🧭🗺️selling entrepreneurial choice/map as Bayes.Entrep

[hyunjimoon]

target audience: Probabilistic Computation evaluators
for BE evaluators, i summarized Probabilistic Computation in sister thread #224

---

when and how it started: 2022, bayesian entrepreneurship conference which ~ three schools of thought has prepared

• white paper.pdf

• referenced papers in white paper

• human summary of conference

• compilation of all of the above + conference presentations slides with below table of contents

Section	Title	Page
0	Bayesian Entrepreneurship Conference Agenda	1
0	White paper	2
1	BEC_Kickoff Gambardella BEC 2024	57
2	What is Bayesian Entrepreneurship - Stern BEC 2024	63
2a	What is Bayesian Entrepreneurship - Zenger BEC 2024	90
2b	What is Bayesian Entrepreneurship - Novelli BEC 2024	106
3	BEC Priors in Entrepreneurial Decisionmaking - Camuffo BEC 2024	114
3a	BEC Priors in Entrepreneurial Decisionmaking - Astebro BEC 2024	129
3b	BEC Priors in Entrepreneurial Decisionmaking - Cohen BEC 2024	151
4	Entrepreneurial Experimentation - Gambardella BEC 2024	161
4a	Entrepreneurial Experimentation - Krieger BEC 2024	174
4b	Entrepreneurial Experimentation - Kim BEC 2024	200
5	Entrepreneurial Persuasion - Gans BEC 2024	248
5a	Entrepreneurial Persuasion - Nanda BEC 2024	297
6	Entrepreneurial Pedagogy and Practice - Scott BEC 2024	313
6a	Entrepreneurial Pedagogy and Practice - Spina BEC 2024	343
6b	Entrepreneurial Pedagogy and Practice - Coali BEC 2024	362

---

based on this, i organized as below in 2024

1. hierarchical structure

• 1.1 🧢operation caps market understanding (rev = max(ops, market)) 📝1
• 1.2 🪵test2choose1 (amortized learning),
• 1.3 🧬pivoting (spandrel 📝4)

2. 🧭experiment design for hierarchical structure

• 2.1 low vs high bar
• 2.2 bayesian workflow

3. theory-driven/scientific approach/bayesnet with experiment infra

4. 🗺️ecosystem

• entrepreneurship as perception of economic opportunity
• role of thinker and doer ENT as system1 vs 2 for knowledge production chain brain (knowledge chain) 📝5
• unified theory linguistic meaning: statistical and pragmatic

---

building on, i organized as below in 2025

Framework & Components		Individual Choice	Societal Choice	Computational Extensions
1. Strategic Structure
	Operation-Market Fit	Best-foot-forward experiments to validate fit (e.g., Zappos testing rapid delivery model)	Industry standards emerge for minimum viable operations	Bayesian optimization to efficiently search operation-market fit space
	Choice Between Alternatives	Critical tests between strategic options (e.g., Commonwealth Fusion's magnet vs reactor decision)	Dominant designs emerge through collective experimentation	Optimal stopping problems for experiment design
	Strategic Adaptation	Three-lens adaptation (probabilistic/relational/social factors)	Industry-wide learning and adaptation patterns	POMDP modeling for sequential pivot decisions
2. Experiment Design
	Experimental Standards	Low bar (self-learning) vs high bar (external validation) experiments	Development of shared experimentation norms	Multi-agent Bayesian persuasion models
	Learning Process	Theory-driven experimentation with clear success/failure criteria	Collective knowledge accumulation	Bayesian workflow diagnostics (prior/posterior checks)
3. Ecosystem Impact
	Market Formation	Development and testing of contrarian beliefs	New market creation through belief alignment	Equilibrium models with heterogeneous beliefs

[hyunjimoon]

2. experiment design for hier.structure with 2.1 low vs high bar and 2.2 bayesian workflow

• Bayesian Workflow and Startup Operations and gps2navigation
  
• vikash agreed Testing Entrepreneurial Hypothesis and Experiment
  

low and high bar theory is applied to my situation in Data4DM/Tool4Ops4Entrep#21 (comment)

Entrepreneurial experiments and their success measures.

• conducting low BAR experiments in entrepreneurship is important, which are easy to pass and provide valuable insights into what works and what doesn't.

• people often understand how to interpret low BAR experiments differently than high BAR experiments, with low BAR experiments providing affirmation and high BAR experiments providing the truth.

