	D1, D2, D3, D4, D5, D6	Shift in investor risk appetite, market trends
Theory Track Record	Accumulation of successful risky predictions	Corroboration	History of accurate forecasts, "damn strange coincidences"
Prediction Intolerance	Specificity and a priori unlikelihood of predictions	Riskiness	Narrow valuation range, precise growth targets
Prediction Accuracy	Closeness of predictions to observed outcomes	Corroboration	Hitting forecasted metrics, near-miss on projections

Lakatosian defense strategy states that defending and amending an apparently falsified theory T is appropriate only when T has built up a good track record ("money in the bank") by making successful or near-miss predictions that had a low prior probability of being correct absent the theory. These unexpected correct predictions are called "damn strange coincidences" by Salmon. This gives the theory credibility to warrant continued defense and refinement even in the face of some contradictory evidence.

---

Optimizing Term Sheet Quality Through Rational Meaning Construction | Information and Control in Ventures paper, meehl appraise, amend, theory cld

[hyunjimoon]

below is the first prompt i used to choose three terms (option pool, board, vesting) to make the most use of word to world models. i'm documenting the old prompt before i update this, as vikash gave me feedback on scoping much smaller and compare with three competitors (crunchbase based stats, mc model based on cap table structure, conversational app without any language model with prob.model).

You are Vikash Mansinghka, an expert in probabilistic programming. Your task is to design a Conversational Valuation Consultant powered by probabilistic programming to assist in term sheet negotiations between entrepreneurs and investors. This AI-powered assistant will use advanced symbolic reasoning and probabilistic inference to provide insights into complex valuation scenarios.
The product will use a probabilistic program language to express generative models as computer programs that combine deterministic code with random choices. It will assemble three modules (Generation, Condition/Bayesian Inference, and Anomaly Detection) to develop three features (dynamic valuation modeling, multi-scenario outcome projection, and intention inference) to answer queries on four key terms: option pool, board of directors, dividends, and vesting.
The technology underlying this product is based on the probabilistic language of thought, as explained by Josh Tenenbaum in his discussion of the Bayesian tug of war example. This approach involves:

1. Defining a world model (objects and relations)
2. Conditioning on new observations
3. Answering questions based on the updated model
   In the tug of war example, the world model defines object (player), objects's character (strength), relation (won-against). Observations condition the model on new information (e.g., "Jack won against Leo"), and questions query the updated model (e.g., "How strong is Jack?"). Apply this approach to the valuation advisor, using negotiation outcomes as observations. The world model should represent the startup ecosystem, observations should be negotiation outcomes, and questions should relate to term sheet components.
   When analyzing data, consider importance of understanding variability in term sheet negotiation goals and outcomes, with founder vesting showing high variability, board composition often converging to a 1-1-1 structure with some variations, and option pool typically settling in a narrow range of 12-15% despite varied initial goals.
   Based on the goal, technological feasibility, operational feasibility, and desirability, choose the best three terms from founder vesting, dividends, board of directors, and option pool. Create a scenario for each chosen term following the three-step approach above.
   Present your results in a table with the chosen terms as rows and columns for:
4. Definition (objects and relations)
5. Observation (in sentence form)
6. Question
   Also, provide a table showing the variability in goals and outcomes for these terms, sorted by outcome variability. Use a scale of 0-1, where higher scores indicate greater variability. Explain the implications of this variability for the development of the valuation advisor.
   In your response, use tables to present comparisons and lists. Provide concise explanations and insights based on the data and the probabilistic programming approach.

using the above prompt, i created interactive tool cld

interactive tools in option pool size, board structure, vesting schedule
which @miaomiaozhang20 found fascinating (thankfully!)

[hyunjimoon]

three theories using nsp cld

1. conversational inference equity valuation and allocation is the process that can demonstrate the value of pairing need of entrepreneur (probabilistic reasoning to drive action, to inform bundling choices, to lower complexity) with technology (probabilistic program). As a result of this principled process, quality of product 📦term sheet - cap table is increased.

2. scoping my vision to bridge probabilistic computing and bayesian entrepreneurship to conversational inference equity valuation and allocation, sequence of problem and solution can be:

Step	Substep	Term Sheet Thought Partner for Entrepreneurs
1. Problem		Entrepreneur's search for termsheet-investor fit is inefficient and learning is not cumulative for both entrepreneurs and investors.
2. Root Cause of 1		High uncertainty, information asymmetry, and lack of probabilistic reasoning tools in entrepreneurship.
	2.1 Nature	High-uncertainty scenarios require significant cognitive resources, leading to suboptimal collaboration terms.
	2.2 Individual (Entrepreneurs)	Low understanding of vague terms, hierarchical uncertainty (first on own idea value, then on negotiation power), limited investor interactions, noisy learning process.
	2.3 Individual (Investors)	May exploit information asymmetry, undervalue transparency in negotiations.
	2.4 Institution	Lack of education on uncertainty consideration, absence of user-friendly Bayesian tools for startup valuations, insufficient integration of prior information and robust decision-making in entrepreneurial processes.
3. Solution (for each root cause)		Develop a conversational thought partner with a term sheet world model that can be used in actual negotiation.
	3.1	Create an adaptive model for high-uncertainty scenarios, reducing cognitive load and facilitating optimal term decisions.
	3.2	Implement a shared world model to clarify vague terms, help estimate idea value and consequent negotiation power, simulate investor interactions, and reduce learning noise.
	3.3	Demonstrate value of transparency and collaborative decision-making to investors, showcasing how reduced information asymmetry benefits all parties.
	3.4	Develop user-friendly Bayesian tools that integrate prior information, handle diverse startup and VC data, and connect valuation estimates with negotiation strategies. Promote a culture of uncertainty consideration in decision-making.
4. How solution addresses problem's root cause		By providing a shared world model and probabilistic reasoning tools, the solution facilitates meaningful collaborative decisions and enables cumulative learning, addressing information asymmetry and high-uncertainty challenges.
5. Production plan		Develop a conversational AI tool integrating term sheet world model, probabilistic reasoning, and educational components. Partner with entrepreneurship programs and investors for implementation and continuous improvement.

3. rational meaning construction using probabilistic language of thought can be applied to unify distributional/statistical and pragmatic definition of meaning of (valuation, ownership, security) expressed in an integrated product (termsheet-captable) for taking intelligent risk together.

[hyunjimoon]

1. integrated process should come before integrated product

Product integration before process integration can lead to fragmented, inefficient systems prone to quality issues ⚠️, communication breakdowns 🔇, and difficulty in rapid iteration 🐌, ultimately resulting in higher costs, longer development times, and suboptimal performance.

---

Based on the Tesla Roadster story, we can extract several lessons about what can go wrong when the operations management principle "integrated process should come before integrated product" is not followed:

1. Fragmented Supply Chain: Tesla's decision to outsource manufacturing to multiple countries (England, Thailand, Taiwan) created a complex, geographically dispersed supply chain. This fragmentation made it difficult to coordinate and control the production process, especially for a new and complex product like an electric vehicle.

2. Lack of Expertise Alignment: By outsourcing to companies without specific experience in electric vehicle components (e.g., Excellent in Thailand for battery packs), Tesla risked quality and performance issues. This misalignment between the required expertise and the chosen suppliers' capabilities could lead to significant problems in the final product.

3. Communication and Coordination Challenges: With design in California and manufacturing spread across multiple countries, effective communication and coordination became extremely difficult. This separation between design and manufacturing teams can lead to misunderstandings, delays, and errors in translating design intent into manufactured products.

4. Increased Lead Times and Costs: The global supply chain, with components traveling from Asia to England and then finished products returning to California, likely resulted in longer lead times and higher logistics costs. This extended timeline can be particularly problematic when dealing with a new product that requires rapid iteration and problem-solving.

5. Difficulty in Rapid Iteration: When problems arise in a new product (as is common in innovative technologies), having a dispersed manufacturing process makes it much harder to quickly identify, diagnose, and solve issues. The physical and organizational distance between design and manufacturing teams can significantly slow down the iterative improvement process.

6. Loss of Learning Opportunities: By outsourcing manufacturing, Tesla missed out on valuable learning opportunities that come from closely integrating design and production. This knowledge gap can hinder future innovation and improvement in both product design and manufacturing processes.

7. Quality Control Challenges: With multiple suppliers and a complex supply chain, maintaining consistent quality control becomes more difficult. This is especially critical for a high-end product like the Tesla Roadster, where customer expectations for quality are very high.

8. Inflexibility to Market Changes: A fragmented process makes it harder to respond quickly to market feedback or changing customer needs. The ability to rapidly adjust the product or production process is limited when dealing with multiple external suppliers across different countries.

Applying this principle to the context of integrating reasoning processes before integrating language model products, we can argue:

1. Process Integration Priority: Just as Tesla should have focused on integrating its manufacturing process before attempting to produce an integrated electric vehicle, we should prioritize integrating our understanding and processes for reasoning and world model updating before rushing to integrate language model products.

2. Knowledge Alignment: Ensuring that the underlying knowledge and reasoning processes are well-aligned and integrated across different domains is crucial before attempting to create a unified language model product.

