Yudame Research Podcast

Algorithms for Life: Ep. 6, Letting Go

OVERVIEW Gene Kranz didn't care what the hose was designed to do — he cared what it could do. That Apollo 13 moment launches a deep dive into constraint relaxation and strategic randomness, two computer science strategies that explain why the most successful pivots in history all required deliberately breaking the rules. But the same principle that saved Apollo 13 killed 346 people on the Boeing 737 MAX, revealing a critical asymmetry: individuals are systematically too cautious, while institutions are systematically too reckless. WHAT YOU'LL LEARN * How "half a camera" solves impossible problems — the counterintuitive math of LP relaxation and why breaking reality fixes reality * Why people who flip a coin and make a change are 2.2 points happier on a 10-point scale six months later (Levitt, N=22,500) * The jam study is wrong: a 50-experiment meta-analysis found the effect size of choice overload is virtually zero * Why decision fatigue isn't about glucose — the replication crisis busted the willpower-as-fuel model (Hagger 2016, 23 labs) * The asymmetry framework: individuals need to be pushed to let go, institutions need to be forced to hold on KEY TIMESTAMPS * [00:00] - Apollo 13: Gene Kranz's famous order and why it's actually a computer science operation * [02:13] - NP-Hard Problems: Why 100 intersections creates more combinations than atoms in the universe * [04:35] - Voyager, JWST, and Startups: Constraint relaxation patterns from space exploration to Silicon Valley * [06:31] - Simulated Annealing: Why algorithms deliberately go downhill — and why humans refuse to * [09:17] - Satisficing vs. Maximizing: The paradox of richer but more miserable decision-makers * [10:22] - Myth-Busting: The jam study correction and the death of the glucose model of decision fatigue * [12:57] - Boeing 737 MAX: When constraint relaxation becomes lethal * [14:38] - Five Practical Protocols: The calibrated coin flip, constraint inventory, strategic satisficing, engineered serendipity, and environment design RESOURCES RESEARCH PAPERS * Levitt 2020 [https://gwern.net/doc/psychology/cognitive-bias/2020-levitt.pdf] — Coin-flip experiment: making changes leads to +2.2 happiness points at 6 months (N=22,500) * Scheibehenne et al. 2010 — Choice overload meta-analysis: effect size ~0 across 50 experiments * Hagger et al. 2016 — Ego depletion failed to replicate across 23 labs (N=2,000+) * Rajkumar et al. 2022 [https://www.science.org/doi/10.1126/science.abl4476] — LinkedIn weak ties: moderately weak ties maximize job mobility (N=20M) * Schwartz — Maximizers earn 20% more but correlate with regret at r>0.50 FRAMEWORKS & TOOLS * Five Practical Protocols — Calibrated coin flip, constraint inventory, strategic satisficing, engineered serendipity, environment design * Asymmetry Framework — Individuals: relax constraints to grow. Institutions: tighten constraints to protect * Theory of Constraints — Label rules as physics, legal, or self-imposed. Test one self-imposed constraint FURTHER READING * Algorithms to Live By (Christian & Griffiths) — Constraint relaxation, simulated annealing, and randomized algorithms * Quit (Annie Duke 2022) — Strategic quitting framework FULL RESEARCH REPORT Read the complete research synthesis with all citations at: research.yuda.me [https://research.yuda.me/podcast/episodes/algorithms-for-life/ep6-relaxation-randomness/report.md] Try the constraint inventory this week: list the reasons you "can't" do something, then label each one — physics, legal, or self-imposed. Test one.