• importance of experimentation in product development, highlighting the value of minimal viable products (MVPs) in reducing false positives and startups.

• the concept of "high BAR" experiments, which require minimal effort and provide valuable insights into customer preferences. (??)

Optimizing experiments for entrepreneurs.

• Entrepreneurs prioritize low-bar experiments over high-bar experiments, despite potential biases.

• Speaker discusses the use of noise in decision-making, with a focus on the Bayesian approach. (??)

Entrepreneurial optimism and experimentation.

• Entrepreneurs choose high bar experiments to convince investors, as they are more optimistic and want to maximize upside potential.

• Minimum Viable experiments are valuable for businesses and investors, as they provide convincing evidence and demonstrate potential for growth.

• Entrepreneurs' optimism influences likelihood of starting a venture, with lower probability of success for those who need to persuade others.

• Pioneer of experiments on unbiased data to gain insights, but challenges remain in interpreting results and understanding data-generating process.

[hyunjimoon]

Scott's 👍 on connecting lowbar-highbar with ops VS market action on Oct. eval(charlie-scott, angie)1,

Aspect	Thinker	Doer	Optimal Balance
Approach	Heavy analysis, perfect information	Quick execution, learning by doing	Adjust based on stage & context
Speed	Slow, thorough	Fast, iterative	Match speed to market pressure
Risk	Lower error rate, higher opportunity cost	Higher error rate, lower opportunity cost	Balance based on reversibility of decisions
Market Understanding	Deep analysis before action	Learn through customer feedback	Only gather info you can operationalize
Operational Focus	Planning & optimization	Building & testing	Must have capacity to execute insights
Resource Cost	High thinking cost, delayed action	High correction cost, quick action	Stop analysis when marginal value < cost
Best For	Complex, irreversible decisions	Fast-moving markets, reversible decisions	Most startups need mix of both
Example	Analyzing all customer segments deeply	Launching MVP to initial segment	Tesla: Started luxury (doable) before mass market
Key Risk	Analysis paralysis	Preventable mistakes	Missing optimal stopping point
Success Rule	"Think until value of new insights < delay cost"	"Act until cost of mistakes > learning value"	"Only gather info you can execute on"

Remember: "Segmenting between 2-person and 4-person families only matters if your operational capacity can serve different segments."

[hyunjimoon]

The left plot from bayesian entrepreneurship paper illustrates how the optimal strategy changes based on the expected value (V) of theory Θ, showing three distinct zones: commit to alternative theory when V is low (V < T⁰), experiment with alternative theory for intermediate values (T⁰ < V < T), and commit to theory Θ when V is high (V > T). The value of experimentation (V^E) increases with the expected value V, while Q^E represents the value of the alternative theory, suggesting that experimentation becomes more valuable as the potential payoff increases.

[hyunjimoon]

example of low -high 🍫 for tesla model 2 and cybertruck

We design a situation where high and low bar experiment is possible in both production strategy and market segment. The tradeoff between higher mean for lowbar but higher sigma for highbar experiments manifests in three dimensions: (1) Market size dynamics - Model 2 targets larger but saturated market vs Cybertruck/Robotaxi's uncertain but potentially larger new segments; (2) Competitive landscape - Model 2 faces intense competition from other EVs and used vehicles leading to lower margins, while Cybertruck/Robotaxi occupy less competitive niches enabling higher margins despite uncertainty; (3) Unit economics - Model 2 offers predictable but lower per-unit returns vs Cybertruck's higher variance but potentially higher per-unit returns due to premium pricing in novel segment. Market expansion during growth...

This updated text better captures the key insights from Tom's discussion about competition's role in experiment choice and the relationship between market maturity and return distributions, while maintaining the original structure about Tesla's evolution.