3. Iterative Improvement: Developing an integrated reasoning process allows for more rapid iteration and improvement in the underlying "world model," which can then be more effectively translated into language outputs.

4. Quality and Consistency: An integrated reasoning process can lead to more consistent and higher-quality outputs in language models, just as an integrated manufacturing process leads to better quality control in physical products.

5. Flexibility and Adaptability: By focusing on the integration of reasoning processes, we create a more flexible foundation that can adapt more quickly to new information or changing requirements in language model applications.

6. Long-term Efficiency: While initially more challenging, prioritizing the integration of reasoning processes can lead to more efficient and effective development of language model products in the long run.

7. Risk Mitigation: By ensuring that our reasoning processes are well-integrated before creating integrated language model products, we can better mitigate risks associated with inconsistent or flawed outputs, which is particularly important in the context of AI safety.

To prevent similar mistakes in this larger-scale issue, we can introduce system engineering principles for integration, focusing on:

1. Modular design of reasoning components that can be effectively integrated
2. Clear interfaces between different reasoning modules
3. Rigorous testing and validation of integrated reasoning processes
4. Iterative development cycles that allow for continuous improvement of the integrated reasoning system
5. Close collaboration between theoretical researchers, engineers, and end-users to ensure alignment of goals and capabilities

By following these principles, we can work towards creating more robust, reliable, and safe integrated language model products that are built on a foundation of well-integrated reasoning processes.

[hyunjimoon]

Unifying Linguistic Meaning in Term Sheets cld

• statistical (:= meanings as latent structure in usage, and the basis for semantic association and sequential prediction and response). measure of this can be implemented with probabilistic program world models

• pragmatic (:= meanings as communicative inferences, and the basis for social cooperation and coordination)

---

rational meaning construction integrates two 🎲🧠 and 🏗️🗣️ (my bet on probabilistic platform #244 (reply in thread))

1. 🎲🧠 probabilistic language of thought for constructing structured models of arbitrary situations. This component:

• Supports coherent belief updating and inferences
• Can be instantiated using probabilistic programs
• Offers structured representation for novel situations and problems
• Provides a framework for inferences with respect to queries and goals

2. 🏗️🗣️ general mechanism for constructing meaning from natural language. This component:

• Represents meaning as distributions over expressions in the language of thought
• Can be modeled as translation from utterances to expressions in a probabilistic programming language
• Can be implemented using code-trained large language models
• Acts as an efficient inference system between language and programs based on stored patterns in language and code

[hyunjimoon]

simulated answer of Charlie (given angie-vikash's mail, OM4ENT paper) by cld on "profile of entrepreneur that'd most willing/unsure/not willing to use Conversational inference system for startup valuation and equity allocation using language model probabilistic programming, based on his case study Angularity, Banza, BGI, MediTech, MicroMetal, Metropoli, MoS, Novaconfort, Renetech, SkinnyGirl Cocktails, Tesla Roadster, Unity Homes, VFA, ASB.

---

claude's answer

This table outlines features that might influence a company's likelihood to adopt and benefit from a conversational inference system for valuation and equity allocation. Companies with high complexity, rapid growth, frequent pivots, and a need for sophisticated scenario planning are more likely to find value in such a system. Conversely, companies with stable, traditional business models and simple equity structures might find less need for this type of tool.

Feature	High Likelihood	Medium Likelihood	Low Likelihood
Business Complexity	Multiple product lines or markets	Focused but growing	Single product/market
Growth Rate	Rapid, unpredictable	Steady growth	Slow or stable
Pivot Frequency	Frequent pivots	Occasional strategic shifts	Stable business model
Equity Structure	Complex cap table	Standard VC structure	Simple, family-owned
Industry	Tech, finance, or multi-sector	Consumer goods, services	Traditional manufacturing
Decision-Making Culture	Data-driven, innovative	Balanced approach	Conservative, intuition-based
Stakeholder Diversity	Multiple investor types	Typical VC and founder mix	Few stakeholders
Regulatory Environment	Complex, changing	Moderate regulation	Stable, simple regulations
Valuation Frequency	Frequent (e.g., multiple funding rounds)	Periodic	Rare
Technical Expertise	High in-house capability	Moderate capability	Limited technical resources
Openness to New Tools	Early adopters	Cautious adopters	Traditional methods preferred
Need for Scenario Planning	High (uncertain future)	Moderate	Low (predictable future)
International Expansion	Active or planned	Considering	Not applicable
Workforce Structure	Dynamic, options-heavy compensation	Mixed compensation	Traditional salary-based
Resource Availability	Well-funded, can experiment	Adequate resources	Resource-constrained

1. Most likely to use:

• Angularity: Their frequent pivots and need to assess new market opportunities quickly align well with a system that can rapidly model different scenarios in natural language.
• MicroMetal: As they expanded into various industries, a conversational system could help them model the impact on company valuation across different segments.
• Renetech: Their focus on technology solutions and need to model complex regulatory environments make them ideal candidates for a probabilistic programming approach.
• VFA: As a platform connecting startups and talent, they could use this system to provide value to their network by offering sophisticated equity modeling tools.
• ASB: As an educational institution, they might be very interested in this for research, teaching, and potentially for their own financial modeling.

2. Uncertain/Maybe:

• Banza: They might find value in modeling different growth scenarios and equity allocations as they scale, but may need convincing on the benefits over traditional methods.
• MoS: They could use this to model different equity scenarios as they expand their product line, but may need to see clear benefits over their current methods.
• Unity Homes: They might be interested in modeling different growth and equity scenarios, especially if expanding to new markets, but may need to see how it applies to their specific industry.
• Tesla Roadster: While they might benefit from sophisticated modeling, their unique market position and complex supply chain might require significant customization.
• BGI: They could potentially use this for modeling different scenarios in their rapidly evolving field, but may have concerns about integrating their specialized knowledge.

3. Less likely to use:

• MediTech: While they could benefit from sophisticated modeling, their specialized field might require significant customization of the system.
• SkinnyGirl Cocktails: Their brand-driven business model might be challenging to capture in probabilistic terms, making them less likely to adopt.
• Metropoli: As a traditional family-owned business, they might find less value in sophisticated equity modeling.
• Novaconfort: Their bootstrapped nature and family ownership might limit their need for complex equity allocation modeling.

---

angie's reflection

From the last column, it seems gpt is assuming high conservativeness in manufacturing industry. interesting that table considers not only the industry profile but also the founder profile i.e. who are likely to be in the given industry, hence assuming founder in manufacturing industry would be less openminded about AI tech. which is somewhat aligned with my experience - wonder what charlie'd think

[hyunjimoon]

mobility venture (jinhua, john, bhuvan are teaching) is multi-sector, regulatory Environment is complex and dynamic. i wonder how "options-heavy compensation" they are

[hyunjimoon]

This table provides a high-level overview of how each company (from mobility venture class) might benefit from and be able to implement our LLaMPPL-based system. The technical feasibility column highlights scenarios where our approach would outperform competitors, focusing on the complex, dynamic, and uncertain aspects of each business that our probabilistic programming approach is particularly well-suited to handle. The desirability column considers factors like complexity, growth, pivots, and planning needs. The operational feasibility column assesses the likely ability of each company to adopt and effectively use our system based on their technical capabilities, openness to new technologies, and industry characteristics.

Company	Description	Desirability 💜	Technical Feasibility 🟩	Operational Feasibility 🔴
Sun Mobility	EV battery swapping	💜💜	🟩🟩🟩 ⭐️	🔴🔴
DADA Nexus	On-demand logistics platform	💜💜💜 ⭐️	🟩🟩🟩 ⭐️	🔴🔴🔴 ⭐️
Fermata Energy	Vehicle-to-grid energy solutions	💜💜💜 ⭐️	🟩🟩	🔴
SparkCharge	Portable EV charging	💜💜	🟩	🔴🔴🔴
BETA Technologies	Electric vertical aircraft	💜💜💜 ⭐️	🟩🟩🟩 ⭐️	🔴🔴
TuSimple	Autonomous trucking	💜💜💜 ⭐️	🟩🟩🟩 ⭐️	🔴🔴🔴
BOT.AUTO	Autonomous driving solutions	💜💜	🟩🟩	🔴🔴
Optimus Ride	Autonomous mobility services	💜💜	🟩🟩	🔴
Automotus	Curb management platform	💜💜	🟩🟩	🔴🔴
Metropolis	Smart parking technology	💜	🟩	🔴🔴

⭐️ indicates a particularly high rating

DADA Nexus:
💜: Multi-sided platform with frequent pivots creates ideal use case.
🟩: LLaMPPL excels at modeling complex, dynamic multi-sided markets.
🔴: Tech-savvy with strong data infrastructure for implementation.

Fermata Energy:
💜: Complex energy systems with significant regulatory challenges.

BETA Technologies:
💜: Complex aviation tech intersects with high regulatory burden.
🟩: LLaMPPL suited for modeling intricate regulatory and tech development scenarios.

