Souk's Notes - Unofficial Summaries

2026-10-16 · open the original ↗

• The bull market is intact despite weeks of headline-driven chop. Since mid-August the S&P 500 has repeatedly dropped a few percent on news, then reversed to new local highs, building higher lows and higher highs in a volatility contraction pattern (VCP). The author sees no blue-chip crash and no break of structure, and dismisses insider activity and Middle East conflict as irrelevant to investing.
• Buying dips is a bad habit; hold built positions instead. The author argues dip-buying gets rewarded a few times, which reinforces it, and then crushes the buyer when a real bear market arrives. Their stated approach is to have already built positions and wait for the next leg up rather than add on pullbacks.
• Pokemon cards may turn down before equities, as NFTs did. The hypothesis rests on the 2021 precedent: risk-on NFTs (CryptoPunks peaking October 2021, Axie Infinity around November) crashed roughly three months before traditional markets. The author's "Charizard Index" tracks 1,000 popular, liquid cards, deliberately excluding the rarest, illiquid ones as outliers. This is their own speculation, not an established indicator.
• Card buyers look like late-stage bubble investors, not early ones. People buying at absurd prices for retirement or kids' college funds mirror NFT and $100k+ Bitcoin buyers who believed they were early. The author calls the market "clearly in bubble territory," contrasting newcomers with Gary Haase, who began collecting in 1999 and holds a Guinness-recorded $10M collection.
• The index has not broken, so no bear signal yet. It wobbled over the past month but resumed its bull trend, which the author reads as pointing to continuation in stocks. The trigger they are watching for is a steep single-week collapse breaking structure, with news of scalpers and collectors panicking, implying equities follow lower in roughly 2–3 months.

2026-10-16 · open the original ↗

• A model trained on teacher probabilities converges to the teacher, not the truth. In the author's paper with Denis Burakov (SSRN), exact OLS risk under teacher labels is teacher bias + teacher noise × d/(n−d−1): the noise term shrinks with data but the bias is a floor, so teacher labels win at small n and real outcomes win beyond a crossover (about 19 observations in their worked example). They conclude the optimal weight on teacher labels in a blend goes to zero as data grows. Scoring the student against the teacher cannot reveal the teacher's errors; only realized outcomes can, so they recommend Brier and log loss on outcomes, compared against conventional models fitted to the same outcomes and to the same teacher probabilities.
• A frozen pretrained representation pays off only below some sample size. Projecting to r of d dimensions cuts variance but permanently loses signal outside the retained directions ("representation loss"); compression beats the full model only when that loss is below σ²(d−r)/(n−d−1), a threshold that falls toward zero as n grows (crossover near n = 39 in their example). Combined with teacher labels, only the representable part of the teacher's bias appears as bias, while the rest turns into noise-like error in the fitted head — harder to spot, not smaller — and the components don't add neatly (at n = 27, best mixed model risk 0.207 compressed against 0.201 full).
• Temperature calibration fixes a score's scale, not its direction. In a Gaussian probit model, a teacher with the right direction but wrong magnitude is exactly repaired by temperature scaling; a teacher pointing in a wrong direction still gets a perfect reliability diagram after recalibration but stays wrong (a 60° rotation: excess Brier 0.086, only cut to 0.063 by exact recalibration). The author's point for validators: a calibrated model can still rank cases poorly, which only discrimination against outcomes or comparison with an outcome-fitted model will flag; under a logistic link temperature scaling is an approximation, but the direction conclusion holds.
• Candidate-scoring models must give one event one probability across menus. Softmax enforces summing to one within a menu, not agreement across menus: if "30+ days past due" gets 40% but its split into "30–59" and "60+" sums to 55%, the model is incoherent. With independent per-candidate scoring (Luce's choice axiom, which the author says matches how TypeSafe's Jev describes high-cardinality scoring), coherence holds exactly when a combined event's score is the log-sum-exp of its parts; with shared-context scoring there is no guarantee, so the author says to test the same event under different candidate sets, orders and wordings.
• The exact formulas are optimistic for heavy-tailed data at small samples. Rerunning with a multivariate t (5 degrees of freedom), simulated risk was 35% above the formula at n = 12 and about 7% above at n = 95 (uniform inputs came in slightly below); the author notes credit, fraud and AML data are usually heavy-tailed. Stated limits: linear heads on frozen backbones only (not LoRA), a teacher not trained on the student's sample, and population-level calibration results; and a model that predicts which option an annotator or past policy would pick has not been shown to pick the best outcome, which needs a causal model of outcomes given actions.

2026-10-16 · open the original ↗

• Student AI use is now near-universal. The host cites a survey last year by Chegg finding 80% of undergraduates in wealthy countries using AI in their studies, and more recent polls in Britain and Germany putting it at 93–94% of students; he says rigorous evidence on the effect on learning has only recently arrived.
• AI raised Chinese students' homework scores while their exam scores fell. A Stockholm University and University of Hong Kong study published in June, using a difference-in-differences design on grades 7–12 (about 80% adopted chatbots such as DeepSeek or Doubao), found homework scores up 18% and time per assignment down about 30% (64 to 45 minutes), but monthly exam scores down 20% within 6 months and entrance-exam scores down 18% and 24%, with the full penalty emerging after about 2 years. Homework scores flipped from predicting exam success to predicting worse exams; losses were largest in social sciences, then STEM, then languages, and for junior students, high achievers and boys.
• The losses came from homework outsourcing, not AI use as such. In the same study the losses were concentrated among the roughly 80% of users with very short completion times and high homework scores; users spending 50–65 minutes, about the same as non-users, had exam scores effectively identical to non-users. The host argues this is why he would not rush to ban AI in schools as New York City recently did.
• Tool design decides whether AI helps or harms learning. In a Wharton-led field experiment with nearly 1,000 Turkish high-school math students, standard GPT-4 lifted practice performance almost 50% and a safeguarded "GPT Tutor" (prompts over 500 words versus about 50) lifted it 127%, but on the unaided exam the standard group scored 17% worse than no-AI controls while the tutor group was statistically indistinguishable; AI users also overrated how much they had learned. The host also cites a Middlebury College experiment (immediate scores up 0.27 standard deviations, persisting a week, with gains larger for students who had AI explain concepts than for those who had it generate text), a Harvard physics study of nearly 200 students learning more than twice as much from a custom tutor fed the lecturer's solutions, and a World Bank trial in Nigeria where a 6-week Copilot/GPT-4 program raised scores 0.31 standard deviations — a result its authors attribute to the whole package, not the AI alone.
• Watch time on task and weight unaided assessment more. The host's own takeaways: a student whose homework quality jumps while time collapses is probably outsourcing; turn on platforms' study or learning modes; and closed-book, in-class assessments matter more now because homework grades no longer show what a student knows. He mentions oralexam.ai, co-created by his Tulane co-instructor Professor John Renne, and says the same trade-off applies to adults at work: try the problem first and use AI to augment thinking.

2026-10-16 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-16 · open the original ↗

• High valuations and profits blind management to their own weaknesses. The guest, who is not named in the clip, argues that valuation and profitability in themselves shape how a company's people behave. The example given is Nike: very high valuation and profitability bred a corporate culture "totally detached from the customer."
• Nike's headquarters showed the detachment that leads to poor decisions. The guest describes visiting the Portland, Oregon HQ about five years ago and finding "a playground for adult children," with staff playing basketball. They tie that culture to running a business priced at 25 times earnings with very high returns, and argue this is what leads to poor decisions.
• The guest's firm is skeptical of high valuations by design. In the guest's view, a high valuation does damage in two ways. Capital cycle theory says it attracts competition, and the people inside the company also "drink their own Kool-Aid" and make mistakes on capital allocation and pricing. The conclusion is that investors have to be very careful buying into high valuations.