[hyunjimoon]

todo: reach out to aticus for continued chat on the third (meaning of learning and persistence and choice given our answer of the two)

Aticus<>Angie agenda123.pdf
Aticus __ Angie_learn, persistence, experiment.txt
Adner15_pace_sub_techscurve.pdf

[hyunjimoon]

from scott's seminar summarized in #234 (comment)

Comparison of Zappos vs Commonwealth Fusion Experiments

Aspect	Zappos (Low Bar)	Commonwealth Fusion (High Bar)
Experiment Type	"Best Foot Forward"	"Raising the Bar"
Core Design	Optimized narrow circumstances for success (inventory, customer service)	Single critical technical milestone (20 Tesla magnet)
Purpose	Validate market viability	Validate technical feasibility
Test Cost	Moderate (inventory + operations)	Very high ($100M+ range)
Success Criteria	Easy to pass - just needed decent customer response	Hard to pass - required achieving specific technical threshold
Risk Profile	Low risk of false negatives - if it works here, likely works broadly	Low risk of false positives - if magnet works, fusion likely works
Prior Belief	Founders highly optimistic about online shoe retail	Team confident about fusion but faced skeptical industry
Key Stakeholders	Founders + early investors	MIT + external investors
Audience	VCs (Sequoia specifically)	Broader scientific/investment community
Time Horizon	6-12 months	Multiple years
Exit Decision	Clear exit if experiment fails	Success critical for next phase funding
Information Value	Demonstrated market demand	Proved fundamental technical feasibility
Follow-on Impact	Led to major VC funding	Would enable billion-dollar next phase
Strategic Choice	Chose easiest path to show viability	Chose hardest but most definitive test
Learning Type	Market learning through real customers	Technical learning through R&D
Organizational Impact	Built culture around customer service	Required MIT institutional support

[hyunjimoon]

1. hierarchical structure: 1.1 🧢operation caps market understanding (rev = max(ops, market)), 1.2 🪵test2choose1 (amortized learning), 1.3 🧬pivoting (spandrel)

pivoting started from lean startup movement

Crilly18_fix_pivot.pdf

Kirtley20_pivot_entrep.pdf
Chen22_Programs of Experimentation and Pivoting for (Overconfident) Entrepreneurs.pdf

two papers by Steven Yoo

yoo-et-al-2021-a-theoretical-analysis-of-the-lean-start-up-method.pdf

yoo16_opt-time-alloc-process-improvement-for-growth-focused-entrepre.pdf

scott recommended (detail)

Key Distinction	🧠 Theory-Based Search	🤜 Practice-Based Search
1. Guiding Principle	Guided by formal hypotheses or conceptual frameworks (theories)	Guided by on-the-ground actions, routines, and emergent adaptations
2. Primary Search Approach	Structured, hypothesis-driven tests across multiple domains or markets	Iterative, trial-and-error learning driven by local insights, tacit knowledge, and incremental tweaks
3. Role of Large Sample or Data	Tends to rely on broader, more systematic data (e.g., “big” samples such as Amazon’s)	Often works with smaller, situation-specific data; draws on direct practice in local contexts
4. How Learning Updates Occur	Learns globally via formal updating (akin to hierarchical Bayesian: pooling across many)	Learns locally; each new venture or action is treated as a distinct experiment with limited pooling
5. Overfitting/Underfitting Trade-off	More emphasis on generalizable findings; risk is missing local nuances (underfitting)	More emphasis on local fit, but risk of being siloed or ignoring global patterns (overfitting)
6. Example Contexts	Large platforms systematically testing new ideas (e.g., A/B testing at scale)	Small-scale entrepreneurs iterating prototypes and pivoting based on direct feedback
7. Relation to Hierarchical Modeling	Naturally maps onto multilevel models: each idea/venture is partially pooled via a global theory	May not explicitly pool across contexts; tends to treat each action as unique unless aggregated later

This table encapsulates how Theory-Based Entrepreneurial Search (left column) naturally aligns with the hierarchical Bayesian perspective from “Models With Memory,” pooling evidence systematically from many contexts. In contrast, Practice-Based Search (right column) learns from smaller, direct experiments and is less likely to use an aggregate theory across multiple data clusters—though it can still benefit from partial-pooling ideas when insights from different local experiments must eventually be combined.

erin recommended josh kreiger's bec talk image

[hyunjimoon]

M3S Poster Angie Charlie .pdf is the poster Charlie will present at MIT-Singapore AI workshop this July. Please don't distribute yet!