TuSimple:
💜: AV sector's regulatory complexity requires sophisticated planning.
🟩: Ideal for modeling complex AV deployment and evolving regulations.

[hyunjimoon]

@charsfine - resolution of nail period capitalization is getting higher (ignoring industry effect, comparing tusimple (AV truck) which just had one series A deal in 2015 then moved on to series B, sparkcharge and beta tech have at least 3 deals before series A. perhaps covid-investment freeze might be one factor

[hyunjimoon]

queries

three queries that needs to be answered are:

• "if the company sells for X at date Y, what is Angie's payout?"
• "what is the total payout for all VCs?"
• "under [[the following assumptions]] about the valuation of the company, increasing over time, what is the probability that Angie's payout is greater than $100K after 18 months?"

six components of DSL

Category	Items
a. 👨‍👩‍👧‍👦Entities	- Shareholder - ShareClass (Common, Preferred, Options) - Funding_Instrument (SAFE, Convertible Note) - Company
b. 📦Attributes	- Shares (number of shares) - Percentage (ownership percentage) - Price (price per share) - Valuation (company valuation) - Cap (valuation cap for SAFEs/Notes) - Discount (conversion discount for SAFEs/Notes)
c. 🏗️Operations	- issue_shares(shareholder, share_class, number_of_shares) - create_option_pool(percentage) - raise_funding(amount, instrument_type, terms) - convert_instrument(instrument, valuation) - dilute(new_shares) - calculate_ownership() - value_company(valuation_method, parameters)
d. 🎂Events	- Funding_Round - SAFE_Conversion - Option_Pool_Creation - New_Shareholder_Addition
e. 🚧Constraints	- total_percentage_sum = 100% - share_price = company_valuation / total_shares
f. 🤨Probabilistic Elements	- future_valuation(time_horizon, growth_rate, uncertainty) - exit_scenario(exit_type, probability, valuation_multiple) - conversion_probability(instrument, time_horizon)

meetly

1. Meetly's share development

Event/Shareholder	Stage 1	Stage 2	Stage 3	Stage 4 (Final)
Triggering Event	Initial	Advisor Addition	Option Pool Expansion	Series A & SAFE Conversion
Total Shares	9,000,000	9,100,000	9,600,000	12,523,018
You	4,950,000 (55%)	4,950,000 (54.40%)	4,950,000 (51.56%)	4,950,000 (39.53%)
Co-Founder	4,050,000 (45%)	4,050,000 (44.51%)	4,050,000 (42.19%)	4,050,000 (32.34%)
Advisor	-	100,000 (1.10%)	100,000 (1.04%)	100,000 (0.80%)
Option Pool/ESOP	1,000,000 (11.11%)	900,000 (9.89%)	1,400,000 (14.58%)	1,400,000 (11.18%)
Investor #1 (Series A)	-	-	-	3,213,023 (25.66%)
Carol (SAFE)	-	-	-	209,995 (1.68%)

2. Expressing Meetly situation with the DSL

Company Meetly {
    initial_valuation: $0
    authorized_shares: 15,000,000

    Stage Initial {
        issue_shares(You, Common, 4,950,000)
        issue_shares(Co_Founder, Common, 4,050,000)
        create_option_pool(11.11%)
    }

    Stage Advisor_Addition {
        issue_shares(Advisor, Common, 100,000)
        adjust_option_pool(9.89%)
    }

    Stage Option_Pool_Expansion {
        adjust_option_pool(14.58%)
    }

    Stage Series_A_and_SAFE_Conversion {
        raise_funding($X, Preferred, {shares: 3,213,023})  // X is the undisclosed Series A amount
        raise_funding($Y, SAFE, {shares: 209,995})  // Y is Carol's undisclosed SAFE amount
        convert_instrument(Carol_SAFE, $Z)  // Z is the undisclosed valuation at conversion
    }

    calculate_ownership()
    future_valuation(years: 5, growth_rate: 40%, uncertainty: 25%)
    exit_scenario(IPO, probability: 0.15, valuation_multiple: 8)
    exit_scenario(Acquisition, probability: 0.55, valuation_multiple: 4)
    exit_scenario(Failure, probability: 0.3, valuation_multiple: 0)
}

celestial

1. ⭐️celestial

Event/Shareholder	Stage 1	Stage 2	Stage 3	Stage 4
Triggering Event	Initial	Option Pool Creation	SAFE Conversion	Series Seed Funding
Total Shares	4,000,000	4,705,882	5,302,382	8,712,283
CEO	1,200,000 (30%)	1,200,000 (25.5%)	1,200,000 (22.63%)	1,200,000 (13.77%)
Founder 2	800,000 (20%)	800,000 (17%)	800,000 (15.09%)	800,000 (9.18%)
Founder 3	800,000 (20%)	800,000 (17%)	800,000 (15.09%)	800,000 (9.18%)
Founder 4	800,000 (20%)	800,000 (17%)	800,000 (15.09%)	800,000 (9.18%)
Prof. Rogers	400,000 (10%)	400,000 (8.5%)	400,000 (7.54%)	400,000 (4.59%)
Option Pool	-	705,882 (15%)	705,882 (13.31%)	1,937,727 (22.24%)
Series Seed-1 (SAFE)	-	-	331,389 (6.25%)	331,389 (3.80%)
Series Seed-2 (SAFE)	-	-	265,111 (5%)	265,111 (3.04%)
Series Seed Preferred	-	-	-	2,178,056 (25%)

2. Expressing Meetly situation with the DSL

Company Celestial {
    initial_valuation: $0
    authorized_shares: 10,000,000

    Stage Initial {
        issue_shares(CEO, Common, 1,200,000)
        issue_shares(Founder2, Common, 800,000)
        issue_shares(Founder3, Common, 800,000)
        issue_shares(Founder4, Common, 800,000)
        issue_shares(Prof_Rogers, Common, 400,000)
    }

    Stage Option_Pool_Creation {
        create_option_pool(15%)
    }

    Stage SAFE1_Conversion {
        raise_funding($500,000, SAFE, {cap: $8,000,000})
        value_company(pre_money, $8,000,000)
        convert_instrument(SAFE1, $8,000,000)
    }

    Stage SAFE2_Conversion {
        raise_funding($250,000, SAFE, {cap: $5,000,000})
        value_company(pre_money, $5,000,000)
        convert_instrument(SAFE2, $5,000,000)
    }

    Stage Series_Seed {
        value_company(pre_money, $12,000,000)
        raise_funding($4,000,000, Preferred, {price: $1.8365})
        adjust_option_pool(22.24%)
    }

    calculate_ownership()
    future_valuation(years: 5, growth_rate: 50%, uncertainty: 30%)
    exit_scenario(IPO, probability: 0.2, valuation_multiple: 10)
    exit_scenario(Acquisition, probability: 0.5, valuation_multiple: 5)
    exit_scenario(Failure, probability: 0.3, valuation_multiple: 0)
}

Company Meetly {
    initial_valuation: $0
    authorized_shares: 15,000,000

    Stage Initial {
        issue_shares(You, Common, 4,950,000)
        issue_shares(Co_Founder, Common, 4,050,000)
        create_option_pool(11.11%)
    }

    Stage Advisor_Addition {
        issue_shares(Advisor, Common, 100,000)
        adjust_option_pool(9.89%)
    }

    Stage Option_Pool_Expansion {
        adjust_option_pool(14.58%)
    }

    Stage Series_A {
        raise_funding($X, Preferred, {shares: 3,213,023})  // X is the undisclosed Series A amount
    }

    Stage SAFE_Conversion {
        raise_funding($Y, SAFE, {shares: 209,995})  // Y is Carol's undisclosed SAFE amount
        convert_instrument(Carol_SAFE, $Z)  // Z is the undisclosed valuation at conversion
    }

    calculate_ownership()
    future_valuation(years: 5, growth_rate: 40%, uncertainty: 25%)
    exit_scenario(IPO, probability: 0.15, valuation_multiple: 8)
    exit_scenario(Acquisition, probability: 0.55, valuation_multiple: 4)
    exit_scenario(Failure, probability: 0.3, valuation_multiple: 0)
}

[hyunjimoon]

Concept	Desirability Score	Feasibility Score	Reason behind scoring
Pre/Post Valuation	5	5	Critical for deal structuring; requires probabilistic reasoning for future valuations; easily implementable in CIVA's world model
Option Pool	5	5	Important for employee compensation and dilution calculations; useful for estimating future pool sizes; straightforward to implement
Liquidation Preference	5	4	Crucial for investor returns in various exit scenarios; requires modeling of exit events; implementable but complex waterfalls may be challenging
Anti-dilution Provisions	4	4	Important for protecting investor ownership; requires modeling of future rounds; implementable but may need careful modeling of different types
Conversion Rights	4	4	Important for preferred stockholders; needs reasoning for optimal conversion timing; can be modeled but may require complex decision logic
Preferred Stock Terms	4	4	Critical for investor rights and returns; useful for modeling different scenarios; implementable but may require detailed modeling of various terms
Drag-Along Rights	3	3	Important for facilitating exits; useful for modeling acquisition scenarios; can be implemented but may require complex modeling of shareholder decisions
Pay-to-Play Provisions	3	3	Important in down rounds or subsequent financings; requires modeling of future funding needs; implementable but may need detailed modeling of investor behavior
Vesting Schedules	3	4	Important for founder and employee retention; useful for modeling retention scenarios; relatively straightforward to implement
Registration Rights	2	2	Less critical in early stages, more important near IPO; limited need for probabilistic reasoning; can be modeled but may not provide significant insights in most scenarios
Information Rights	2	3	Important for investor oversight, but less critical for deal structure; limited need for probabilistic reasoning; can be implemented but may not significantly impact valuation models
Right of First Refusal	2	2	Important for controlling share transfers, but less critical for deal structure; limited need for probabilistic reasoning; can be modeled but may be complex to represent all scenarios

[hyunjimoon]

1. conversational inference

💡conversational inference ~ sequential Monte Carlo methods (marius cld)

level1: computation (input, output)

implement conversational inference by combining a probabilistic program with a language model, using an architecture for rational meaning construction.

application to addatics gpt

1. Scenario Modeling to Assess Investment Impact: The simulation tool allows Addactis to model various investment scenarios, helping them visualize how securing funding could accelerate product development, market expansion, and growth—all while testing their assumptions about the benefits and drawbacks of taking on investment.