2026-10-15 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-15 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-15 · open the original ↗

• Ukraine has resumed drone strikes on Russian refineries. Overnight into October 15th, drones hit Rosneft's Ryazan refinery, about 300 km from Moscow, setting it on fire; the host says the plant makes about 5% of Russia's fuel, much of it for Moscow. Zelensky told the Financial Times the three-day pause is over because Russia never stopped hitting Ukrainian power plants — the first Ukrainian strike on Russian energy since President Trump announced an energy ceasefire on Sunday, which the host says neither side confirmed.
• Hormuz attacks now sink container ships, not just tankers. The Singapore-flagged GFS Pride was hit near Oman on Tuesday and sank the next day (all 21 crew got off safely); the Singapore-flagged GFS Gizelle was hit Wednesday, damaged on its side, and reached safe water. As of October 15th, the UKMTO count for October stood at around 24 warnings, and OPEC ministers say freight and insurance now cost more than one-third of a cargo's value — a figure the host attributes to OPEC, not himself.
• Washington may let previously sanctioned ships carry non-sanctioned oil. A US official told Reuters the White House is weighing a plan under which such ships could carry "legal oil" — crude and products not subject to US sanctions — with companies applying to the Treasury, aiming to move barrels out of the Persian Gulf and lower freight costs. The host stresses it is not a pass for Russian or Iranian barrels; oil fell about 1.5–2% on the news on the morning of October 15th.
• OPEC cut this year's oil-demand growth forecast again. As reported October 15th, OPEC now sees world demand this year growing by only 180,000 bpd, down from 380,000, while raising next year's number (the captions garble OPEC's stated rationale). The IEA reports Russia pumped 65,000 bpd more in September, reaching 8.43 million bpd — still 870,000 bpd below a year earlier.
• US diesel stays expensive heading into winter. As of October 15th, New York diesel wholesale futures were near $4.93 a gallon and US distillate inventories sat about 11% below the 5-year average. The host closes on a Mizuho analyst's line: "It hasn't even gotten cold yet."

2026-10-15 · open the original ↗

• Once a takeover is announced, sell the whole position immediately. Answering a viewer already long a target stock, the hosts say to take the win rather than hold to collect the remaining spread, because the deal can break: funding falls through, the buyer walks after reviewing the books, and the stock drops below its pre-announcement price. They point to that day's tape (Nasdaq down 525, S&P down 46) as an example of how fast conditions turn.
• A deal stock is worth its market price, not the offer. Their example: with a $50 takeover price and the stock at $48, the stock is worth $48, since even an 80–90% chance of completion leaves the $2 as a risk premium the hosts argue is not worth chasing for an ordinary holder.
• Risk arbitrage belongs to specialists, and amateurs blow up trying it. One host says they have lost friends in the business who blew out thinking it was easy, and that every deal they have played was sized small and was among the most uncomfortable positions they have held. The host compares waiting out the last few dollars to sitting through the final 20 minutes of a 0DTE strangle.

2026-10-15 · open the original ↗

• This post is a schedule of paid-subscriber staff events. It announces sessions where the Prof G Media staff who work behind the scenes on the podcasts present on their own topics: analyst Dan Chiolan on winners and losers in the autonomous vehicle wars (Friday, October 16, noon ET); senior producer Jennifer Sanchez and research lead Mia Silverio on how the podcasts are made (Tuesday, October 20, 1:30 ET); analyst Kristin O'Donoghue on whether America is losing its scientific edge (November 13, 1pm ET); and senior producer and creative director for video Billy Bennett on using AI (November 20, 1pm ET). All are for paid subscribers only.
• What Washington leaves unfunded now may cost breakthroughs later. O'Donoghue's session is framed around the Golden Goose Awards, which honor federally funded research that looked obscure at first but later led to major advances. Her question is what goes unfunded today that could be changing lives 20 years from now. The post poses this as a question for the session and does not answer it.
• The author dismisses the AI job apocalypse but uses AI in-house. He restates that he has "called bullshit" on predictions of mass AI job losses, while saying AI is part of the Prof G Media workflow. He gives no evidence for either point here.

2026-10-15 · open the original ↗

• A time machine would need energy comparable to a star. One speaker, who says they give a blueprint in their book, describes a gigantic atom smasher converging that energy on a single point to open a gateway through space and time.
• Time-travel paradoxes are resolved by the universe splitting. On the kill-your-parents paradox, the speaker argues time is a river that may have forks, so a traveller to the past would enter a parallel universe and kill people genetically identical to their parents, not the parents themselves. This is offered as what "we think", not as established.

2026-10-15 · open the original ↗

• War-driven energy inflation has started bleeding into non-energy prices. The government reported September consumer prices up 0.6%, lifting annual inflation to 3.7% from 3.4%. He stresses that only half of this was energy, against earlier reports where nearly all of it was. He attributes the housing rise probably to higher interest rates, and expects heating oil to make it worse into winter.
• The producer price index is a flashing warning of consumer increases still coming. PPI rose 1.3% in September, the largest monthly rise since March 2022, and 6.3% year over year. Producer gas prices jumped about 20% and transportation and warehousing about 3.5%. He calls it a warning light, not a guarantee of what consumers will pay.
• Restoring crude flows fixes nothing without refining and delivery capacity. Citing the IEA's October report, Middle Eastern refined product shipments are running at 1.9 million barrels a day against 4.7 million before the war. He argues the region has surplus crude it cannot refine, and that US production milestones have never meant immunity from world prices.
• The cost shock is reaching consumers on a lag, alongside tariffs. The Fed's Beige Book reports elevated diesel prices passed through supply chains as fuel surcharges, with businesses describing squeezed margins. He had predicted roughly six months of lag; the war is now in its seventh or eighth month. The Fed lists tariffs first, then insurance, which he says compounds rather than excuses the war's effect.
• He expects real incomes to keep shrinking and tightening to continue past the war. Private workers' average hourly earnings after inflation are 0.7% lower than a year earlier. He expects further rate increases, before or after the election, and tight diesel and gasoline markets into next year as inventories are rebuilt. He calls on decision-makers to say how shipping becomes reliable and how the war ends.

2026-10-15 · open the original ↗

• Single-name SRT imports pool pricing onto a binary bet. Synthetic risk transfer's economics, including the regulatory spread that makes investors willing buyers, came from granular pools of 200 to 2,000 loans with smooth losses. A single data-centre SPV loan with one tenant pays zero for years and then may lose the whole tranche in one event. The author argues this mismatch, not the instrument, is the risk.
• Banks use SRT because these loans are too big to move. They are too large to syndicate, too relationship-critical to sell at a discount, and breach single-borrower limits. Reported: a US$38bn Oracle-leased construction package sat with its arrangers for over six months, Oracle-leased loans are quoted at 89–91 cents, and JPMorgan, Morgan Stanley, SMBC and MUFG are exploring SRT variants (FT, May 2026).
• The SRT is a licence to originate, not a hedge. In the illustrative deal, a bank holding US$4bn of a US$20bn loan sells the 2–20% tranche (US$720m) as a funded credit-linked note at about SOFR+10%, ~14% all-in. Equity needed falls from ~US$420m to ~US$76m, freeing ~US$344m and the limit for the next syndication. Morgan Stanley's ~US$1.5tn data-centre funding gap through 2028 cannot otherwise be warehoused.
• Investors are paid for capital relief, and recovery is the real variable. The bank pays more than the economic cost of the risk because buyers aren't bound by bank risk weights. In the sketch, 3% cumulative default times ~US$650m loss gives ~US$20m expected loss against ~US$505m of coupon. But a 30% loss on default wipes out all US$720m, and 50–70% back-leverage pushes equity returns toward 20%+.
• It breaks at circularity, the bank's call, and weaker tenants. Risk re-enters banks through their lending to the non-bank SRT buyers. If the fund won't roll at the call, which coincides with lease renegotiation and chip residual values, risk-weighted assets snap back to US$4bn. Neocloud-leased paper (7.0–9.9% vs 5.7–7.9% for hyperscaler) would need a thicker tranche; the author projects that is why no such deal has been reported closed. Whether any data-centre SRT has closed at all is unverified.

2026-10-15 · open the original ↗

• SRTs on unplaceable data-centre debt mark a synthetic-CDO-bubble stage. That is his framing of a data point that JPMorgan and Morgan Stanley issued synthetic risk transfers on US$38bn of Oracle-related debt they could not place. How much was sold is undisclosed; he assumes about US$20bn is still held. Selling at ~90 cents would crystallise the loss and anger the client, so banks keep the loan and sell the credit risk.
• SRTs were built for transparent pools, not one loan. After the financial crisis, mortgage pools were modelled loan by loan, so losses were statistically predictable. A single construction loan to an AI data-centre SPV with one tenant keeps the legal form but has no statistical foundation. Losses are binary, so the sold tranche must run 2–20% versus 0–8% or 2–10% on a pool.
• The bank's early call is where GPU values get tested. In the worked example an investor pays US$720m upfront, held in Treasuries or cash, for about 14% a year (~US$101m) over five years, absorbing losses in the 2–20% band. The bank can call at year three, when refinancing is due. If GPUs carried at 75% of value turn out worth 25%, he says, "we got a big problem."
• Banks are buying capital relief, not credit relief. On a US$4bn holding, required common equity falls from about US$420m to US$76m, releasing US$344m. He multiplies that by roughly 15× bank leverage to show the lending capacity freed. He calls it American financial engineering to bail out the banks and leaves open whether it works.