[hyunjimoon]

#243 is prototype that competes human and near-optimal bayes - it's goal is to

1. show belief update visually
2. enhance human's belief on machine though improved thought product

[hyunjimoon]

entrepreneurial choice with freedom (starbuck-peets), constraint (body shop), uncertainty (amazon on why it chose books to test selling something online), noisy learning (third love)

strategy2operations for tesla roadster
cladue template: elon's pitch to investors

[hyunjimoon]

jeff dotson

using bayesian diversification of corporate diversification

whether the paper's findings demonstrate managers' ability to make value-maximizing diversification choices ex-ante, or whether the apparent rationality in the data is primarily a result of ex-post selection effects:

Does your statistical evidence reflect managers making good diversification choices in real time, or just show the survivors who made choices that worked out?
How do you know it's managers predicting value accurately versus just seeing data from the "winners" who survived their diversification choices?
Can your Bayesian methodology distinguish between good decision-making and selection bias from seeing only firms that succeeded with their chosen strategy?

[hyunjimoon]

Angie connected 🧠 Theory-Based Search who keeps learning with Partial Pooling. however both have limitation in evolutionary (exaptation on neural reuse)

Below is a simplified two-column table that aligns Theory-Based Entrepreneurial Search with partial pooling (multilevel/hierarchical Bayesian perspective) in one column, and Practice-Based Entrepreneurial Search with no pooling in the other. This concise layout highlights key differences in how each approach deals with learning, data, and variation.

Key Aspect	🧠 Theory-Based Search + Partial Pooling	🤜 Practice-Based Search + No Pooling
Core Principle	Relies on an overarching theory or conceptual model that adapts to each venture’s data via partial pooling	Relies on action-by-action learning without systematically linking ventures or contexts
Data & Sampling	Aggregates information from multiple ventures or experiments, borrowing strength (shared information)	Treats each venture or trial as mostly isolated; local evidence does not readily inform other ventures
Variability Across Clusters	Recognizes clusters (e.g., different markets, teams) share similarities; uses hierarchical priors to combine info	Assumes clusters are basically independent, making separate inferences that do not “remember” past or external data
Under/Overfitting Trade-off	Balances global vs. local fit by adaptively shrinking extreme or data-scarce clusters toward the overall mean	Tends to overfit local contexts (small data) or else underfit by ignoring any broader pattern altogether
Representative Example	Large platform systematically tests & updates a central theory (partial pooling of results across sub-studies)	Small startup “pivots” frequently, but rarely uses aggregated learnings from past prototypes or markets

• Partial Pooling (Theory-based side) means each cluster/venture’s estimates are informed by a shared population distribution.
• No Pooling (Practice-based side) means each cluster/venture is treated independently, so estimates do not inform one another.

[hyunjimoon]

3. customized tool for theory-driven/scientific approach (current: bayesnet with experiment infra)

@mugamma was looking for bayes-net based causal logic structure

• todd zenger's slide mentions griffiths and tenenbaum's research on causal induction from small samples (i don't understand why todd is framing "bayes net to causal theory" - isn't bayesian workflow (inference, model checking, extension) with bayes net model, what causal model is?)
  
• tomas zellweger explained here smith friston and josh tenenbaum's research to frame strategizing as active inference
  

---

to gain insights on enhancing game environments for pivot-game developed in #243 , I played Bocconi univ.'s theory building games. arnaldo shared his hope to add multi-armed bandit (which was suggested in chat with @mugamma) and graphical belief representation which i feel prob.prog. can contribute (arnaldo said he'd discuss with scott on ways mit and bocconi team can collaborate)

JB who has expertise on porsche helped me play porshce theory building game - @jeanbaptiste may i ask for your input on below four probabilities - reason we couldn't proceed the game was because we further need to set probability on the edge:

---

below is prompt to extract gpt's causal logic from >20 game's situation (i feel supply, product, demand is too limited and am extending to @chasfine 's seven gear model dynamics ((🏢corporate strategy, 💜customer preference, 🧑‍🔬technology&innovation , 👮🏻regulatory policy, 𝍁industry structure, 🔄business cycle, 🏦capital market):

Please analyze the following generative model description and create a causal graph representation:
[Insert generative model description here]
Your task is to:

1. Identify the key premises or assumptions in the model.
2. Determine the main strategic domains or learning modules.
3. Establish the causal relationships between these elements.
4. Create a graph structure showing how these elements connect to form the overall strategy or model.
5. Indicate how these elements contribute to the fitness or effectiveness of the strategy.
6. Present your analysis as a textual description of the graph structure, clearly stating the nodes (elements) and edges (relationships) between them. If possible, also provide a visual representation of the graph, example below
   graph TD
   A[Rent room to make money] --> D[Supply of housing]
   B[Hosts need to have airbeds] --> D
   C[Platform to search for roommates] --> E[Product]
   F[Peer to peer payment] --> E
   G[Conference in town] --> H[Demand for housing]
   I[Need for cheap accommodation] --> H
   D --> J[Fitness of the strategy]
   E --> J
   H --> J
   image