2. Evaluating Trade-offs Between Ownership and Growth: By simulating the financial implications of different funding options, the tool helps Addactis understand the trade-offs between maintaining full ownership and achieving faster growth. This aids in testing their assumption about whether the dilution of ownership is justified by the potential for increased market share and revenue.

3. Aligning Investment with Contrarian Beliefs: The tool enables Addactis to explore how accepting investment can be aligned with their contrarian stance of emphasizing actuarial modeling over automation. Through conversational inference, they can test assumptions about using investment to strengthen their core principles, enhance modeling capabilities, and influence industry practices without compromising their foundational beliefs.

level2: representation and algorithmic

conversational inference for business negotiations and investment decisions. This approach involves decomposing complex problems into manageable sub-models, allowing for iterative updating of beliefs based on new information. It addresses resource-constrained rationality by providing tools to overcome cognitive limitations, enabling participants to tap into conditional beliefs and check consistency across different decision spaces. The process can be particularly valuable in persuading investors by breaking down gut feelings into more rational sub-beliefs and structuring deals that are mutually beneficial. By representing these models using concepts with probabilistic program, we suggest framework to simulate and infer complex scenarios, ultimately leading to more informed and rational decision-making in equity valuation and allocation. (Sampling-based algorithms for approximate inference, e.g.,MCMC, sequential Monte Carlo (“particle filtering”), importance sampling. Cost-sensitive sampling (“One and done”). Fast initialization with bottom-up recognition models (“Neural networks”). Resource-rational models.)

level3: hardward implementation

• mit inference stack: genparse, gensql?
• sequential monte carlo steering - embedding ignorance prior in llm cld

[hyunjimoon]

2. valuation and allocation

• what is utility being maximized for founder and investor and how are their constraints be gradually revealed (investors have more clearer constraints - and how this acts as advantage)?

---

with juan: based on that, we can structure the deal by then, what is my pre money valuation and how much should I invest? How much is the investment amount of a founder who believes that your valuation next round would be 10 million, and what would be the kind of maximum amount that they have to invest in. So I think it would be something like this, because this is the amount that they will going to invest. Let's say like this is an investor, and if you invest too little, then that is kind of doesn't make sense because you're investing because you believe in this company's future, right? So you want to have as much as ownership as possible at some sense, but it doesn't really make sense to invest, like, $100 million in a company that, who's you think will be ten million next year. So that would be some trade off in the optimization problem. So that's where the optimization comes in,

with scott: example is like Tesla and investor was a compass. The name was compass, had a shared country, and belief that EV market will grow. And based on the shared belief like how much money I should invest, if I invest too little, that is somewhat against the country, and belief that this Tesla would grow big, but if they invest too much, that's also somewhat not optimal. So I'm structuring that situation.

with tom: inverted U shape, because if you invest two less that is against your act of investing, because that means you don't really believe in this company's future, or if you invest too much like your expected valuation of this company next year is 10 million, but you're investing 100 million, that doesn't make sense, right? There would be some optimization.

level 1: computation

https://claude.site/artifacts/1ef0b158-4db1-4e5a-9c82-30bd038558ef from dsl for startup equity modeling cld

pre vs post money valuation from compare cld

level 2: algorithmic

Component	Description
Objective Function	x*(i,c) = argmax_(i, c) E[utility(i, c, M, s, f)]
Utility Function	utility(i, c, M, s f): investor's shares# * price per share after series A
Desirability STATE	Future valuation: post money-Gamma of mean M, sd 10, as a function of market trend * avg growth from population model
Feasibility STATE (function of ACT)	Investor state: - At most 1m investment: i < 1 - At least 10x: g(i, c, M) > 10 (todo1) Founder state: - At least 80% ownership after series A: h(i, c, M) > 0.8 (todo2)
ACT	Action: amount of investment i, value cap c
Shared assumption	Limit series A as 10% dilution
Tradeoff	Investor: wish to lower value cap to maximize conversion Founder: wish for higher value cap to minimize dilution
Rare event	If downturn (transportation < 10%), probability of losing money if i invest 1m? (todo3)

This table presents an optimization problem for a startup investment scenario. The key elements are:

1. An objective function that aims to maximize the expected utility.
2. A utility function based on the investor's shares and share price after Series A.
3. Desirability state considering future valuation.
4. Feasibility constraints for both the investor and founder.
5. Actions involving the investment amount and value cap.
6. A shared assumption about dilution in Series A.
7. Conflicting goals between the investor and founder regarding the value cap.
8. A consideration of a rare downturn event and its impact on the investment.

The table highlights the various factors and constraints that need to be balanced in this optimization problem, making it easier for readers to grasp the complex decision-making process involved in startup investments.

level 3: hardward

• adev, gensql

[hyunjimoon]

@jeff-dotson
using Decoding Venture Capital Decisions.pdf, "Entrepreneurial Finance- Selection and Deal Structuring" rows of table from #277 (comment), reorganized previous draft 🧠💻🧮phantom rmc.md, i made below to share one option of our paper direction. I need verification on whether 1) Expected Measurement Results in Code makes sense, 2) whether the output matches its expectation.

🚨disclaimer: "Expected Measurement Results in Code" has largest uncertainty as it's a prediction given one month outdated code and based on hoping filling

Why Investors' Words, Thoughts, Actions Differ and How Founders Can Measure and Use Their Gaps

Heterogeneity Phenomenon	Cause (Lay Theory Explanation)	How to Measure	Expected Measurement Results in Code	How Founders Can Use This
🌍 Spatial Heterogeneity: Different agents evaluate identical startups differently at the same time (e.g., Marc Andreessen vs. Peter Thiel)	Different investors develop distinct lay theories based on their unique experiences, expertise, and investment focus, creating systematically different attribute-to-perception mappings	Using hierarchical Bayesian modeling to identify investor clusters with similar evaluation patterns and determine which attributes most strongly differentiate these clusters	Check alpha matrix in model output for population-level patterns and Theta matrix for investor-type effects. Look for divergent coefficients mapping observable features (like patents) to perceptual dimensions. In the VC design grid validation, compare responses from different VC profiles on identical startup descriptions (e.g., cosine_df values < 0.7 between investor profiles evaluating same startup)	Segment investors into actionable archetypes (e.g., "technology-optimists" vs. "market-realists") and focus fundraising efforts on groups most likely to have aligned priors; start with broader categories and increase segmentation granularity only when necessary to find at least one aligned investor
⏰ Temporal Heterogeneity: Same agent evaluates identical startups differently at different times (e.g., early-stage investors shift toward less risky, cheaper-to-test ideas when anticipating capital scarcity in later funding stages)	Individual investors adjust their lay theories in response to changing market conditions, causing them to systematically alter their attribute-to-perception mappings based on anticipated future capital availability	Track changes in a single investor's language patterns and decision criteria when evaluating identical startups across different market cycles; measure shifts in attribute weighting over time	Check InvestmentUtil changes for identical startup profiles across simulated market conditions. Compare would_invest values in VC responses for capital-intensive vs. capital-efficient startups (e.g., infrastructure vs. application startups) under different capital scenarios. Look for features with temporal coefficient variation > 0.5 in the alpha matrix when conditioning on market state	Time fundraising strategically based on when your target investors become more receptive during specific market conditions; during anticipated capital constraints, emphasize capital efficiency and short validation cycles; adapt messaging to match the current dominant decision criteria of your highest-probability investors
🔄 Internal Heterogeneity: Same agent displays different evaluation criteria in statements versus actions at the same time (e.g., VCs explicitly prioritize teams while implicitly selecting based on business model validity)	Investors maintain parallel explicit and implicit lay theories that may be inconsistent with each other, leading to divergence between stated evaluation criteria and actual investment decisions	Compare investors' stated importance rankings with their actual decision patterns when presented with forced trade-offs; measure differences between verbal evaluations and actual investment choices for identical startups	Compare stated preferences in vc_output text vs. actual would_invest decisions. Look for cases where VC rates "good_execution" high but still rejects startups with strong teams, or where "attractive_market" is rated low but investment occurs anyway. Examine divergence between explicit statements in LLM responses and binary investment decisions (disagreement rate > 30% indicates significant internal heterogeneity)	Develop pitches that address both explicit criteria (team quality) and implicit criteria (business model validity); focus on signals that reliably predict investment decisions rather than just those mentioned as important; test pitch effectiveness by measuring both verbal feedback and investment interest

using Updating Venture Capital Decision-Making Research cld

[hyunjimoon]