2026-10-15 · open the original ↗

• Basic research must be curiosity-driven, or it misses what matters. Brassard argues practical innovation is the end of a pipeline that starts with research done only to understand nature, citing relativity and quantum theory, without which there would be no electronic computers, and number theory, long called "the queen of mathematics" because it was beautiful and useless until RSA and Diffie-Hellman in the mid-1970s. Bennett adds that if you aim research at what you want to build you will probably miss what turns out to be important. Brassard says their own biggest results, quantum cryptography and teleportation, came by serendipity, from physics meeting computer science after the two met at a 1979 conference in Puerto Rico.
• Quantum key distribution is already deployed at scale in China. Brassard says China has a 10,000-kilometre quantum-cryptography backbone integrated with satellites and, last he heard, 5 million users, and that receiving stations which once took months to build now fit in a truck and deploy in hours. He dates the first satellite experiment to 2017 ("I think") and notes he and Bennett built the first prototype about 5 years after inventing the protocol and 10 years after they met.
• Post-quantum cryptography rests on belief, not proof. Brassard distinguishes quantum cryptography, which he and Bennett invented in 1984 ("83 actually") and which is secure by physics if quantum theory is correct and the apparatus is built correctly, from post-quantum cryptography, classical schemes he calls "wishful thinking" that could fall even to classical computers. Bennett pushes back that there is not "absolutely no reason" to trust them if one-way functions exist, notes all classical cryptography except the one-time pad is provably insecure but may be practically secure, and says quantum cryptography cannot easily replace digital signatures or public-key exchange between strangers.
• Harvest-now-decrypt-later means the reason to act is already here. Brassard argues adversaries can store today's RSA- and Diffie-Hellman-encrypted traffic, which covers almost everything since the browser padlock appeared, and decrypt it all retroactively once they have a quantum computer, with nothing anyone can do to save the past. Switching to post-quantum schemes would take years and guarantees nothing; quantum cryptography's real advantage, in his view, is that an eavesdropper must exploit a flaw during transmission or not at all. He will not predict when a quantum computer powerful enough to break cryptography arrives, says one could already exist unknown, and cites unbelievably fast recent progress. The practical advice from both guests is not to say anything online you would regret if it leaked.
• Treat AI like Merlin: untrustworthy, but able to convince you. Bennett uses interactive proofs, where a limited but honest Arthur questions an all-knowing but untrustworthy Merlin, and a result proven about six years ago that two entangled quantum provers can convince Arthur of the halting of any Turing machine. Because the provers could prove a counterexample to Goldbach's conjecture if one exists but could not convince Arthur that none does, he argues we should not trust what AI says but should learn to question it so that it either tells the truth or gets caught lying; machines take the less important part of the job, and ours is to be "good Arthurs".

2026-10-15 · open the original ↗

• Lump-sum investing beats easing in, and scaling into trades backfires psychologically. The hosts cite Vanguard's finding that investing at once beat spreading over three months about two-thirds of the time, and say their own think-tank research matched it, citing positive drift and how hard timing is. Pressed by a viewer, Sosnoff concedes he does add on down moves, buying Robin Hood at 101 and more near 95, but says that is not dollar-cost averaging.
• Conviction only pays if the position is sized to survive. Sosnoff's long bonds / short Nasdaq trade hurt for weeks, then paid in days: bonds ran from about 102 to 106 after a hot CPI, and the Nasdaq fell roughly 5.5% in six or seven sessions. He covered his short S&P and Nasdaq futures overnight because he won't stay short into expiration Thursday. His forecast is that the selloff is short-term overdone, but any rally should be sold.
• Take the money on an all-cash takeover; leave the spread to arbitrageurs. The hosts say a stock bid $48 against a $50 deal price is worth $48, because the $2 premium pays for deal-break risk, and say they have seen friends blow out on risk arbitrage. Scott Sheridan adds the general test: if you would not open the position at this price, why are you holding it?
• AI buildout asks communities to bear costs without sharing upside. Scott frames the spending as 3.6% of a $32 trillion-plus GDP, roughly 50% above the railroad peak, and cites Meta's Hyperion vehicle: about $27 billion of debt against nearly $30 billion of assets, a coverage ratio of about 1.15, with one analyst saying a 10% cash-flow drop wipes out the equity. Tony Battista wants some environmental checks but would rather the US than its enemies lead.
• A risk reversal is a doubled-up directional bet, not free upside. In the worked example, with the stock at $100 and 60 days to go, you sell the 95 put for $3 and buy the 105 call for $2. That keeps $1 between the strikes, loses dollar-for-dollar below a 94 breakeven, and opens at roughly 70 deltas. The hosts call it "the stupid trade" and warn that the short put requires margin.

2026-10-15 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-15 · open the original ↗

• Shrinking invested capital is the return lever most companies avoid. Davidson frames return on capital as profits over invested capital: most firms chase the numerator with new products and growth, while cutting the denominator means laying off staff, closing factories and exiting countries, which carries heavy social friction.
• Japan is running that denominator cut economy-wide, and it has far to go. Davidson describes a country-level capital cycle with broad cultural buy-in to unwinding overcapitalized balance sheets — cross-holdings, idle cash, duplicate head offices, unused plant — through buybacks at a discount and sales of non-core assets. He reports corporate Japan's returns on capital have been rising for several years and, in his view, the trend has "a long way to run," which is his case for investing there.

2026-10-15 · open the original ↗

• Basic research must be curiosity-driven, or it misses what matters. Brassard argues practical innovation is the end of a pipeline that starts with research done only to understand nature, citing relativity and quantum theory, without which there would be no electronic computers, and number theory, long called "the queen of mathematics" because it was beautiful and useless until RSA and Diffie-Hellman in the mid-1970s. Bennett adds that if you aim research at what you want to build you will probably miss what turns out to be important. Brassard says their own biggest results, quantum cryptography and teleportation, came by serendipity, from physics meeting computer science after the two met at a 1979 conference in Puerto Rico.
• Quantum key distribution is already deployed at scale in China. Brassard says China has a 10,000-kilometre quantum-cryptography backbone integrated with satellites and, last he heard, 5 million users, and that receiving stations which once took months to build now fit in a truck and deploy in hours. He dates the first satellite experiment to 2017 ("I think") and notes he and Bennett built the first prototype about 5 years after inventing the protocol and 10 years after they met.
• Post-quantum cryptography rests on belief, not proof. Brassard distinguishes quantum cryptography, which he and Bennett invented in 1984 ("83 actually") and which is secure by physics if quantum theory is correct and the apparatus is built correctly, from post-quantum cryptography, classical schemes he calls "wishful thinking" that could fall even to classical computers. Bennett pushes back that there is not "absolutely no reason" to trust them if one-way functions exist, notes all classical cryptography except the one-time pad is provably insecure but may be practically secure, and says quantum cryptography cannot easily replace digital signatures or public-key exchange between strangers.
• Harvest-now-decrypt-later means the reason to act is already here. Brassard argues adversaries can store today's RSA- and Diffie-Hellman-encrypted traffic, which covers almost everything since the browser padlock appeared, and decrypt it all retroactively once they have a quantum computer, with nothing anyone can do to save the past. Switching to post-quantum schemes would take years and guarantees nothing; quantum cryptography's real advantage, in his view, is that an eavesdropper must exploit a flaw during transmission or not at all. He will not predict when a quantum computer powerful enough to break cryptography arrives, says one could already exist unknown, and cites unbelievably fast recent progress. The practical advice from both guests is not to say anything online you would regret if it leaked.
• Treat AI like Merlin: untrustworthy, but able to convince you. Bennett uses interactive proofs, where a limited but honest Arthur questions an all-knowing but untrustworthy Merlin, and a result proven about six years ago that two entangled quantum provers can convince Arthur of the halting of any Turing machine. Because the provers could prove a counterexample to Goldbach's conjecture if one exists but could not convince Arthur that none does, he argues we should not trust what AI says but should learn to question it so that it either tells the truth or gets caught lying; machines take the less important part of the job, and ours is to be "good Arthurs".