Todd Zenger's "Entrepreneurial learning as seeking to gain confidence in a theory’s usefulness" from his talk on What is Bayesian Entrepreneurship - Zenger BEC 2024.pdf

From BEC in #234 (comment)
image
is from
Eisenmann, T.R., Ries, E., & Dillard, S. (2012). Hypothesis-driven entrepreneurship- The lean startup.pdf

is from Strategic Management Journal - 2024 - Camuffo - A scientific approach to entrepreneurial decision‐making Large‐scale.pdf

experiment infra

Year	Location/Partner	Name/Website	Participants (startups)	Experimental Arms	RQs/Counterfactuals
RCT1 - 2017	Italy – Bocconi	The startup training	116	2	Scientific vs. natural heuristics
RCT2 - 2018	Italy – Bocconi	The startup training	250	2	Scientific vs. natural heuristics
RCT3 - 2019	Italy – Politechnique	The startup lab	210	2	Scientific vs. natural heuristics
RCT4 - 2019	UK – City Bayes	Strategy insight lab	220	2	Scientific vs. natural heuristics
RCT5 - 2020	Italy – Bocconi – 2 Politechniques	Innoventure lab	380	3	Scientific vs. Effectuation vs. natural heuristics
RCT6 - 2020	India – Indian School of Business	Idea to startup lab	380	3	Scientific vs. Effectuation vs. natural heuristics
RCT7 - 2021	Shanghai – Shenzen University	Idea to startup bootcamp	140	3	Scientific vs. Lean vs. no training
RCT8 - 2021	Tanzania – Sugeko – Milan – Bocconi – University of Maryland	Idea Booster Lab	120	3	Scientific vs. Lean vs. no training
RCT9 - 2021	Italy – Bocconi	Idea Booster Lab	240	3	Scientific vs. Lean vs. no training
RCT10 - 2021	Netherlands – RSM	Idea Booster Lab	120	3	Scientific vs. Lean vs. no training
RCT11 - 2022	UK – City Bayes	Idea Booster Lab	120	3	Scientific vs. Lean vs. no training
RCT12 - 2022	Colombia – UniAndes	Idea Booster Lab	160	3	Scientific vs. Lean vs. no training
RCT13 - 2022	Spain – ESADE	Idea Booster Lab	140	3	Scientific vs. Lean vs. no training
RCT14 - 2023	India – Madras Christian College	Idea Booster Lab	130	2	Scientific vs. Lean

[hyunjimoon]

test2choose1
with gpt

[hyunjimoon]

using gpt to capture tradeoff in valuecap and probability of finding investor in due time

[hyunjimoon]

one of the top five ent paper due to its logic of 😲surprise (can be measured by entropy) and ✂️dimension reduction

camuffo24-theory-driven-strategic-management-decisions.pdf

Definition C. 1 (Classical Decision Problem). A decision problem is a quartet $(A, S, C, \xi)$ in which

• $A$ is a collection of available actions
• $S$ is the space of all payoff-relevant contingencies called states of nature
• C is a collection of consequences
• $\xi: A \times S \rightarrow C$ is a consequence function that details the consequence $c=\xi(a, s)$ of action $a$ when state $s \in S$ obtains.

Definition C. 2 (Attributes). An attribute $X_j$ is the set of all the alternative outcomes of one element of a problem. A space of attributes $X=\prod_{j \epsilon]} X_j$ is the space of all the attributes $j \in J$ that DMs believe are relevant for their problem. An attribute $X_j$ and a space of attributes $X$ are maps $D \rightarrow X$ that ✂️reduce the dimensionality of the exploration problem within the decision makers' domain $D$.

Definition C. 3 (Domain). A domain is the set of all attributes $\mathcal{D}=\prod_{j \epsilon /} X_j, J={1,2, \ldots N}$ that decision makers know or could be aware of at any given moment in time given their knowledge, experience, or reference points. It does not necessarily contain $S$. This depends on whether decision makers are aware of all the relevant contingencies for the consequences of their problem. Domains change as decision makers learn.