Algorithms for Life: Ep. 4, How to Communicate

OVERVIEW What can computer networking protocols teach us about human communication? Discover how your brain's 10-bit-per-second bottleneck shapes every conversation, why exponential backoff is "the algorithm of forgiveness" for flaky friends, and the counterintuitive science showing that lossy communication often beats lossless precision. From TCP handshakes hiding in your phone greetings to Walmart's billion-dollar protocol mismatch in Germany, this episode maps the hidden parallels between network engineering and human connection. WHAT YOU'LL LEARN * Why your brain processes only 10 bits per second through a billion-bit firehose — and how to work with this bottleneck instead of against it * How the "algorithm of forgiveness" (exponential backoff) provides a mathematically elegant strategy for dealing with flaky friends * What a billion-dollar protocol mismatch at Walmart Germany teaches about cross-cultural communication failures * Why experts make better decisions using "lossy" gist-based thinking than precise verbatim details (Fuzzy Trace Theory) * How to use the "two-exchange rule" to know exactly when to switch from text to a phone call KEY TIMESTAMPS * [00:00] - The TCP Handshake: Why your "hello" ritual is identical to how computers establish connections * [02:00] - Your 10-Bit Brain: The landmark 2025 study revealing conscious thought processes at just 10 bits per second * [06:20] - Context Switching Costs: Why "quick questions" destroy up to 40% of your productivity * [08:50] - Conversational ACKs: The science of "uh-huh" — how 51% of conversational turns begin with acknowledgments * [11:30] - Walmart's Protocol Tragedy: The billion-dollar lesson in cross-cultural communication mismatch * [13:20] - The Algorithm of Forgiveness: How exponential backoff from ALOHAnet solves the flaky friend problem * [16:00] - Async vs Sync Debate: Healthcare data showing 58.8% time savings clashes with trust-building research * [21:00] - Amazon's API Mandate: How treating teams like network nodes accidentally created AWS * [27:30] - Fuzzy Trace Theory: The biggest surprise — why lossy beats lossless in human cognition * [33:20] - Where the Metaphor Breaks: Why Shannon excluded meaning and what strategic ambiguity teaches us * [36:05] - Three Takeaways: Diagnose the protocol, respect the bottleneck, and don't mistake the map for the territory FULL RESEARCH REPORT Read the complete research synthesis with all citations at: research.yuda.me [https://research.yuda.me/podcast/episodes/algorithms-for-life/ep4-networking/report.md] Try the two-exchange rule this week: the next time you find yourself going back and forth more than twice in an email or Slack thread, stop typing and pick up the phone.

Algorithms for Life: Ep. 5, When to Scout, When to Settle

OVERVIEW A mathematician proposes to the statistically optimal woman — and gets rejected. That story launches a deep dive into the 37% rule, the explore/exploit tradeoff, and multi-armed bandits as frameworks for life's biggest decisions. From the satisficing paradox that makes maximizers richer but more miserable, to Kodak's fatal exploitation trap and Amazon's $170M failure that birthed Alexa, discover why the principle of "explore then commit" matters far more than any magic number. WHAT YOU'LL LEARN * Why the famous 37% rule is almost always wrong — but the principle behind it is pure gold * How maximizers earn $7,500 more but feel significantly worse about their jobs (the satisficing paradox) * What Kodak's suppression of the digital camera and Nokia's culture of fear reveal about the exploitation trap * Why Amazon's $170 million Fire Phone disaster was actually the key to dominating the smart speaker market * How the Malamud study of England vs Scotland proves late specializers win the career marathon KEY TIMESTAMPS * [00:00] - Welcome: The universal anxiety of commitment and missing out * [01:06] - Michael Trick's Story: The operations researcher who optimized his love life * [04:58] - The Secretary Problem: The elegant math behind look-then-leap * [08:17] - When 37% Breaks: How recall and "good enough" goals change everything * [15:05] - The Satisficing Paradox: Why richer maximizers are more miserable * [21:44] - The Exploitation Trap: Kodak, Nokia, Amazon, and organizational ambidexterity * [30:09] - Practical Frameworks: Multi-armed bandits, stopping test, and Plan ABC * [37:58] - Three Takeaways: Principle over number, strategic satisficing, calibrate by domain RESOURCES RESEARCH PAPERS * Iyengar, Wells & Schwartz 2006 — Maximizers secured ~$7,500 more but were significantly less satisfied * Petrucci 1993 — With 50% recall probability, optimal exploration jumps from 37% to 61% * Malamud — England vs Scotland: early specializers switch careers more; late specializers find better fit * O'Reilly & Tushman — Structured exploration units: >90% success vs 0% for permission-based models TOOLS & FRAMEWORKS * Five-Question Stopping Test — Break decision paralysis: Can you define great? Still learning? Met your threshold? Best guess stable? Unresolved mystery box? * Plan ABC (80,000 Hours) [https://80000hours.org/career-planning/] — Structure career risk with exploit mode, pre-set pivot triggers, and a lifeboat plan FURTHER READING * Algorithms to Live By (Christian & Griffiths) — The foundational book on applying computer science to everyday decisions * Range (David Epstein) — The case for broad sampling and late specialization FULL RESEARCH REPORT Read the complete research synthesis with all citations at: research.yuda.me [https://research.yuda.me/podcast/episodes/algorithms-for-life/ep5-optimal-stopping-explore-exploit/report.md] Try the five-question stopping test this week: the next time you're stuck in a decision loop, ask yourself — am I still learning anything new, or am I just stalling?