2026-10-15 · open the original ↗

• Judge a data agent by the analytics workflow, not its features. Shane frames the live lab around the lab's workflow (goal, question, metrics, analysis, insight, decision, story, closing the loop). He says that in his experience with agentic systems, AI can carry the execution at every stage, but people still own which questions are worth answering, what a metric means locally, the level of rigour a decision needs, and the decision itself.
• The agent's data profiling caught the planted quality problems. Given a synthetic grocery-delivery database (six tables, uploaded to ChatGPT's work mode with the data plugin), the agent reported that many non-cancelled orders did not reconcile. It advised exploratory use with clear metric definitions, not treating the financial fields as reconciled. Shane praises its review gates before reporting.
• The same revenue question gave three different numbers. Asked last month's revenue in four parallel chats across models, three answered 278,993, one 276,941, and two noted that excluding tax and tips would give 242,684. Most did not state a definition and none stopped to ask for one, which is what Shane says he would want. A regional breakdown gave Northeast growth of 12.1% in one chat and 13.1% in another, though the recommendation was the same in both (back the Northeast, investigate the Midwest and Mountain). His lesson is to run the same question several times to expose unreliable numbers.
• Calculations are strong; business judgment and pushback are the weak points. The co-host says that across earlier runs they saw no calculation errors and that the agent handled traps like Simpson's paradox, but that it does not always state its assumptions and gives way when strong-armed. Shane says that in an earlier session he told it a CFO-reported revenue figure and it readily aligned to it. He warns that it can become a tool for stakeholders to slice the data until it agrees with them.
• Use it for metric alignment and directional exploration, not reported numbers. Its dashboards were generic big-number-plus-charts templates, and neither of Shane's would open, so he would keep a backup before a meeting. He would use it to pull numbers for several candidate definitions of a metric ahead of a team alignment meeting, and for directional exploratory analysis. The co-host would use it for drilling into segments and for recurring reports once definitions are codified.

2026-10-15 · open the original ↗

• A working query can still use a definition your team rejects. Shane, opening the context series with a fictional outdoor gear retailer, says an AI analyst will find a monetary table, write SQL that runs and return a revenue number built on the wrong definition. Context engineering supplies the company's definitions, date fields and exclusions, and where a question leaves a choice open the agent should ask rather than guess. He says the explanations also need a structured home with a record of changes, rather than meetings, personal notes and Slack.
• Start from existing work, then trace it upstream. A finance report's query shows the filters and calculation in use, a refund policy gives business rules, a data catalog describes the records, and transformation and application code explain what was removed or recorded in the first place (test orders dropped before the reporting table, whether an order is logged at attempt or completion). Shane warns that reused queries carry filters for their original purpose, such as a single promotional campaign, and says to ask the owner about anything unexplained.
• Describe table grain, and endorse the sources the team trusts. In his example, a $100 order of a $30 shirt, a $20 bottle and a $50 day pack, joined to its three item rows and summed, reports $300. Shane recommends endorsements (reviewed-by labels or gold/silver/bronze tiers), last-updated dates, and named owners who update context when, say, finance changes a definition, so that stale reports aren't weighted equally with current ones.
• Keep broad context always loaded and specific guides retrieved. Shane separates company-wide conventions, which Hex includes in every conversation as workspace context, from a library of guides pulled in only for the relevant question. A guide called "sales notes" gives an agent nothing to choose by, while "revenue and orders" with a description in the words people actually use does. His sample guide confirms the period and gross versus net with the asker, defines net revenue as excluding cancelled, refunded and test orders, counts each order total once, and ties out to the daily revenue summary, which also gives a rubric to evaluate against.
• Too much context in one file is the failure mode. In Q&A Rachel of Hex says organised context with front matter and proper retrieval lets models skip what is irrelevant. Shane says gaps are found iteratively by running analyses, asking why they went wrong and running evals. On whether reading code assumes the code is correct, he says bad data foundations just produce wrong answers faster, and that executives using AI analysts may be the forcing function to fix them.

2026-10-15 · open the original ↗

• Most North Stars are defaults, not decisions. The co-host, presenting, says metrics usually arrive through a top-down mandate, a dashboard tool or another team, and that asking AI for "a North Star for my company" returns a pick without the trade-offs. In "weekly active users" each word is a decision: how often value arrives, which action counts, and who the unit is. A copied metric that moves regardless of what the team does is a vanity metric.
• Derive the metric from the customer promise in five steps. The chain runs promise, moment, action, unit, cadence. A language app promising a daily habit gets daily active learners. For a small-business hiring platform, weekly active users would fall whenever a customer succeeded and stopped logging in, so roles filled, with earlier predictors, fits better. For a tax app, weekly use fails on seasonality, and the right metric depends on which promise it makes: done right with no stress, every dollar owed, done in one sitting, or money handled all year, which would favour monthly engaged households.
• The demo prompt has AI interview first and choose last. In a plain Claude chat, the prompt has Claude interview the user one question at a time and push back on vague answers. It then researches comparable companies, returns four to six candidates with what each incentivises, how it could be gamed and when it would be outgrown, and picks one, saying what would change its mind. For the fictional Tidy Tax (promise "done right", moment a federal return accepted on first e-file, unit the household, value once a year from January to April), Claude cited Intuit, H&R Block, the IRS and Lemonade, gave six candidates, and chose households accepted on the first submission over the default of accepted returns. When told the CEO wanted next-season return households, its adversarial review still disagreed.
• Claude's "what would change my mind" list is the analysis backlog. The conditions it gave were that first-try acceptance does not predict customers coming back, that rejections are outside the company's control, and that the rate is already near its ceiling. The co-host says a data team can test each of these against historical data before adopting the metric.
• Pair a lagging North Star with leading drivers, and revisit it. The co-host says the CEO's retention preference is a lagging indicator, so the North Star should be timely and broken into drivers: households starting a return, accepted, accepted first time, and returning. The co-host lists four reasons to revisit it: the promise changes (a pivot to handling money all year would make the monthly metric right), the goal is met, the team learns to game it, or it stops predicting retention and revenue. Asked about zero-to-one products, the co-host says to treat the metric as a placeholder and not to get attached to it.

2026-10-15 · open the original ↗

• Ground every answer in your own experience, not a generic one. The presenter, working through a lesson built around a fictional analyst, "Priya", contrasts a generic answer to "Why do you want to work at Nextdoor?" ("I love your mission of building community…") with an example answer about a neighbour's missing dog, found within an hour through a Nextdoor post, tied to a measurable product moment: how fast a post gets a real response. Because AI can write the generic answer for anyone, the presenter says you need to write down your own personal and technical stories for it to work from. The caption text stops after about 700 words, just as the presenter introduces three reasons interview prep goes wrong, so the rest of the roughly 55-minute lesson is not in the source.

2026-10-15 · open the original ↗

• Recommendations die after the meeting, not in the analysis. The co-host, presenting, names four traps: the action item with no owner, metric or date; "we should", where shared ownership means none; "let's monitor it" with no baseline or target, so monitoring never ends; and "by Q3", a deadline nobody can put on a calendar.
• The close-the-loop checklist has six fields. They are a single named decision owner (not a team), a success metric close to the change (not NPS for a hotfix), a dated baseline, a specific target (fewer than 200 payment tickets, not "reduce significantly"), a real calendar check-back date, and a fallback of revert, escalate or pivot. Shane adds that if it isn't written down it didn't happen, and that treating decisions as bets with a pre-agreed fallback heads off finger-pointing when one fails.
• The demo's skill turned a root cause into a trackable plan. The co-host uses the lab's free AI analyst repo, whose close-the-loop skill triggers whenever an analysis in Claude Code ends with a recommendation, on fictional e-commerce data from "Novamart". Claude Code found a June 2024 spike of about 2x in payment tickets, traced it to app version 2.3.0, and found no double charges and no blocked transactions. The skill then asked who owns the decision, by when, and who follows up, and wrote a markdown plan with owners, check-back dates, guardrail thresholds, fallbacks and a confidence score. The co-host was on Opus 4.7, and says the human remains accountable for the plan.
• Validate the action by triangulating methods, not just the number. Shane lists the validation layers built into the system: reviewing the SQL and Python, tying numbers to ground truth where it exists, and sanity checks such as values that should never be negative. Triangulation sends several subagents to test the same hypothesis with different methods (regression, propensity score matching, difference-in-differences), and agreement on direction is what gives confidence.