Definition C. 4 (Model). A model $\theta$ is the realization of a parameter (or vector of parameters) that identifies a specific probability distribution $p_\theta$ within a family of probability distributions $P={p_\theta}$ on the space of attributes.

Definition C. 5 (Theory). A theory is a family of probability distributions $P_{\Theta}={p_\theta}{\theta c \Theta}$ such that the parameter set is a subset $\Theta$ of the set all possible realizations of the parameters $\theta$. We call theory $P{\Theta}$ or $\Theta$ interchangeably.

Definition C. 6 (Experiment). An experiment is a map $f: \Theta$ $\rightarrow \Delta(Y)$ from models $\theta$ to probability distributions $f(y \mid \theta)$, $y \in Y$, where $f(y \mid \theta)$ is the probability of receiving signal $y$ under model $\theta$.

---

[hyunjimoon]

4. entrepreneurship as perception of economic opportunity

1. visuals in their project webpage which i brainstormed how genParse can help it improve Bayesian Entrepreneurship: 🗺️map for startup compass🧭 #159 (comment)

2. "bayesian approach is needed ... favorable election outcome induces more optimistic assessment of opportunity"

3. as can be seen from reference 4,5,6,10, (dynamic) quality improvement is the key which is @chasfine's expertise!

[hyunjimoon]

☝️One and Still: The entrepreneurial strategy is a dynamic process where initial hypotheses about technology and market fit are continuously refined through feedback loops - as entrepreneurs execute (theta) and learn from market responses, they reduce uncertainty about future technological possibilities (sigma_phi), which then becomes the foundation for next iterations.

👴 Portfolio View: Individual entrepreneurs may abandon promising technologies prematurely due to early failures, while institutional investors like Buffett with diversified portfolios can sustain longer experimentation periods necessary for breakthrough innovations, highlighting the tension between individual and social optimal experimentation levels.

🔬🔭Experimental Lens Adjustment: This metaphor frames pivoting as dynamically adjusting your "magnification level" based on experimental data - just as scientists switch microscope lenses to find the right scale of observation, entrepreneurs should flexibly adjust their market focus based on emerging data about supply and demand uncertainties.

⛏️Mining Test quantities: This appears to connect the idea of discovering reusable patterns through systematic testing with industry-wide experimentation - just as individual mining operations contribute to mapping out a geological formation, individual startup experiments contribute to mapping out the possibility space of viable business models.

[hyunjimoon]

Below is the summary table from projecting scott's scott be first seminar transcript.txt to my BE map and compass from #234 and #159.

Framework Components	Individual Decision-Making	Market/Social Impact	Scott's Framework + Computational Extensions
1. Strategic Structure
1.1 Operation-Market Fit	Design of experiments to validate operation-market alignment through automated testing	Formation of industry standards for minimum viable operations	Scott: Best-foot-forward experiments (e.g., Zappos) Extension: Bayesian optimization of operation-market space
1.2 Choice Between Alternatives	Simulation-based comparison of strategic alternatives considering learning costs	Emergence of dominant designs and industry architectures	Scott: Critical tests between alternatives (e.g., Commonwealth Fusion)\nExtension: Optimal stopping in experiment design
1.3 Strategic Adaptation	Three-lens (probabilistic, relational, social) approach to adaptation decisions	Industry-wide learning and adaptation patterns	Scott: Strategic pivots based on market feedback Extension: POMDP modeling of pivot decisions
2. Experiment Design
2.1 Experimental Standards	Choice between self-learning vs external validation experiments	Development of industry experimentation norms and standards	Scott: Low bar (self) vs high bar (external) validation Extension: Multi-agent Bayesian persuasion
2.2 Learning Process	Individual belief updating through systematic experimentation	Collective knowledge accumulation and diffusion	Scott: Theory-driven experimentation Extension: Bayesian workflow diagnostics
3. Ecosystem Impact
3.1 Market Formation	Development and testing of contrarian beliefs	Creation of new markets through belief alignment	Scott: Matching of optimistic entrepreneurs with investors Extension: Equilibrium from heterogeneous beliefs

I used discussion threads from Scott, Andrew, Josh to construct the table:

A. Test2choose1 (discussed with Scott):
while entrepreneurs often face choices between multiple options, the optimal strategy is to automate testing of many scenarios (64) first, then manually test two options before choosing one. This contrasts with the traditional "choose 2, test 1" approach. The automation allows for broader exploration while conserving resources for deep evaluation of the most promising alternatives.