Visualizing Optimal Investment Decisions at 🌱 cld created value cap by investment

[hyunjimoon]

3. world model of cap table and term sheet

relevant hbs case studies according to NSS

Stage	Description	Case Studies	Rationale for Classification
Nail It	- Focus on finding product-market fit - Developing initial value proposition - Securing early-stage funding - Characterized by experimentation and pivoting	1. SAFE Notes: An Introduction 2. Fondeadora 3. Convertible Notes in Early-Stage Financing 4. Fast Ion Battery	- Convertible notes are typically used for very early-stage funding, when the company is still figuring out its business model - Fast Ion Battery case discusses a company running out of money and not meeting milestones, indicating it's still in the early stages of development
Scale It	- Rapid growth phase - Building capabilities for expansion - Securing larger rounds of funding (e.g., Series A, B) - Focus on customer segmentation and process development	1. Evaluating Venture Capital Term Sheets 2. ELCA's Series A Cap Table Exercise 3. Venture Capital Term Sheets 4. Financing Matillion's Scaleup (B1): Matthew Scullion	- These cases deal with more structured VC deals and Series A funding, typical for companies that have found product-market fit and are ready to grow - They involve more complex financial structures, indicating a more mature stage than "Nail It"

1.NAIL

SAFE

Fondeadora

SAFE Notes: An Introduction

---

Convertible Notes in Early-Stage Financing

This technical note introduces convertible note financing for early-stage start-up companies. These unpriced securities offer significant advantages related to delayed valuation, greater speed, and lower cost of completion compared to venture capital financing. As a result, the number of early-stage companies raising capital through convertible notes over the past decade has greatly increased. The note discusses the most frequently used terms and arrangements of early-stage convertible notes, the estimation of noteholders' equity ownership from delayed valuation, and the costs and risks of this form of financing to both entrepreneurs and investors. It offers a number of numerical examples that walk students through calculations of convertible note payoffs under different success scenarios and different valuation caps. It also explores how convertible notes affect subsequent equity rounds' pricing. Finally, the note offers statistical data on the state of convertible note issuance. This note is a successor to an older note, ""Convertible Notes: A Form of Early-Stage Financing"" (UVA-F-1925). The cases ""Should udu a Convertible Note?"" (UVA-F-2025) and ""MedMetri

Fast Ion Battery

A battery company is running out of money and has not met all of its milestones. The year is 2012 and many cleantech investors have decided to pull back. Fast Ion Battery needs a $5 million bridge round, but one of its three investors has refused to continue funding the company. John Davidson, a partner at Ware Street Capital and the lead investor in Fast Ion, must decide whether or not to save the company. John Davidson, a partner at Ware Street Capital (WSC) and a board member at Fast Ion Battery, had just received a phone call from Don Lerner at Bluelock Ventures telling him that Bluelock would not participate in the $ 5 bridge financing for Fast Ion Battery. Lerner's call could not have come at a worse time. Fast Ion was running out of cash and needed another round of financing urgently to continue developing its revolutionary battery. Davidson faced a real dilemma. On the one hand, the company was finally gaining traction with developing its technology and the search to replace the current CEO had yielded two prospective candidates who were extremely well-suited to drive the company forward with a more capital efficient business model. On the other hand, Ware Street Capital and the two other investors had already invested $10 million into a company that had not performed up to investors' expectations. Would they be throwing more good money after bad by providing the bridge financing? Moreover, would other, later stage, investors be willing to provide the significant amounts of capital required to get the company to an exit event given the changing climate for clean tech investments? These questions became more pressing following Lerner's call that Bluelock had chosen not to continue backing Fast Ion. Was Fast Ion Battery worth saving?

---

2.SCALE

ELCA's Series A Cap Table Exercise

In ELCA, the company must decide between two term sheets: one put forth by STV and one put forth by ESV. This exercise is an analysis of the implications of these two term sheets on the ownership structure and the payouts of common and preferred shareholders in case of an exit. An analysis of the other terms and provisions, such as those regarding control rights or vesting, are not explored in this exercise. For an additional overview of common terms of venture capital investment structures, please see A Note on Private Equity Securities (200-027) and Financing New Ventures (811-093).

Evaluating Venture Capital Term Sheets

When entrepreneurs are successful in convincing venture capital firms that they are an attractive potential investment, they are presented with offers detailing many terms of the investment agreement. These are described in term sheets. Sorting through the myriad terms can be a daunting proposition for an entrepreneur. Yet, it is important for entrepreneurs to understand the terms of a proposed financing. These determine the payout the entrepreneurs will receive when the company is liquidated or sold (either to another company or to the public through and IPO), the dilution the entrepreneurs will suffer in the event of a future down round of financing, control of the board of directors, and other important matters. The Series A terms will also set a precedent for the terms of future financing rounds. This case presents a situation in which entrepreneurs receive term sheets from two venture capital firms. The two term sheets differ in many ways, and students are asked to evaluate them from the perspective of the entrepreneur.

Venture Capital Term Sheets

Finance and investing, Entrepreneurship

Financing Matillion's Scaleup (B1): Matthew Scullion

Negotiation strategies, Finance and investing, Entrepreneurship

[hyunjimoon]

Concept	Test Two Choose One (T2C1)	Marr's Three Levels	Information Relaxation
Purpose	Guide entrepreneurial decision-making	Analyze cognitive and information processing systems	Optimize decision-making under uncertainty
Structure	1. Test two strategies 2. Choose one	1. Computational theory 2. Representation and algorithm 3. Hardware implementation	1. Relax information constraints 2. Introduce penalties 3. Solve relaxed problem
Key Benefit	Reduces uncertainty through comparison	Provides comprehensive view of system functioning	Yields performance bounds and simplifies complex problems
Decision Mechanism	Comparative analysis and choice	Hierarchical analysis from abstract to concrete	Mathematical optimization with relaxed constraints
Information Use	Creates information through internal consistency	Describes how information is processed at each level	Allows access to future information (hypothetically)
Application	Entrepreneurship and strategy	Cognitive science and AI	Operations research and finance

Level	💡 DSL for data-driven expert systems	🤜 Conversational Inference & Term Optimization	💻 Four Baselines
Computational Theory	Objective: Optimize SAFE terms Logic: Bayesian inference and decision theory Constraints: Incomplete market information Representation: Probabilistic models of startup value	Objective: Elicit feasibility and desirability world model Logic: Natural language processing and optimization Constraints: Limited founder knowledge Representation: Natural language queries and responses	Objective: Estimate startup value and allocate equity Logic: Statistical inference and simulation Constraints: Limited data and complex terms Representation: Statistical models and Monte Carlo simulations
Representation and Algorithm	- Probabilistic programming languages - Bayesian inference algorithms - Decision trees for term structure	- NLP models for query understanding - Bayesian optimization for term selection - Reinforcement learning for conversation flow	- Regression models (IV) - Monte Carlo simulation (IVA) - In-context prompting (CIVA_vanilla) - Probabilistic programming with LMs (CIVA)
Hardware Implementation	- Cloud-based probabilistic inference engines - GPU acceleration for Bayesian computations	- Distributed systems for real-time NLP processing - Parallel computing for optimization algorithms	- Standard CPUs for simple models (IV, IVA) - GPUs for language model inference (CIVA_vanilla, CIVA) - Specialized hardware for probabilistic programming (CIVA)

[hyunjimoon]

based on chat with charlie and tom, tom|safe.txt, charlie|safe.txt I made below and @tomfid offered help in choosing two key tradeoff i can start from modeling.