Algorithms for Life: Ep. 3, How to Delegate

82% of hiring managers admitted they saw the warning signs during interviews—arrogance, defensiveness, rigidity—and hired anyway. Within 18 months, 46% of those new hires had failed. And here is the part that should make every leader uncomfortable: 89% of those failures had nothing to do with technical skill. The failures were almost entirely about attitude—coachability, emotional intelligence, motivation, temperament (Leadership IQ, 3-year study, N=20,000+ across 312 organizations). This finding sits at the heart of a much larger problem with how we think about delegation. The most popular delegation advice in the world rests on an evidence base that is extraordinarily thin. The "70% rule"? Zero empirical validation. Situational Leadership? A 2025 systematic review called it a "fundamental paradox"—widely adopted but lacking strong empirical support. Key findings: * The 89% Attitude Gap: 46% of new hires fail within 18 months. Coachability (26%), emotional intelligence (23%), motivation (17%), and temperament (15%) dwarf technical skill (11%) as failure causes. * Learning Agility Predicts Success: Meta-analysis of 20 field studies: ρ = 0.74 with leader performance. Virtually uncorrelated with IQ (r = 0.09 across 60,000+ participants). The "smartest person in the room" may be the worst person to delegate to. * The 70% Rule Is Folklore: Traced to a single consultant's intuition (Jim Schleckser, Inc.com, 2014). No RCT, no field experiment, no peer-reviewed validation. Works as a psychological trigger—a "satisficing heuristic"—not as a scientific law. * Founder Mode vs. Manager Mode: Brian Chesky took 40-60 direct reports at Airbnb during COVID crisis → $4.1B revenue, 50% EBITDA margin. But Wasserman's data on thousands of startups shows founders who cede control build companies worth 80-100% more. Resolution: it's a gear shift, not an identity. * Cross-Cultural Reversal: Empowerment was associated with negative satisfaction and lower performance in India (Robert et al., 2000). High power-distance cultures performed better when disempowered (Eylon & Au, 1999). Norwegian retail chain saw 5.6% sales increase when autonomy was reduced (Gjedrem & Rege). * The Evidence Void: 66-90% of leadership studies fail to address endogeneity (Antonakis et al.). Good to Great: 6/11 companies underperformed S&P 500 by 2012. Circuit City went bankrupt. Five Protocols for Evidence-Based Delegation: 1. Protocol 1 — Hire for Learning Agility: Ask "Tell me about a time you had to learn something completely new." Red flag: vague answers. Green flag: systematic process. Give real-time critical feedback during the interview and watch their response. 26% of failures are coachability—testable in 30 seconds. 2. Protocol 2 — Calibrated 70% Heuristic: Slide the threshold based on reversibility. Low-stakes/reversible tasks: delegate at 50% competence (learning-first). High-stakes/irreversible: wait for 90-95%. Use a graduated autonomy timeline: supervised → checkpoint → periodic review → exception-based. 3. Protocol 3 — OPPTY Framework: Observation → Practice → Partnering → Taking responsibility → You're on your own. 9-12 week cycle for transferring tacit knowledge. Stolen from medical residencies and Toyota Production System. 4. Protocol 4 — Cultural Adaptation: Look up your team's Power Distance Index (Hofstede/GLOBE). High PD: be more directive, provide templates, avoid open-ended questions. Low PD: focus on outcomes, give freedom on methods. The Golden Question: "Would you prefer detailed directions or just the end goal?" 5. Protocol 5 — Selective Founder Mode: Keep the soul (vision, product definition, cultural values, hiring standards). Delegate the body (operations, execution, logistics). Applies to AI delegation too: high performers use founder mode with AI tools, iterating 20 times rather than throwing tasks over the wall. Three Key Takeaways: 1. Who > What: Delegation success is 89% about the person (learning agility, attitude) and only 11% about the task or technical skills. Hire the metabolism, not the resume. 2. Tools, Not Rules: Frameworks like the 70% rule are useful thinking tools and heuristics, not scientific laws. Use them as starting points for judgment. 3. Context Is King: No universal "best way" to delegate. Adapt to culture (power distance), stakes (reversibility), and lifecycle (crisis vs. stability). Full research report: report.md [https://research.yuda.me/podcast/episodes/algorithms-for-life/ep3-how-to-delegate/report.md] Key Sources: * Leadership IQ — 3-year study of 20,000+ new hires across 312 organizations: 46% failure rate, 89% attitudinal * De Meuse et al. — Learning agility meta-analysis: 20 field studies, ρ = 0.74 with leader performance * Wasserman (2008/2012) — Founder's Dilemma: 50% of founders replaced by year 3, Rich vs. King tradeoff * Robert et al. (2000) / Eylon & Au (1999) — Cross-cultural empowerment reversal studies * Antonakis et al. (2010/2014) — 66-90% of leadership studies fail causal standards * Paul Graham, "Founder Mode" (2024) [https://paulgraham.com/foundermode.html] — Sparked Silicon Valley delegation debate * Gjedrem & Rege (2017) — Norwegian retail: reducing autonomy increased sales 5.6% * Blunden & Steffel (2024) — Delegated decisions perceived as burden, not opportunity