B. Three types of bias (discussed with Charlie, Andrew, Josh):
The framework adapts machine learning's inductive bias concepts to entrepreneurship: approximation bias (difference between infinite vs finite function space), statistical bias (infinite vs finite samples), and optimization bias (convex vs nonconvex optimization). This provides a structured way to understand different sources of error in entrepreneurial decision-making and experiment design.

C. Probabilistic, relational, social reasoning (discussed with Charlie, Josh):
This framework analyzes operational choices through three lenses: probabilistic (quantitative uncertainty), relational (network/system effects), and social (human/organizational factors). Each operational decision (facilities, capacity, integration, etc.) can be evaluated through these three perspectives, providing a comprehensive approach to understanding entrepreneurial choices.

D. Low vs high bar experiment (discussed with Charlie, Scott):
Entrepreneurs design different experiments based on their purpose: low bar experiments are easier to pass and used for self-validation, while high bar experiments are more rigorous and used to convince external stakeholders. This distinction reflects how experimental design varies based on whether the goal is self-learning versus external persuasion.

E. Three major diagnostics (discussed with Andrew):
The approach adapts Bayesian workflow diagnostics to entrepreneurship: prior predictive checks (testing initial assumptions), simulation-based calibration (validating learning process), and posterior predictive checks (evaluating updated beliefs). This provides a systematic framework for testing business hypotheses.

[hyunjimoon]

some slides to discuss from scott's slide,
What is Bayesian Entrepreneurship MIT BE Reading Group Session 1 2025 FINAL.pdf

meaning of w vs n - biased?

[hyunjimoon]

need analysis of BE group based on transcripts and slides from this dropbox
BER25_1scott_otter_ai.txt
BER25_2luca_otter_ai.txt
BER25_3amir_otter_ai.txt

tl; dr Reason to collaborate with scott is, he can and wish to answer two WHYs: 1) why has common priors dominated economic theory, 2) why is revisiting this assumption matters for entrepreneurship