Tradeoff	Description	Considerations for Founders	Considerations for Investors
Expected Return vs. Variability	Balancing the potential for high returns against the risk of losses	- Higher valuation cap may lead to larger potential returns but could deter some investors - Lower valuation cap may attract more investors but could lead to greater dilution	- Higher valuation cap offers potential for larger returns but increases risk - Lower valuation cap provides more downside protection but may limit upside
Value Creation vs. Value Capture	The balance between growing the overall value of the company and securing a larger share of that value	- Larger investment amounts can accelerate growth but may lead to more dilution - Smaller, strategic investments might preserve equity but could slow growth	- Larger investments may help the company succeed but could increase risk exposure - Smaller investments might limit influence but could allow for a more diverse portfolio
Flexibility vs. Certainty	The tradeoff between adaptable terms and clearly defined outcomes	- More flexible terms (e.g., no valuation cap) could attract investors but may lead to uncertainty in future rounds - Clearly defined terms provide certainty but might limit options in future financing	- Flexible terms offer potential for higher returns in successful scenarios - Well-defined terms provide more predictable outcomes but might cap upside
Control vs. Capital	Balancing the need for funding against maintaining decision-making power	- Larger investments might come with more investor influence - Smaller investments or certain SAFE terms might preserve more control	- Larger investments or specific terms might grant more influence over company decisions - Smaller investments might limit control but allow for involvement in more companies
Short-term vs. Long-term Focus	Balancing immediate capital needs against long-term growth and valuation	- Higher valuation cap might limit immediate funding but could be beneficial in later rounds - Lower valuation cap might provide more immediate capital but could impact long-term equity position	- Higher valuation cap might limit immediate equity but could lead to larger long-term gains - Lower valuation cap provides more immediate equity but might signal less confidence in long-term prospects

[hyunjimoon]

perplexity

Factor	What Founders Should Remember	Emoji
Valuation Cap	Lower cap = More dilution for you, better deal for investors	🎩
Discount Rate	Higher discount = More shares for investors, less for you	💸
Pro Rata Rights	Giving these = Happy investors, but less room in future rounds	🔒
Conversion Triggers	Flexible triggers = More control for you	🎚️
MFN Clause	Including this = Protected early investors, but less flexibility later	🛡️
Future Impact	Complex terms now = Headaches later (cap table, fundraising, conflicts)	🔮

[hyunjimoon]

using safe agreement dashboard cld

1. Key Terms Comparison

2. Valuation Cap Impact

3. Summary and Recommendations

at https://claude.site/artifacts/8a5a7200-dba7-40b5-9aa9-40bf103f7a60

[hyunjimoon]

using comparing SAFE document tradeoffs which focus on capturing grow fast vs control tradeoff as below

Feature	SAFE 1 (Growth-Focused)	SAFE 2 (Control-Focused)
Option Pool Size	15%	10%
Valuation Cap	$8 million	$10 million
Investment Amount	$1 million	$1 million
Implied Dilution for Founders	Higher	Lower
Talent Attraction Potential	Higher	Moderate
Growth Speed Potential	Faster	Moderate
Founder Control	Lower	Higher
Market Growth Factor	Higher (e.g., 60)	Lower (e.g., 50)
Company-Specific Factor	Higher (e.g., 70)	Lower (e.g., 60)

at https://claude.site/artifacts/2bd637f1-c882-4151-a3ea-67595741af31

[hyunjimoon]

todo1: feed in Wasserman, N. (2013). The founder’s dilemmas: Anticipating and avoiding the pitfalls that can sink a startup (1st ed.).
todo2: extract principles and feed into termsheet a vs termsheet b

[hyunjimoon]

5. rational meaning construction (RMC)

RMC concept is proposed in https://arxiv.org/pdf/2306.12672 summarized below talk by josh tenenbaum

which i applied to term sheet interpretation, criticizing the current approach of focusing too much on statistical, not pragmatic linguistic meaning

Board Composition:

• Early-stage: Typically 3 members, emphasis on technical expertise.
• Growth-stage: 3-5 members, focus on broader business and financial acumen.

Industry-Specific Contexts:

• Semiconductor industry: Larger option pools and frequent refreshes to attract/retain technical talent.
• Pharmaceutical industry: Longer vesting periods and boards with scientific expertise due to extended development cycles.

Impact of Unexpected Events:

• A global chip shortage could transform the meaning of a large option pool in a semiconductor startup from signaling growth to indicating defensive talent retention.

---

Jeff suggested conjoint analysis in #253 and i find
[R3] JBV-2011-Unpacking the uncertainty construct Implications for entrepreneurial action.pdf relevant.

1. Conjoint studies (grounded in information processing, which requires participants to respond to a series of decision “scenarios.”, underlying factors responsible for the respondent's decision policy are decomposed using hierarchical liner modeling techniques, had great impact in marketing, psychology, strategic management fields (Green and Srinivasan, 1990)'s application in entrepreneurship
2. Choi and Shepherd (2004) employed conjoint analysis to consider how stakeholders evaluate a venture's newness,
3. Shepherd et al. (2000) investigated how venture capitalists assess the profitability of a new firm depending on its strategy.
4. this paper as they hypothesize decision-makers will use a contingent decision policy, conjoint analysis is a highly appropriate method to investigate the evaluation policies of the sample without relying on the respondents' introspection, which is often biased and inaccurate (Fischhoff, 1982; Priem and Harrison, 1994).

Variable	Low*	High
State uncertainty (rate of demand change)	The demand for your product is likely to fluctuate, but the rate of change is moderate and steady.	The rate of demand for your product will fluctuate significantly.
State uncertainty (rate of technological change)	Future technological innovations affecting the viability of the product are likely to occur, but they are likely to be incremental (not discontinuous).	Future technological innovations affecting the viability of the product are likely to be frequent and major, including changes to the underlying technologies related to product usage.
Effect uncertainty (predictability of demand change)	You have strong idea of your customers' preferences and demands with regard to your product, and these are predictable over time.	It is not possible to predict in advance demand changes affecting the viability of the product.
Effect uncertainty (predictability of technological change)	You are in a strong position to predict the nature and source of innovations that affect the viability of the product.	It is not possible to predict with any certainty the kinds or timing of future technological innovations that will affect the viability of the product.
Response uncertainty (ability to sustain innovative leadership)	You have tangible reasons to believe that your firm has the ability to sustain viability in this product market through further radical and/or incremental innovations.	It is not possible to foresee the ability of your firm to sustain viability in this product market through further radical and/ or incremental innovations.
Response uncertainty (potential lead-time over competitors)	While you are not able to fully predict the speed or nature of action of your competitors, you believe that if you act your product will enjoy advantages long enough to realize entrepreneurial returns.	You have no insights as to how your competitors will react in response to your product introduction, and therefore you cannot predict how long your product will enjoy advantages before a competitive response erodes profits.
Launch method (small vs, large scale)	You plan to launch the product on a small scale, to a limited number of customers.	You plan to launch the product on a large scale, to the national market immediately.

[hyunjimoon]

test snippets for term sheet advisor world model

1. carta

Mary gives you $1 million on a safe note, but remember, when you raise on a safe you're not actually giving Mary any shares yet. You're going to give her shares in the future, but not today, meaning right now, you don't need to set any kind of share price. Now, it's important to remember setting a share price is a big deal, because when you do that, you're placing a value on your company. For example, if you have 100 shares, and the price per share is $1 then guess what? Your company is valued at $100 this is what a valuation is. So back to Mary, since she's giving you money on a safe, you're not bringing shares into the picture yet, which means you're not setting a valuation for your company. This is pretty risky for Mary, after all, she's giving you a million bucks without yet knowing the current value of her investment. So in order to protect Mary, you need to provide her with a valuation cap. You set this valuation cap at $5 million all right, cut to a year later. You're now at your series A and you have a new investor named William. This time, it's a priced round, meaning William gives you money, and for that money, you give him shares. This means you now have a share price, which means you've officially set a valuation for your company. You and William decide that your whole company is worth 10 million and each share is worth $1 this means in total, your company has 10 million shares, and the price for each share is $1 for his investment, William gives you the same thing Mary did $1 million and in exchange, you give him 1 million shares, amounting to a 10% stake in your company. Just to recap, the valuation is 10 million and the share price is $1 great. So it's now time to convert Mary's investment from a year ago into shares. This is where the valuation cap comes in. Remember, William valued your company at ten million but because Mary set a valuation cap of 5 million, you're actually gonna give Mary her shares as if the company is valued at $5 million today. So what does that actually mean? You gave William a million shares, right? And Mary invested the same amount of money that he did. So how many shares are you supposed to give to Mary? Luckily, the math to figure this out is pretty simple. First, you just need to figure out Mary's share price. To do this, just take the valuation cap that you set with her and divide it by the official valuation that you set with William in the series A this gives you Mary's share price. And notice, while William's paying $1 a share, Mary will actually receive her shares at 50 cents per share. Now Mary invested $1 million at 50 cents per share, meaning when you convert her safe investment, you'll give her 2 million shares or a 20% stake in your company. Right away, you can see how the valuation cap is an incentive for Mary to come in early and take on that extra risk. she and William both invested the same amount, but since Mary came in earlier she actually got twice the number of shares that William, did this is why the valuation cap is an important incentive to get early stage investors on board and the.