Stablecoin Series: Ep. 8, Post-Launch Operations - Stablecoins Are Banks Disguised As Software

Circle, the company behind USDC, pays Coinbase $908 million per year. Not for technology. Not for custody. For distribution. That single line item from Circle's S-1 filing with the SEC tells you more about what running a stablecoin actually looks like than any whitepaper ever could. This is the finale of our 8-part deep dive into stablecoins. We've covered regulation (the GENIUS Act), tokenomics, technical architecture, reserve management, liquidity partnerships, and go-to-market strategy. But we saved the most pragmatic—and frankly most startling—piece for last: what happens after the launch press release. Key findings: * The $908 Million Reality Check: Circle's largest expense is distribution and transaction costs at $1.01 billion annually, with $908 million paid to Coinbase alone. Technology is the easy part. Getting stablecoins into users' hands is where the real cost lies. * Two Viable Operating Models: Tether manages $115B with ~150-235 employees ($93M profit per employee). Circle manages $60B with 815-1,200 employees ($292K average cost per employee). Both work—they represent different strategic bets on automation vs regulatory positioning. * Four Monitoring Layers Operating 24/7: Reserve composition tracking (hourly reconciliation of on-chain vs off-chain), transaction flow surveillance ($27T annual volume), counterparty health assessment (custodians, validators, bridges), and systemic risk detection (stablecoin issuers now hold $127B in U.S. Treasuries—17th largest globally). * Monthly Attestation Gauntlet: GENIUS Act requires monthly verification by independent auditors. Process: snapshot on-chain supply across all chains at midnight UTC, obtain custodian confirmations from BNY Mellon et al., auditors independently verify, publish report within 10-15 business days. * Enforcement Philosophy Determines Org Structure: Tether: high-throughput freeze-burn-reissue model (Sept/Nov 2025 spikes of $25-30M destroyed tokens, continuous blacklist updates). Circle: judicially-anchored model (actions cluster Oct-Nov 2024, Mar-May 2025, each requires legal review, no burn-reissue). * Multi-Chain Operational Burden: Circle supports 28-30 chains, Tether ~14 (after deprecating 5 legacy networks in Sept 2025). Each chain requires: full/archive node, real-time monitoring, multi-sig wallets, gas estimation, chain-specific risk management. Tether deprecated Kusama with only $250K remaining from $3.5M lifetime issuance after 2+ years of decline. * Cross-Chain Transfer Protocol (CCTP): Circle's burn-and-mint model eliminates bridge risk. $110B+ volume, 5.3M+ transfers. Standard transfers: 13-19 minutes (source chain finality). Fast transfers: seconds (with fees). * Operational Cost Structure: For $1B-$5B issuer: Personnel $10-25M, tech infrastructure $1-3M, compliance vendors $100K-$500K, legal/regulatory $500K-$2M, banking/custody $200K-$1M, attestation/audit $200K-$500K. Scaled issuers: $30M-$150M+ annually. * Vendor Ecosystem Lock-In: Compliance (Chainalysis, TRM Labs, Elliptic: $30K-$100K/year), custody (Fireblocks: $200B monthly volume, 120+ chains, SOC2 Type II), payments (Rain.xyz: $3B annualized, Visa Principal Member), nodes (Alchemy/QuickNode: $1K-$30K/month). * Payment Processor Integration (Stripe): Customer redirected to crypto.stripe.com for wallet connection, funds settle in merchant Stripe balance in USD (not stablecoins), no disputes supported, refunds yes, manual capture no. US businesses only, customers worldwide. USDC on Ethereum/Solana/Polygon/Base. * Reserve Yield Profitability: Circle's $1.6B revenue (2024) primarily from 5% Treasury yield on $60B reserves = ~$3B gross income. Tether: $5.7B profit (H1 2025). Profitability depends on Federal Reserve rates—normalization to historical averages significantly impacts revenue. * Compliance vs Growth Trade-Off: Circle spent $263M on personnel (28% engineers, only 4% compliance staff—automation bet). Tether's lean model may reflect reduced compliance investment rather than pure efficiency. GENIUS Act effective Jan 2027 forces all issuers toward Circle-style infrastructure. Three Operational Playbooks: 1. Multi-Chain Expansion Must Be Compliance-Led: If you cannot reliably freeze and enforce on a chain, don't launch there. Deprecation trigger: usage declining 2+ years, ruthlessly cut operational overhead. 