2026-10-15 · open the original ↗

• A deck's titles alone should tell the story. The presenter contrasts topic titles ("Revenue overview", "Channel breakdown") with claim titles such as "Revenue grew 78% in H2, but December growth nearly stopped", "TikTok drove all of it" and "AOV fell in all six channels". Each slide must pass four questions: the so-what (a finding, not data), the stakes for this audience, two or three pieces of evidence rather than 15, and an ask rather than "continue monitoring". The framework is tool-agnostic: Claude Code reaches Google Slides, or PowerPoint, through an MCP connection or a CLI.
• Have Claude write a deck brief before it builds slides. Starting from a long Google Doc that the lab's larger, paid skills-and-agents repo produced on fictional Novamart 2024 data, the presenter asked for a four-sentence arc (situation, complication, answer, action) and a slide plan for a 20-minute meeting with a VP of growth. Claude said the document led with a performance overview and buried the complication: H2 growth was bought with traffic, largely from TikTok, and the AOV decline was real and probably not a TikTok mix effect. It recommended holding the FY25 channel budget and closing three data gaps, and planned six slides where a straight conversion would have run to 20 or more.
• A critique skill caught a false headline in Claude's first pass. The lab's deck-critique skill graded the built deck C, 6.8 out of 12 across nine slides, scoring each slide 3, 1 or 0 on the four questions. It flagged that slide seven's headline was false (AOV did not fall every month). The presenter calls the first pass a draft, and the fix step had not finished during the session. The presenter attributes the clean, uncluttered charts to chart rules written into the skills, drawn partly from a storytelling-with-data book.
• Check each pipeline step on new data; collapse it once trusted. Asked why the analysis and the deck were two steps, the presenter says they run as one prompt in practice. On an unfamiliar question, though, you stop at each step to confirm the questions and methods before spending tokens, and merge steps only over later iterations. On PII, the presenter says access should be restricted before Claude Code ever sees the data, with a must-pass rule in CLAUDE.md as a backstop.
• The lab dropped Gamma because Claude Code is more flexible. This is the co-host's opinion: the hosts stopped paying for Gamma two or three months before the session because Claude Code can be made more tailored. The co-host thinks the "SaaS apocalypse" narrative is "overstretched by a mile", though some software can't beat general-purpose models.

2026-10-15 · open the original ↗

• The exhaustive deck answers the wrong question. The co-host, presenting, shows a 17-slide churn deck for a fictional B2B company (executive summary, methodology, cut after cut, key findings, recommendations) and says it answers "what data do we have?". An executive asking "how do the numbers look?" usually means "what should I do?". The co-host says the analysis underneath can be correct and still lose its impact because of where things sit.
• Lead with the insight: make each title the claim. The co-host replaces labels with findings: "Churn by segment" becomes churned accounts having skipped onboarding at a ratio of 4 to 1, "Q3 revenue overview" becomes "we lost $10 million in ARR and most of it was preventable", and "Key findings" becomes the decision to fund the onboarding push. The co-host then begins a structure of problem and stakes first, then insights backed by only the most salient evidence. The caption text stops there, after about 2,100 words of a roughly hour-long session, so the rest of the framework and the Claude Code demo are not in the source.

2026-10-15 · open the original ↗

• Without definitions, the model picks a column and sounds sure. The co-host, presenting, says an LLM knows what a customer is in general but not at your company, so it picks a column, a meaning for "week", and a view on whether a June dip is real, then answers confidently. Fixing this in the prompt works once and is gone next session, so the co-host says the prompt is never the place for it, pointing back to an earlier lesson in which asking AI the same question five times gave five different numbers.
• Onboard Claude like a new hire, on four things. These are what the data is (each file, what one row represents, what each column means), what the metrics mean, how the company reports (time windows, filters), and "what you would only know from the hallway", such as past incidents behind anomalies. In the co-host's example coffee subscription, "Brley", with two CSVs (metrics and ad spend), an activation means the day the first paid box is delivered, which is not the same as a signup.
• A definition you can't write plainly hasn't been agreed. The co-host offers this as a test for metric files: if you can't write it in plain, simple words, your team hasn't agreed on it: "a user who is active" defines nothing. The caption text stops here, after about 2,250 words of a roughly hour-long session, so the rest of the framework and the CLAUDE.md demo the title promises are not in the source.

2026-10-14 · open the original ↗

• Claims that Hormuz flows are near pre-war levels are false. He grants that September flows rose somewhat, partly as oil prices eased into early October, but says a reported October average of 2.9 million barrels a day as of October 11 is "just barely better than completely closed", against official talk of 80–90% of pre-war levels.
• October is the worst month of the war for shipping attacks. He counts roughly 30 tankers struck in the first 14 days, plus a UAE cargo ship that hit a mine and sank despite claims the strait was cleared. Turning off AIS doesn't help, he says, because spotters on fishing boats and gunboats call in coordinates for Russian-supplied weapons; 12 tankers refused to re-enter yesterday even under US escort, and freight is reportedly moving from $1.2 million toward $1.4 million a day.
• The US has answered none of the attacks, which he blames on election fears. Zero responses to the 30 strikes, he says, while the Houthis have hit an estimated dozen Saudi energy sites, including Saudi Aramco headquarters, Yanbu and the pipeline, which is still running. He reads the inaction as the Trump administration protecting itself before the election, and expects it to make things worse.
• The Russia diesel deal has not lowered prices, and stocks keep falling. Retail diesel has risen for about three days and wholesale diesel rose 25 cents a gallon yesterday. The penny-or-two dip in national gasoline prices he credits only to state gas-tax suspensions (around 40 cents a gallon in North Carolina, he believes). He cites an 11-million-barrel crude inventory draw (attributed to the API, then the EIA) and a 1.5-million-barrel diesel draw, leaving diesel 13–14% below the five-year average, by his estimate.
• He predicts a sharp oil price rise as shortfalls surface before the election. The government has not reported takers for the 40-million-barrel SPR loan eight or nine days on, against a usual three to five, which he takes to mean little or no oil moved. Once September's cargoes are delivered and October's fail to arrive, he expects the official flow numbers to be exposed and fundamentals to beat jawboning.

2026-10-14 · open the original ↗

• The US pulled its bombers from Fairford over lax British security. He reports that on October 4th, 2026 the US withdrew all B-1 Lancers from RAF Fairford in Gloucestershire, its main European bomber hub and host to many aircraft used against Iran. He ties this to the September 27th arrest of six people, one an Iranian-British national, over an alleged drone plot against the bombers, and says perimeter fencing and universal ID checks had not been maintained.
• Britain has lost a major bargaining chip with Washington. His inference is that the UK will no longer be the preferred platform for US power projection into the Middle East, so it can no longer trade Fairford basing permissions for a say in where the US strikes in Iran, or use the base as leverage in negotiating tariff relief.
• He expects Poland to host the bombers instead. He presents this as a guess: Poland has long asked for more US presence and, in his view, is not permissive toward jihadist infiltration. He concedes the risk of Russian infiltration is higher there, but argues Russian tampering with US equipment would be a red line Russia would pay heavily for, and allows that he could be wrong.

2026-10-14 · open the original ↗

• Dot-product attention is a kernel-weighted average, not something new. Naming the exponentiated score κ(q, kᵢ) = exp(q⊤kᵢ/√d_k) rewrites softmax attention exactly as a normalized kernel smoother; softmax only makes similarities positive and the weights sum to one, and with labels as values it is Nadaraya–Watson (NW) regression (Nadaraya and Watson, 1964). The author's toy case: labels 10, 20, 40 with weights 0.6, 0.3, 0.1 give 16. The argument comes from the author's SSRN tutorial and a theory paper written with Ming Yuan, with a companion notebook that checks each identity.
• The weights and the values are two separate choices. The same weights averaged over labels give a prediction, over feature vectors a representation, over residuals a correction (residuals −2, 2, 5 average to −0.1 in the worked example). Changing the similarity rule is a separate step and a modeling change: Gaussian attention with T = 2√d_k equals scaled dot-product attention plus a key-norm bias, so it matches plain dot-product attention only when all keys have equal norm.
• Stacked residual attention layers compute iterated NW, i.e. boosting. Context self-attention updates the fit, query cross-attention applies the averaged residual at new inputs, and the skip connection adds it, which is the BoostNW scheme studied by Di Marzio and Taylor; both updates must read the same incoming residual, a bug the author warns is easy to write. In a full transformer each head is still a kernel smoother, but the output projection, layer norm and token-wise feed-forward network mean the output is no longer an average of the original features. The construction shows attention can implement iterated NW, not that a trained transformer does.
• Initializing a trainable network as NW beat random initialization. On a four-layer restricted residual model (not the full block), test RMSE over 128 episodes was 0.290 after 250 updates and 0.263 after 1,000 with NW initialization, against 0.548 and 0.394 for PyTorch's default and 0.380 for untrained iterated NW, across three seeds. The author concedes the runs had not converged, the test functions came from the training family so nothing is shown about transfer, and shuffling labels (RMSE about 1.0) shows the network uses label–input pairing without explaining why.
• The kernel reading has limits, and validation must set depth. Context error keeps falling with layers while validation error levels off and rises, so depth acts like early stopping; a positive, row-stochastic attention matrix with eigenvalues 1 and −0.8 makes residual iteration diverge (×1.8 per step) at any step size; and iterated NW converges to an interpolating fit that still differs from ridgeless kernel ridge regression (KRR) at new inputs. Worked examples, not benchmarks: on the synthetic data NW scored 0.241, iterated NW 0.205, KRR 0.209; on a California Housing split iterated NW won on validation yet lost on test (1.105 vs 0.756 for NW and 0.632 for KRR), which the author keeps as the case for a separate test set.