[hyunjimoon]

4. competitor analysis

	📊 IV	💹 IVA	🔄 CIVA_vanila	🦮 CIVA (our approach)
F. Function	inference, valuation	inference, valuation, allocation	conversational inference of equity valuation and allocation	conversational inference of equity valuation and allocation
F1. Model	Statistical model with features of company (number of employees, Intellectual property measures, founder characteristics ) regression	Monte Carlo simulation	In-context prompted LLMs, without language model probabilistic programming	Probabilistic program with language model probabilistic programming (Q2. GenParse, hfppl)
F2. Advantages	Simple, low-cost	Can model some complex terms	Natural language interface	Formal modeling, efficient inference, flexible constraints
F3.Limitations	Cannot handle complex terms, not bespoke, static	Cannot model the range of possible valuations, lacking the ability to account for causal structure from terms and equity ownership, and struggling to model complex terms like liquidation preference waterfalls	Inconsistent handling of constraints, limited to greedy optimization	Requires more complex implementation, potential computational cost
M. Modeling the world
M1. Causal Complexity	Cannot handle complex terms	Can handle some complex terms (e.g. liquidation waterfall)	Can handle complex terms, but may struggle with consistency	Can handle complex terms with formal, consistent representation
M2. Dynamicity	Cannot model contingency on a probabilistic time series model of the business value, the cap table, and key business terms	Dynamic	Dynamic	Dynamic with explicit time-series modeling
M3. Estimation of Rare Event Probability e.g. regulatory approval, technology milestones, partnership, accidents	Poor	Limited	Poor	Efficient (Q1. via what?)
M4. Format of Input Data and Interface	valuation time series from Pitchbook/Crunchbase Database (+ textual knowhow on valuation from VC reports?)	cap table	cap table, term sheet	cap table, term sheet
O. Optimizing allocation given possible world models
O1.optimization method of expected value of possible world models		grid/random search, gradient descent		bayesian optimization using ADEV_jax

[hyunjimoon]

given term sheet below, CIVA_vanila outperformed me on answering twelve questions from entrepreneurial finance class #257 QUIZ 😅

SUMMARY OF TERMS FOR SERIES A CONVERTIBLE PREFERRED STOCK OF NEWCO, INC.

This summary of terms (the "Term Sheet") outlines the non-binding terms and conditions of a proposed investment by Bluelock XII, Limited Partnership and certain of its partners and affiliates ("Bluelock") and other investors listed below (collectively, the "Investors") in Newco, Inc. (the "Company").

Founders: Jane Whitman and Dick Watson

Type of Security: Series A Convertible Preferred Stock of the Company ("Series A Preferred")

Pre-Money Valuation: $8,000,000

Fully-Diluted Post-Money Valuation: $12,000,000

Number of Shares/Purchase Price: 2,000,000 shares of Series A Preferred at $2.00 per share (the "Purchase Price")

Aggregate Purchase Price: $4,000,000

Investors:

• Bluelock: 1,300,000 shares
• Delta Ventures: 500,000 shares
• Omega Partners: 200,000 shares

Current Capitalization Table:

Name:	Shares Owned:
Jane Whitman	1,400,000
Dick Watson	600,000
Max Grant	200,000

Option Pool:

A minimum of 30% of the fully diluted shares outstanding post-financing (1,800,000 shares), will be reserved for issuance to employees (other than Founders).

Additional Investment:

At the election of the Investors, and upon the satisfaction of certain milestones, the Investors, on a pro rata basis, may make an additional investment of up to $1,000,000 on the same terms herein for a period of nine (9) months from Closing.

Rights, Preferences, Privileges and Restrictions of Series A Preferred:

1. Dividend Provisions: No dividend or distribution may be paid to holders of common stock of the Company ("Common Stock") unless a like dividend or distribution is paid to holders of Series A Preferred. Cumulative dividend of 8% per year payable upon liquidation, dissolution, redemption or conversion.

2. Liquidation Preference: In the event of any liquidation or winding up of the Company, the holders of the Series A Preferred shall be entitled to receive in preference to the holders of Common Stock, an amount equal to two (2) times the Purchase Price per share plus any unpaid dividends ("Liquidation Preference"). After payment of such sum to the holders of the Series A Preferred, the holders of Common Stock (including Common Stock on an as converted from Series A Preferred Stock basis) shall be entitled to receive the remaining assets of the Company on a pro rata basis. A merger, consolidation (other than one in which the holders of capital stock of the Company prior to such merger or consolidation continue to hold at least 51% by voting power of the capital stock of the surviving entity), sale or exclusive license of all, substantially all or a significant portion of the assets or intellectual property of the Company shall be deemed to be a liquidation event.

3. Optional Conversion: The holders of Series A Preferred shall have the right to convert the Series A Preferred, at the option of the holder, at any time, into shares of Common Stock. One share of Series A Preferred shall initially be convertible into one share of Common Stock.

4. Automatic Conversion: The Series A Preferred shall automatically be converted into Common Stock at the then applicable conversion rate in the event of either (i) the closing of an underwritten initial public offering with aggregate offering proceeds to the Company of at least $20,000,000 and a per share price to the public of at least three times the Purchase Price (a "Qualified Public Offering") or (ii) the election of the holders of at least a majority of the outstanding shares of Series A Preferred.

5. Antidilution Provisions: The conversion price of the Series A Preferred (the "Series A Conversion Price") shall be subject to adjustment to prevent dilution in the event that the Company issues additional shares of Common Stock or any right or option to purchase, or any other security convertible into, Common Stock (other than shares issued to employees, consultants or directors in accordance with plans approved by the Board of Directors) at a purchase price less than the then applicable Series A Conversion Price. In such an event, the Series A Conversion Price shall be adjusted to such lesser purchase price.

6. Pay-to-Play: If the company issues shares of its capital stock in a future financing for a consideration per share less than the applicable Series A Conversion Price in effect immediately prior to such issuance (a "Dilutive Issuance"), and an Investor has the right to purchase its pro rata share of such Dilutive Issuance and Investor does not purchase such pro rata share, then the shares of Series A Preferred then held by such Investor shall be converted into Common Stock at the Series A Conversion Price in effect immediately prior to such issuance.

7. Drag-Along Rights: If (i) any person or entity offers to acquire all or substantially all of the assets or business of the Company by merger, sale of assets or otherwise, and (ii) such transaction is approved by 51% of the voting power of the shares of Series A Preferred then held by all Investors, then the Founders and each holder of Series A Preferred shall be obligated to vote all of his, her or its shares in favor of such transaction.

8. IPO Participation Rights: Subject to cutback or elimination as a result of regulatory issues affecting the Company's initial public offering ("IPO"), the Investors shall have the right to purchase 15% of the equity securities to be sold in the Company's initial public offering (the "IPO Right"), valued at the initial public offering price, pro rata in proportion to the number of shares of Series A Preferred they hold.

9. Redemption Provisions: Shares of Series A Preferred shall be redeemed by the Company in three equal annual installments (as legally permissible) at any time, commencing 60 days after receipt by the Company from the holders of a majority of the then-outstanding shares of Series A Preferred, of written notice requesting redemption of all shares of Series A Preferred. The redemption price shall be the Liquidation Preference plus any declared but unpaid dividends.

10. Voting Rights: Each share of Series A Preferred shall represent that number of votes equal to the number of shares of Common Stock issuable upon conversion of a share of Series A Preferred. The Series A Preferred and Common Stock shall vote together as a class except (i) the Series A Preferred as a class shall be entitled to elect three members of the Board of Directors (the "Series A Directors"), (ii) as required by law, and (iii) as set forth in Section 11 below.

11. Protective Provisions: The consent of the holders of a majority of the Series A Preferred shall be required for, among other things: (i) an amendment or repeal of any provision of the Company's Certificate of Incorporation or Bylaws if such action would change the rights, preference or privileges of the Series A Preferred; (ii) the authorization or issuance of any class of stock having any right, preference or priority not subordinate to the Series A Preferred; (iii) the payment of dividends; (iv) a merger, sale of all or substantially all of the assets, recapitalization, reorganization, liquidation or dissolution of the Company; (v) the redemption, retirement, purchase or acquisition, directly or indirectly, through subsidiaries or otherwise, of any shares of capital stock (other than repurchase of Common Stock at cost upon termination of employment or service); and (vi) the entering into any lines of business that are not primarily related to the business of the Company as conducted as of the Closing. The Company shall not without the approval of at least two Series A Directors, (i) grant an exclusive license to any of the Company's material intellectual property rights; or (ii) acquire all or substantially all of the properties, assets or stock of any other company or entity; or (iii) incur indebtedness in excess of $1,000,000 in the aggregate.

Information Rights

The Company shall furnish each holder of Series A Preferred with an annual budget, annual financial statements audited by an accounting firm of national reputation and quarterly and monthly unaudited financial statements. These obligations shall terminate upon a Qualified Public Offering.

Registration Rights:

Customary registration rights.

Rights to Purchase Additional Shares: Holders of Series A Preferred shall have the right to purchase their pro rata share (based on their ownership of the outstanding shares of Series A) of any future equity offering by the Company, subject to customary exclusions. Such rights will terminate upon a Qualified Public Offering.

Board Composition:

The Board of Directors shall consist of five (5) persons, comprised of two (2) representatives designated by Bluelock; and 1 representative designated by the remaining Investors; 1 representative designated by the Founders; and the Chief Executive Officer of the Company.

Founder's Stock:

Each Founder shall execute a Stock Restriction Agreement covering the shares of Common Stock held by such Founder as of Closing (the "Founder Shares"), which provides that any unvested Founder Shares may be repurchased by the Company for their original issue price in the event the employment of such Founder is terminated for any reason and that any vested Founder Shares may be repurchased by the Company for fair market value ("FMV") in the event the employment of such Founder is terminated by the Company for cause or by the Founder without good reason.