2. Choose Enforcement Model From Day One: High-throughput (Tether-style: larger ops team, automated blacklist management, burn-reissue capability) or judicially-anchored (Circle-style: heavier legal review, fewer but procedurally constrained interventions). 3. Vendor Stack Is Non-Optional: KYT/AML platform (Chainalysis class), MPC custody (Fireblocks class if not in-house), case management + audit trail, 24/7 monitoring/observability. Budget for it from day one. The GENIUS Act Reality (Effective January 18, 2027): * Monthly attestations from independent registered public accounting firms * Annual GAAP audits under PCAOB standards for issuers >$50B * Technical capability to freeze, seize, and burn tokens when legally required * Federal or state banking regulator oversight (FDIC, Fed, OCC, state banking departments) * Reserves limited to U.S. Treasuries, repos, cash equivalents—no risky yield strategies * 18-month implementation window from signing (July 2025 to Jan 2027) Why This Matters: The popular belief is that running a stablecoin is primarily a technical challenge—deploy the smart contract, set up mint-burn logic, let the blockchain do the rest. The evidence tells a different story. As Circle's SEC filing reveals, compliance, attestation, distribution partnerships, and banking relationships dominate both cost structure and operational complexity. The issuers who understand that stablecoins are banks disguised as software are building financial institutions of the future. The ones who think it's just software won't survive the January 2027 deadline. This concludes our 8-episode stablecoin series covering: 1. GENIUS Act regulatory framework 2. Tokenomics and design decisions 3. Technical architecture and security 4. Reserve management and attestation 5. Liquidity partnerships and market making 6. Go-to-market strategy and distribution 7. Yield strategies and profitability models 8. Post-launch operations (this episode) Full research report: report.md [https://research.yuda.me/podcast/episodes/stablecoin-series/ep8-post-launch-operations/report.md] Key Sources: * Circle S-1 SEC Filing - $908M Coinbase distribution payment, $263M personnel costs, 815-1,200 employees, $60B circulation, $1.6B 2024 revenue * GENIUS Act (Signed July 18, 2025, Effective January 18, 2027) - Monthly attestations, annual GAAP audits, freeze/seize technical requirements, federal/state oversight * AMLBot 2025 Stablecoin Freezes Data - USDT vs USDC enforcement models, Sept/Nov 2025 burn spikes ($25-30M), clustering patterns Oct-Nov 2024 & Mar-May 2025 * Circle CCTP Documentation - Cross-Chain Transfer Protocol, burn-and-mint architecture, $110B+ cumulative volume, 5.3M+ transfers * Tether Company Announcements - September 2025 deprecation of 5 legacy networks (Omni, BCH SLP, Kusama, EOS, Algorand) * GPT-Researcher Industry Analysis - DataIntelo 2024 compliance platforms market research, operational cost structures, vendor ecosystem breakdown * Perplexity Academic Synthesis - Federal Reserve stablecoin systemic risk research, $127B Treasury holdings (17th largest globally), $27T annual transaction volume * Gemini Deep Research Policy Analysis - GENIUS Act vs MiCA operational differences, regulatory timeline, yield-bearing stablecoin classification debates