---

Needs Analysis for Bayesian Entrepreneurship Reading Group

Need Type	Quotes from Transcripts	Context & Relevant Literature
Understanding Heterogeneous Priors	"Entrepreneurs must have sort of reasonably high subjective beliefs of the value of their opportunity relative to others." (Speaker 1, Transcript 3)	The reading group aims to explore how entrepreneurs and stakeholders can hold different beliefs about the same opportunity, challenging the common prior assumption dominant in economic theory. This connects to Van Den Steen's work on heterogeneous priors mentioned in the "Not Shared Priors" image.
	"The Fred Wilson about Airbnb. Airbnb was a sort of good SaaS model, but..." (Speaker 1, Transcript 3)	References the example of investor Fred Wilson rejecting Airbnb while others saw potential, illustrating genuine disagreement among experts with similar information.
	"Very famously, Airbnb had to go through numerous rejections before get funding...Fred Wilson, who exposed sort of like remembers that the failure to invest in Airbnb with this quote, we could wrap our heads around their mattresses on the living room floors." (Speaker 2, Transcript 3)	Demonstrates the need to understand how entrepreneurs and investors can "agree to disagree" about opportunity value.
Bayesian Learning Through Experimentation	"However, just because I believe in something doesn't mean I can't learn and be just waited from that view... how does the fact that I have subjective belief interplay with how I design and then learn and update from experiments to ultimately launch and scale ventures." (Speaker 1, Transcript 3)	The reading group addresses how entrepreneurs update their beliefs through experimentation, covered specifically in the "Optimal Bayesian Experimentation and Learning" session.
	"There are two interesting cases that they talked about in the first chapter, which is an entrepreneur that's predisposed to run an experiment. In other words, they will go ahead and run an experiment if there was no additional information available. And an entrepreneur that is predisposed not to run [the venture]." (Speaker 1, Transcript 2)	Discusses different entrepreneurial types based on their priors and how this affects their willingness to experiment, connecting to the "Entrepreneurial Decision-Making" session topic.
Rational vs. Irrational Overconfidence	"This is exactly how standard economics has treated disagreement since the second basically we have imposed strong assumptions, assuming a way a big part of the disagreement, and specifically the agreement to disagree." (Speaker 2, Transcript 3)	The reading group challenges the standard economic view that disagreement must stem from irrationality when agents have the same information.
	"Even though everyone is very rational, everyone given information updates in a Bayesian way, starting from the Empire, well, first of all, any agent will expect to do better than others." (Speaker 2, Transcript 3)	Addresses the Van Den Steen model where entrepreneurs can be rationally optimistic about their ventures without being irrational.
	"The sort of literature sort of assumes that they will be overconfident also in that kind of job, overconfidence is almost something which is inherent to the individual. Here we have a situation in which these individuals are not overconfident on average." (Speaker 2, Transcript 3)	Challenges the literature that treats entrepreneurial overconfidence as a bias rather than a rational response to heterogeneous priors.
Entrepreneurial Finance and Stage Financing	"But in the effectuation literature, you've got subjective prior so on commercial allowed to be optimistic and, you know, have strong beliefs in your ideas, and others are allowed to, you know, disagree with them and agree to disagree, essentially." (Speaker 1, Transcript 2)	Connects to how different theoretical approaches handle subjective beliefs in entrepreneurial finance.
	"This gives rise to the need for stage financing that you get to fund the startup in stages in order to reveal information about the viability of the venture." (Speaker 1, Transcript 2)	Discusses how investors manage uncertainty through stage financing, connecting to the "Structural Models" session topic.
	"Yeah, and they will update. Of course, the demand for experimentation will depend a little on their prior. So they will update in a in a manner which is optimal given their prior, but they will still update." (Speaker 2, Transcript 3)	Shows how entrepreneurs' priors influence their experimentation approach, which affects financing decisions.
Need for Empirical Studies on Subjective Beliefs	"I think it's important, the point of this conversation is really identifying the opportunities where we can bring in that heterogene head. Header, heterogeneous subjective priors." (Speaker 5, Transcript 3)	Identifies research opportunities for empirically studying heterogeneous priors.
	"My sense is that really, like, up until now, like we've managed to measure beliefs basically through surveys at a sort of like micro level, not only micro level, but like at a nothing salient situation, right?" (Speaker 2, Transcript 3)	Points to methodological challenges in measuring beliefs empirically, which the reading group could address.
	"The reason this hasn't been tested as good as well as it could has more to do with precisely measurement issues." (Speaker 2, Transcript 3)	Highlights the measurement challenges that have limited empirical work on heterogeneous priors.
Interdisciplinary Theoretical Foundations	"There are, like, papers with 1000s in citations in strategy, which may, like, don't take this into account, which basically make the point, like entrepreneurs are privately informed about opportunities." (Speaker 2, Transcript 3)	Identifies gaps in the strategy literature that could be addressed through Bayesian approaches.
	"The whole literature was this whole theory was developed in a time which information economics was gross me, and a great deal of bank good things came out of that literature, and in the words of AU man, sort of assuming agreement to disagree away helped us to zero in on purely information on issues." (Speaker 2, Transcript 3)	Provides historical context for why common priors dominated economic theory and why revisiting this assumption matters for entrepreneurship.
	"I think that the sort of overconfident entrepreneur, sort of literature has not, has not given much justice to [updating]." (Speaker 2, Transcript 3)	Points to limitations in existing literature on entrepreneurial overconfidence that a Bayesian approach could address.

[hyunjimoon]

prep chat with amir

1. seven reading list from chat with rahul
2. has structural model been taken? wish to do with paul

• Angie Moon and Rahul are planning an entrepreneurship seminar that aims to bridge concepts from cognitive science (particularly Bayesian approaches) with entrepreneurship theory.

• They discuss how to structure the session, focusing on mental models, resource rationality, and hierarchical Bayesian approaches that could be applied to entrepreneurial decision-making.

• They select seven papers for the reading list that balance accessibility with relevance to the seminar's goals of connecting Bayesian cognitive science concepts to entrepreneurship.

The seven reading list papers they selected:

1. A paper on information acquisition/experimentation (the "information like experiment one")
2. A paper on abstraction (the "Science" version rather than the longer "Psych Review" version)
3. A paper on resource rational decision making
4. A paper on Bayesian polarization (by Kemp)
5. Chapter 14 from a textbook (about inverse planning or "action understanding as inverse planning")
6. A developmental psychology paper on children as scientists (which they connect to entrepreneurs as scientists)
7. A resource rationality paper (likely the Griffiths paper they discussed at the end)

using ent seminar blending cog.sci/stats and BE cld