2026-10-14 · open the original ↗

• Once intelligence is abundant, vision and ambition become the scarce resource. Wang expects that a decade from now it will look obvious that intelligence and agency became abundant, as current trends continue. He argues that smart people gathering around a shared goal was historically the bottleneck on progress, from the founding of the United States to every YC company, and predicts that this bottleneck goes away.
• AI makes execution far easier, so founders should dream bigger. Wang says agents and AI make acting on a vision perhaps 10 to 100 times easier than a decade ago. What remains is a clear view of how the world should look in 5 to 10 years, and the drive to "go through all the crap" to get it there.

2026-10-14 · open the original ↗

• Risk and consequence are not shared, so "together" is false. He argues that rising prices cost the wealthy money but cost working families necessities, and that factory closures hit workers, local diners and school tax revenue while executives keep their rewards. Politicians give speeches while families send their children to war. He says this is a claim about unequal distribution, not that business or government is evil.
• Even households with identical incomes want fundamentally different countries. His example is two families on the same street with similar pay, schools and commutes. One wants more government, sees taxes as justified and favours collective responsibility; the other calls taxation theft and stresses individual responsibility. He reads the Republican–Democrat split as surface over these competing visions of what America is.
• Unity means accepting disagreement, but only over shared facts. He holds that democracy never required agreement, only arguing, voting and accepting wins and losses, and says Americans can no longer accept losses. Disagreements must be rooted in the same reality. He asserts the country now lives in a "post-truth world" where facts do not matter.
• Americans are stuck with one another, not in it together. For him the test of democracy is whether 300-plus million people who disagree can share one country without deciding the other side has no right to exist. He says a growing, though still minority, share now holds that opponents should be deported or executed, and warns this is "compounding" and "coming."
• He blames comfort and privilege for a culture of grievance. In his reading, being born into privilege in America has left people unwilling to take responsibility, blaming others for unmet expectations. He voices this as a caricature: "I'm white, I'm Christian, I'm American, I'm a man... that's someone else's fault."

2026-10-14 · open the original ↗

• A stock this far down rarely recovers in public. The hosts say Bumble IPO'd in 2021 and is down about 95% from there, and argue the chance of getting a stock back up from 5% of its top is "slim to none". The buyout price they settle on, after some back-and-forth over the figure, is about $1.3 billion; they add that the founder expected to stay on but is not wanted.
• Going private lets owners rework a company out of view. One host's case is that a private company can change its model, raise capital and restructure without explaining each move in quarterly reports to investors; the other agrees and calls this the best of several reasons, framing it as flexibility for owners who still believe in the product.
• The hosts predict a relaunched Bumble within a couple of years. This is their guess, not a reported plan: either a reworked version of the app, or a merger with another dating site sold as a combined package.
• Being public puts a company under a microscope. Drawing on having taken two companies public, the hosts warn founders to be careful what they wish for: listing brings share liquidity but a very different operating world, and they say they often discussed taking their own companies private.

2026-10-14 · open the original ↗

• The September inflation report is unambiguously bad for Republicans. Reported: CPI rose 0.6% for the month, the biggest rise in five months, taking annual inflation to 3.7%, with tariffs and AI investment costs cited; Jessica Tarlov adds gas up 34.2% year on year and $1,043 in extra household fuel costs since the war began, and says it could keep the Fed on track for a December hike. The hosts argue Trump's answer in Alaska (stock market and 401(k)s at record highs, "affordability" mocked) misses voters, and Aaron Parnas notes the pardons he touted there went to people who pleaded guilty to disabling federally mandated pollution monitors under the Clean Air Act, not to "fixing an engine."
• Republicans' money advantage is buying less than expected. Tarlov says Republicans often outspend Democrats three to one but voters remember the Democratic cost-of-living message, and that Republicans have just stopped contesting Henry Cuellar's Texas seat; Parnas cites roughly $300 million from Elon Musk (Tarlov gives $276 million) and $400 million from Trump. Their inference: if Republicans must spend in Mississippi, Iowa and Texas, they have little hope of flipping Michigan or New Hampshire, and Trump is staying out of races like Mike Rogers's in Michigan.
• Parnas calls Musk and Hegseth war criminals over Ukraine. Reported: Ukraine lost partial Starlink access as Russia intensified strikes on its energy grid; Tarlov says the affected system was the Pentagon-run Starshield, whose throughput had been capped below what was needed to counter the latest Russian attack, and that Poland pays for Ukraine's Starlink. Parnas's opinion is that the outage killed civilians and left hundreds of thousands without power, but answering a viewer he concedes manslaughter would be too attenuated to prove and The Hague is out of reach because the US is not an ICC party.
• Ossoff will rout Mike Collins and carry Georgia's governor race. Parnas predicts Jon Ossoff wins by 8–10 points and pulls Keisha Lance Bottoms over the line by 1–2; Tarlov cites Ossoff hitting Collins over a son-in-law she calls a white-nationalist influencer and office staff with KKK ties. Parnas says he is more bullish on Iowa (where Ashley Hinson's post-debate ad claims she voted to release the Epstein files after working to block them) than on Maine, where Troy Jackson improved but faces an uphill race against Susan Collins.
• Talarico still leads Texas despite a Republican-heavy poll sample. Tarlov says Marist's new poll moved James Talarico's lead from 6 points to 2 but on a sample she describes as R+14, which she reads as a good sign. Both hosts argue the Wall Street Journal's report of Ken Paxton spending New Year's Eve 2024 in Dubai with two indicted fugitives and his girlfriend changes nothing, because his scandals are already priced in — a pre-election snapshot that will be overtaken.

2026-10-14 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-14 · open the original ↗

• A receiver swap is a bond position without the bond's cash. Receiving fixed (e.g. 4.5%) and paying floating SOFR gains when rates fall and loses when they rise, so it is synthetic long duration posted against margin rather than paid for in full. The catch the note stresses: that small outlay lets rate sensitivity scale far beyond what cash could buy, and daily variation margin drains liquidity on a losing mark, potentially forcing an exit before any recovery.
• Chile's pension funds scaled a winning bet in a duration race. After pandemic withdrawals drained liquidity, the seven AFPs (~US$229bn) used receiver swaps to rebuild duration and position for U.S. cuts; Fund E reportedly made +7.5% real in 2025 (flagged as not independently reconfirmed). A Citigroup analysis cited by Bloomberg puts the DV01 at ~US$10.6m in early 2024 versus ~US$131m by April 2026, more than 12×; the author's own arithmetic implies ~US$6.55bn lost on a 50bp rise.
• A rate shock flipped the book from gain to deep loss. The note attributes it to the Iran war spiking oil and inflation, pushing the 10-year Treasury above 5.35% on October 7, 2026, a 24-year high. Per Superintendencia de Pensiones figures the position went from +US$1.4bn a year earlier to –US$7.7bn in September; Fund E is –10.4% for Jan–Sep, including –4.9% in September. Notional was ~US$525bn in August, off a ~US$700bn peak; named counterparties include BofA and Citigroup.
• The regulator's deadline turns a valuation loss into a realized one. On April 9, 2026 it set exposure caps of 3% (Funds A/B), 2.5% (C) and 2% (D/E) plus a margin-call liquidity test, and an April 28 notice allowed only risk-reducing trades. The US$7.7bn is negative market value, not yet realized; the author argues the November 4, 2026 effective date forces closing at the worst mark.
• "Conservative" Fund E carried the most rate risk in the system. It is all fixed income, so it avoids equity drawdowns but is by construction a pure duration vehicle, and the AFPs stacked synthetic duration on top. The author's lesson: neither swaps nor the falling-rate view was the error; letting a working bet grow without a size limit was, since risk management exists to keep the position survivable when you are wrong.