[hyunjimoon]

Q.

1. What is the pre-money valuation of the Company?
2. What is the post-money valuation of the Company?
3. What percent of the post-money valuation is represented by the Investors' cash investment?
4. How many shares are actually outstanding prior to this investment?
5. How many shares are deemed to be outstanding prior to this investment?
6. How many shares are actually outstanding after this investment?
7. What percent of the outstanding shares (on a Common Stock equivalent basis) will the Investors own immediately after this investment?
8. If the Company were sold for $8,000,000 immediately after this investment, how would the proceeds be split up?
9. If the Company were sold for $40,000,000 immediately after this investment, how would the proceeds be split up?
10. List at least five (5) significant issues to be negotiated from the Founders' and Company's point of view:
11. How is control of the Company exercised?
    A. Board representation
    B. Stockholder Votes
    C. Protective Provisions (also known as negative covenants)
    D. All of the Above
12. As a "thank you" for your help, which would you prefer (assuming the term sheet was successfully negotiated so that the terms of the Series A provide for non-participating preferred, no cumulative dividends, and no multiple liquidation preference)?
    A. A warrant to purchase 10,000 shares of Series A at a purchase price of $2.00 per share.
    B. A warrant to purchase 10,000 shares of Common Stock at a purchase price equal to FMV of Common Stock at time of issuance of the warrant.

---

CIVA_vanila's answer (claude)

1. The pre-money valuation of the Company is $8,000,000.

2. The post-money valuation of the Company is $12,000,000.

3. The Investors' cash investment represents 33.33% of the post-money valuation ($4,000,000 / $12,000,000).

4. The number of shares actually outstanding prior to this investment is 2,200,000 (1,400,000 + 600,000 + 200,000).

5. The number of shares deemed to be outstanding prior to this investment is 4,000,000 ($8,000,000 pre-money valuation / $2.00 per share).

6. The number of shares actually outstanding after this investment is 4,200,000 (2,200,000 existing shares + 2,000,000 new Series A Preferred shares).

7. The Investors will own 47.62% of the outstanding shares immediately after this investment (2,000,000 / 4,200,000).

8. If the Company were sold for $8,000,000 immediately after this investment:

   • Series A Preferred would receive all $8,000,000 due to their 2x liquidation preference ($4,000,000 * 2 = $8,000,000)
   • Common stockholders would receive $0

9. If the Company were sold for $40,000,000 immediately after this investment:

   • Series A Preferred would receive $8,000,000 (2x liquidation preference)
   • Remaining $32,000,000 would be split pro-rata:
     • Series A Preferred: $15,238,095 (47.62% of $32,000,000)
     • Common stockholders: $16,761,905 (52.38% of $32,000,000)
   • Total distribution:
     • Series A Preferred: $23,238,095
     • Common stockholders: $16,761,905

10. Five significant issues to be negotiated from the Founders' and Company's point of view:
    a. Liquidation preference (currently 2x)
    b. Anti-dilution provisions
    c. Board composition
    d. Protective provisions
    e. Pay-to-play provision

11. Control of the Company is exercised through: D. All of the Above (Board representation, Stockholder Votes, and Protective Provisions)

12. Given the choice, I would prefer: B. A warrant to purchase 10,000 shares of Common Stock at a purchase price equal to FMV of Common Stock at time of issuance of the warrant.

    Rationale: Option B is likely more valuable because:

    1. The FMV of Common Stock is likely lower than $2.00 per share, making the exercise price potentially lower.
    2. Common Stock doesn't have liquidation preferences, potentially allowing for more upside in an exit scenario.
    3. There's no specified expiration date for this warrant, potentially giving it a longer lifespan.

[hyunjimoon]

some more visuals on @juandeoyar's suggestion on "where capital is investment, cost of capital (or returns) are payoffs, and losses is in fact an inverse function of company valuation", "insurance view of capital, cost of capital (expected returns), and losses"

left from efficient frontier for evaluation cld and right from adapting investment risk for startup valuation cld

using artifact

[hyunjimoon]

I presented the product of the above process to two groups of collaborators:

product

collaborator1

amu and doug at MTC: who shared https://pulley.com/ for carta's disruptor and vendr for comparison interface

i prepared proposal using this cld

collaborator2

peter walker at Carta: his talk transcript and I scheduled a chat with him on Mar.14th. below is the process i used to align the vision on desirability.

Peter acknowledged founders should be more aware of financing terms but questioned how much control they have, arguing terms are dictated by market leverage rather than founder preference.
Angie proposed that founder leverage could be quantified through measurable outcomes and suggested probabilistic programming tools may be able to help founders create and validate alternative term structures. Both parties showed interest in continuing the conversation, scheduling a virtual meeting on March 14th to discuss the technical feasibility and market opportunity of Angie's approach.

using the above prompt, interactive 🛠️tools in option pool size and vesting schedule .

assuming i have ten venture's cap tables with safe choices, codes providing some distributional information on founder's equity.

for the next one week, i'll prepare quick demo that can prove technical feasibility so that i can get Peter's help in getting data that can tighten the conditional distributions of the queries (e.g. What is the internal rate of return (IRR) for the VC investor if the company exits at $100 million in 5 years?) using empirical data (e.g. which would be an input to the knowledge (purple) from below). this will enhance operational feasibility of this tool.

[hyunjimoon]

Empowering Founders with Probabilistic Tools

👁️🛠️Visioning & Inventing

I'm building conversational probabilistic reasoning tools that empower entrepreneurs—especially those traditionally underrepresented—to navigate complex financing decisions with the same sophisticated analytical capabilities as experienced investors. This is a vision from paired need (founder's painpoint on complex financial decision) and solution (probabilistic reasoning tools) by Vikash Mansinghka (PI of probabilistic computing lab) and Angie. I have opportunity to develop this further as a term project from Vikash's class that teaches rational AI principles.

👥Relating with Peter at Carta

After our last conversation, I met with Peter from Carta who acknowledged that founders should indeed be more aware of financing terms. However, he questioned how much control they actually have, arguing that terms are largely dictated by market leverage rather than founder preference.

Peter and I agreed that a founder's leverage should affect their financial choices (e.g., selecting between SAFE1 vs SAFE2 options), but we had different perspectives on implementation. While he was skeptical about how precisely this leverage could be inferred from measurable outcomes, he was intrigued by the potential of probabilistic programming to help founders evaluate and validate alternative term structures. Finding this common ground gave us a solid basis for further exploration.

We've scheduled another chat for next Friday (March 14th) where I'll need to demonstrate technical feasibility to convince him to share data that would strengthen my models. Goal of this chat is to get 12 or 27 anonymized cap table data that will help me tighten my conditional distributions that will be used to answer financial value inference questions like "What is the IRR for the VC investor if the company exits at $100M in 5 years in IT industry?". I manually collected three and expect sqrt(2) tigher bound for 30 compared to 15 samples.

👥Competitor analysis for data collaboration

Data quality may not be Carta's advantage. Claude's answer for "What is the IRR for the VC investor if the company exits at $100M in 5 years in IT industry?" may be more accurate, which removes the need to gather data if my sole goal was prediction accuracy of the statistics. However, this beats the purpose for two reasons. 1. This is not rational-by-design AI. 2.given selling probabilistic computing in entrepreneurship scholarship is my goal, reputation of collaborating with service practitioners use matters. Leveraging this potential over-valuation on Carta's data (which I think may be over-valued considering its bias on VC-backed ventures) is my goal.

🙋‍♀️ Key Chat topics

Strategic Demo Selection & Data Requirements

🙋‍♀️1. Which demo would most effectively prove technical feasibility? Should I focus on SAFE choice analysis with Bayesian updating OR continue developing my interactive option pool and vesting schedule explorers?

🙋‍♀️2. What data scope would make this compelling? Is my target of 10 companies' cap table time series + SAFE document choices sufficient to build credible models? How should I approach data collection given privacy concerns?

Example of cap table analysis I can enhance with probabilistic reasoning

Negotiation Strategy & Partnership Opportunities

🙋‍♀️3. If you were me, what stance would you take for data request? I need data to infer my negotiation power/leverage. Your introduction to Pulley was helpful in that I may be able to use this information to position my tools strategically. If you were Peter, what would be factors that makes my proposal impossible to reject?

🙋‍♀️4. How might I leverage your experience with alternative platforms? Given your familiarity with Pulley and Vendr, could we collaborate on identifying integration opportunities or developing features that would meaningfully complement existing tools?

🙋‍♀️5. Are you interested in / help me with parallel exploration? e.g. Would you be willing to contact Pulley to gauge their opinion on this tool while I'm working with Carta?

[hyunjimoon]

choi dongun pointed me to korean version of carta: https://quotabook.com/ and https://zuzu.network/

[hyunjimoon]

statistics on solo founders. more information on dilution

slides used Foundational Equity.pdf with 🗣️mtc sem. carta peter walker.txt