2026-10-14 · open the original ↗

• Chile's pensions bought duration through swaps because cash was short. The host says two or three waves of COVID-era withdrawals left the funds needing duration to match liabilities; a receiver swap (receive fixed, e.g. 4.5%, pay SOFR) behaves like a bond but needs only margin. Their analogy: a $1,000 bond's exposure for perhaps $40–50 down, so a small rate move produces an outsized loss.
• The competing administrators chased each other into ever more duration. With seven AFPs, once one did it the others followed, the host argues; rate sensitivity grew from about $10m per basis point in early 2024 to $131m by April 2026, roughly 12×, implying ~$6.55bn lost on a 50bp rise. They stress notional of ~$525bn against only ~$229bn of assets.
• The Iran-war rate shock turned a gain into a deep hole. The position went from a $1.4bn gain to a $7.7bn loss as the 10-year Treasury yield topped 5.35% on October 7, its highest since 2002, exactly what a receiver book loses on. The regulator had already imposed limits on April 9, 2026, effective November 4, forcing position cuts on a deadline; the host says the true eventual loss is unknown.
• Commercial banks were the wrong counterparties, in the host's opinion. Bank of America and Citi are named as the main counterparties, with others possibly involved. The host says they would have gone to J.P. Morgan or Goldman Sachs instead, arguing that banks "think like a bank" and structure trades accordingly — offered as opinion, without detail on what went wrong in the structuring.
• The missing size limit was the mistake, not the swaps. The no-equity Fund E, nominally the safest, became the most rate-sensitive fund once synthetic duration was stacked on it. The host's lesson: risk management exists so a position is small enough to survive being wrong, and a fund already under pressure is now under more. They flag a follow-up episode on banks' synthetic risk transfers (SRTs).

2026-10-14 · open the original ↗

• Copying star investors means buying what the pros already moved past. One host takes this view of 13F copying, noting that the filings arrive six weeks late. The other argues that a filing's value is that it puts a stock on your radar, and uses this to pitch the hosts' upcoming One Lucky Dog platform as "endless ideas". The prompt was Berkshire's $20 billion net buying, which ended 14 straight quarters of net selling.
• Record retail options activity reflects engagement, not a market top. One host says this even while personally believing the market is near a top. The data: Citadel Securities retail dollar demand about 25% above the 2021 high, July net buying 3.2x average, a record $2.6 trillion of S&P call notional in one May day, and OCC volume up 14.5% year to date. The hosts note that Wells Fargo's contrarian sell signal, given in May, has been followed only by gains.
• Widening AI-debt spreads are a bond-market warning worth heeding. "The bond market doesn't miss," one host says, though the hosts do not call a top. The figures: hyperscaler spreads moved from about 50bp to 80bp, Oracle's 5-year CDS reached ~203bp (an 18-year high), and the five largest hyperscalers issued $121B of bonds against a $28B/year 2020–24 average. They also cite 78 of 91 2026 hyperscaler bonds trading above their issue yields. Both hosts prefer Treasuries.
• A significant pullback looks likely, so positions should be smaller. One host says a 10–15% decline would not surprise them and would be healthy, and is running about 40% of normal size. Their reasoning is that the market has been lifted by AI, so a stumble in one AI name could spread to all of them. They point to Micron falling about $130 from its earnings level.
• A zebra mimics stock cheaply but loses faster in modest declines. In the worked example with the stock at $100, you buy two 93 calls at $9.45 and sell one 100 call at $5.19, a $1,371 debit for about 98 deltas versus $10,000 for the shares. Between 93 and 100 it loses $2 per $1 the stock falls: down $671 at 96.50 versus $350 for the shares, with a maximum loss of $1,371 versus $700 at 93. One host still prefers it for capital efficiency.

2026-10-14 · open the original ↗

• Constant market news-checking damages portfolios more than it helps. He argues that chasing news, Reddit and influencer takes breeds over-analysis that shakes investors out of positions that would have paid off later. He says creators posting daily updates are mostly chasing clicks, and that he will post only when an update is actually warranted.
• The S&P 500 is consolidating in a bullish structure, not breaking down. He reads it as a volatility contraction pattern still making higher lows within the same range as last week, which he says points to an upside move, though no breakout has happened yet. Silver has made a higher high and no lower low since his July update, so he sees nothing to change there.
• Small pullbacks are not grounds for panic. He says -1% or -2% moves need no action; only declines of -30% to -40% in small/mid-caps or blue chips would warrant fresh updates. He calls the Fear & Greed Index nearly useless but says its fearful reading, with the market well above a year ago, shows a market climbing a wall of fear.
• His content surges only at breakouts, breakdowns or buy-list building. As a breakout investor he posts more when the market or individual stocks tap 52-period highs or 52-period lows, and during consolidations he builds buy lists (or short lists, depending on conditions), most recently from the Midweek Movers winners. He frames his expectation of another leg up, including in his own holdings, explicitly as speculation.

2026-10-14 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-14 · open the original ↗

• Federal research cuts are dismantling America's innovation base. The author reports nearly $200 billion in federal funding cut, delayed or disrupted across 17 policy sectors and 21 agencies since Trump returned to office: NSF awarded 45% fewer grants this year than its five-year average (a four-decade low), NIH 13% fewer, and more than 400 CDC grants worth nearly $6 billion were terminated. The stakes the author cites: every FDA-approved drug from 2010 to 2016 traced to NIH-funded research, stealth, night vision, GPS and precision weapons came from government basic research, and nearly a third of U.S. patents rely on federal funding.

• Grants are being politically screened for disfavored words, not waste. NIH staff were told to prescreen applications with a "computational text tool" that staffers describe as an LLM; Nature's Max Kozlov reported a list of 235 flagged terms including "racism," "fossil fuel" and "sexual minorit." Nearly $800 million in NIH grants was cancelled for focusing on or mentioning minorities, including $15 million of Alzheimer's research at Emory and the University of Pittsburgh studying why Black Americans are twice as likely to develop the disease; Noam Ross of Grant Witness calls it banning topics without saying so.

• Restored grants don't undo the damage that delays cause. Of more than 11,000 disrupted NIH, CDC, NSF and SAMHSA grants, 65% were later restored by courts or reviewers, and renewals now take months instead of weeks. The author's illustration: 8-year-old John Paul Macri, diagnosed with DIPG (2% five-year survival) in March 2025, lost access to an mRNA trial when DOGE cut the NIH Pediatric Brain Tumor Consortium five months later, and died on October 4.

• Funding is shifting toward AI and appointees' pet projects. In 2025 health research lost $1 billion while machine learning/AI gained $468 million and networking/IT $368 million; this year the White House announced redirecting $200 billion toward AI, while new NSF grant allocations fell more than 20% year over year. The author argues government should fund what private investors won't, not fields already drawing trillions; Ross flags an NIH metal-toxicity RFP with an unusually short window as cover for studying vaccine aluminum, and warns the lost trust will deter universities from whole research areas — AAAS CEO Sudip Parikh calls it "a 10-year, 20-year, 30-year problem."

• Obscure federally funded science pays off decades later. Golden Goose winner Zhen Xu discovered histotripsy in 2002 quieting an ultrasound on pig hearts; with only NIH and later DoD backing for years, it won FDA approval for liver tumors two decades on, and her company HistoSonics is valued at $4 billion. She credits the U.S. for funding young researchers on data alone, unlike China; fellow winner Sidney Nagel's tabletop work on coffee-stain rings and sand piles improved medical diagnostics and drug-capsule powder filling, and he notes his main cost is training young scientists.

2026-10-14 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-14 · open the original ↗

• Fabric's case is one governed platform instead of integrated pieces. Hitney, a Microsoft partner solution architect and Giard's teammate, describes Fabric as Microsoft's SaaS data and analytics platform spanning warehouses, lakehouses, real-time and unstructured data, with Power BI now part of it. He argues the recurring problem with earlier platforms such as Synapse Analytics was integrating, governing and securing separate data and application layers.
• Fabric Apps exist because Power BI only reads data. First shown at FabCon Atlanta, Fabric Apps are deployed into Fabric to inherit its security, governance and accessibility. Hitney says Fabric now holds operational databases, creating demand for apps that write data or trigger actions, which a Power BI report cannot do.
• Rayfin ships agent skills so coding assistants build apps without guessing. The SDK, CLI and backend-as-a-service lives in a GitHub repo containing skills for Claude Code, GitHub Copilot and others, which Hitney says prevents hallucination about a new tool. Apps are written in TypeScript and become React apps; npx rayfin up compiles and deploys them to Fabric, and they can be run locally first.
• Knowing the data model matters more than knowing TypeScript. Hitney says he barely knows TypeScript himself; what counts is knowing the entities in Fabric that the app connects to and the outcome wanted. Giard puts it as business knowledge outweighing technical knowledge, which Hitney accepts.
• His demos took minutes, though the product is in preview. A plant-operations app showed live sensor data, dispatching maintenance, work orders and an "ask advisor" button that sends telemetry to Foundry, which recommends actions from manuals, service history and parts inventory. A Kanban-like Technology and Friends app took "only a couple of minutes"; Fabric Apps were in preview at recording, with no timeline he knew of.

2026-10-14 · open the original ↗

• Invest on supply discipline, not demand forecasts. The guest's quip is that "supply is measurable, but demand is storytelling": fabs, ships and copper mines can be counted, demand for them can't be predicted, and curtailed or absent new supply improves the odds of a good outcome.
• Memory chips were a textbook capital cycle. He says the memory makers consolidated from about 20–22 players in the 1990s to three, so capital was allocated more rationally while the rising cost of multi-billion-dollar fabs raised barriers to entry.
• The AI memory windfall was luck, not a forecast. He stresses that his early semiconductor investments carried no view that memory would become the AI bottleneck; a supply-discipline approach simply positions you to benefit from that kind of luck.
• He is selling because memory makers trade far above replacement cost. Responding to Frost on a roughly decade-long holding, he says he has sold aggressively over about the last six months, as the companies are now valued at three to four times what it would cost to rebuild their fabs. His call is that the cycle is "much closer to the top than the bottom," though turning will take time.

2026-10-13 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-13 · open the original ↗

• Academia's useful remaining AI role is the ethics of digital minds. Nick Bostrom (the same passage appears in his full interview of 2026-10-16) wants academic groups working on it alongside nonprofits and AI labs. A university base makes the field less likely to be "excised and tabooed", and professors writing papers make it harder to dismiss as nonsense. He adds that leaving university life, especially the administration of running a research institute, has been "a big relief".

2026-10-13 · open the original ↗

• Clickbait titles rarely resemble what is actually in the video. Asked by Yampolskiy for a clickbait title for their interview, Nick Bostrom offers a joke one: tips on how to lose weight, make a million dollars and become a better lover. He adds that he doesn't recommend using it.

2026-10-13 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-13 · open the original ↗

• Managing a losing trade is judgment, not a fixed rule. The host answering calls it "art, not a science": the choice depends on the stock, the position, price and time. The one near-exception is an outright futures trade put on for the day, which the host would exit once it is clearly wrong, and even that is described as not fixed.
• Time to expiration, not price, triggers the management decision. For positions opened at roughly 40 to 60 days to expiration, the host starts looking to adjust around 21 days out, within a loose window of about 15 to 25 days. A second voice closes by saying the point is to have some plan in mind.
• The right action depends on how the position is built. Short premium the host still wants gets rolled from one month to the next. A defined-risk directional trade that went wrong gets left alone ("hands underneath my thighs"). A position with naked exposure that has moved well against them would likely be closed, perhaps replaced with a vertical spread in the preferred direction.

2026-10-13 · open the original ↗

• AI arrived as a flawed human, not a perfect fly. Paul Graham, who studied AI in the 1980s, says the expected path was to build a "perfect fly" and climb through mice, cats and monkeys to humans, flawless at each level. Instead the first ChatGPT versions were "full human but terrible", like an undergraduate bluffing through a paper, so the field now starts at human and works toward perfect, which he says nobody expected.
• Current AI still bluffs, and its competence is uneven. Graham says it keeps much of that bluffing-undergraduate flavor: he hears it is solving famous open math problems, yet he cannot get it to say when restaurants are open. His interviewer calls this the "jagged frontier"; Graham guesses, half-joking, that the people reporting the math results have research versions he lacks.

2026-10-13 · open the original ↗

• China's systemic risk is fiscal; banks are the conduit, not the source. The brief argues that the land-sale-and-off-budget-borrowing model that funded local government for two decades has broken. Beijing is moving the losses onto the public balance sheet, so the open question is the state's capacity to keep recapitalising the banks it owns or controls.
• The local revenue base is collapsing faster, not stabilising. Land-sale revenue fell 31.5% y/y in H1 2026 (977.8bn yuan), after a 14.7% fall in 2025, the fourth straight annual decline. General budget revenue fell 1.7% in 2025, its first drop since 2020, and the 4.7% H1 2026 recovery is tax-driven. Spending obligations did not fall with it.
• The debt trajectory is the problem more than the level. The IMF puts LGFV "hidden" debt at 71.4tn yuan (~51% of GDP) at end-2025, up from 64.4tn at end-2023 and larger than official local debt (54.8tn). General government debt goes from 60% of GDP in 2019 to 107% in 2026 and a projected 124% in 2030 (154% augmented), with a ~7% primary deficit and no consolidation under way.
• Bank clean-ups and new taxes read as fiscal measures. The 10tn yuan debt swap from Nov 2024 covers only about a quarter of hidden debt (Fitch), 500bn yuan of special bonds recapitalised four state banks, and 670 small banks closed in 2025 at provincial cost. Revenue moves include 6% VAT on new bond interest (32bn yuan in 2025, 99bn by 2027) and a 20% tax on foreign individuals' FIE dividends from 1 Sept 2026, ending a 1994 exemption.
• Expect a slow Japan-style absorption, not a default or bank run. With debt in yuan and only 2.2% held by non-residents, the author infers stress shows up as local austerity, heavier taxation, financial repression and weak, deflationary growth (4.3% in Q2 2026). The counter-case is Beijing's view that hidden debt is resolved; watch monthly land revenue, LGFV-list removals and the 2027 transfer budget.

2026-10-13 · open the original ↗

• Attacks on shipping in and near Hormuz are continuing. UKMTO reported six vessels attacked on Sunday, October 11th (five tankers and one container ship): five inside the strait, one by drone near Sohar in Oman. On Monday Iran claimed a third supertanker carrying 750,000 barrels was hit; the host stresses this is Iran's claim and UKMTO has not confirmed it.
• Russia's diesel export promise is not a flood; watch the numbers. Russia's Novak says diesel exports could reach 3.5 million tons a month by year end, back to pre-crisis levels, but Russia's own export ban runs to the end of October and the first shipments in the deal amount to only a few days of use. As of October 13th crude fell about 1% on the news while the diesel crack spread stayed above $100 a barrel, which the host reads as the market "kept paying the fear price."
• Houthi strikes on Saudi Arabia are escalating, and oil majors are noticing. King Khaled airport in Riyadh is closed until October 17th; Saturday's strike killed 13 and Sunday's one more, for 14 dead and more than 300 wounded. On Monday the Houthis claimed new strikes on the airport and an air base, which Saudi Arabia said it shot down; the BP and Shell CEOs pulled out of the World Petroleum Congress in Riyadh, which the host takes as a message in itself.
• Trump's announced energy ceasefire has not taken hold. Moscow and Kiev have said nothing about a formal truce and reports of fresh strikes keep coming; the host's verdict is that "an announcement is not an agreement."
• Europe enters winter with gas storage short of its own mandate. As of October 13th tanks are at 73% against a mandate above 80%, with some countries such as Germany significantly lower, and prices spiked to near €86, the highest in almost four years. The host attributes the thin storage to Qatar's gas being mostly still trapped behind Hormuz, and closes that "the screen looks calmer than the water."

2026-10-13 · open the original ↗

• Smart TVs fingerprint viewing and may leak your device inventory. Citing Gamers Nexus, the host reports that LG TVs scan the LAN for device names, MAC and IP addresses, and scan nearby Wi-Fi to fix location. One used only as an HDMI monitor sent location about 20 times a minute, and LG's ad business claims about 33 million US TVs. That the device list is sent is his suspicion: the traffic is encrypted.
• UniFi allows traffic between VLANs by default, so isolate them. The host recommends three VLANs: default for trusted gear, guest with internet only and Wi-Fi client isolation, and IoT for the TV. IoT gets network isolation but keeps client isolation off, because smart devices need to see each other. Default stays un-isolated, so it can open connections into IoT (e.g. Home Assistant) but not the reverse.
• Wired device isolation fails on switches without ACL support. UniFi lists the switches it does not cover, such as the USW Flex and Ultra. A TV on a Lite 8 PoE downstream of a Pro Max still sees everything on that small switch. Isolation also breaks casting, so he advises casting only within one VLAN and turning mDNS forwarding off: a forwarded device shows up across VLANs but times out.
• Locking DNS to your own Pi-hole best blocks TV telemetry. The host's tool sets DHCP to hand out the approved servers, opens UDP/TCP 53 to them, and blocks DNS and DNS over TLS (port 853) everywhere else. Streaming still works while telemetry domains fail; Roku is the noisiest device in his Pi-hole. DNS over HTTPS cannot be blocked, hard-coded DNS breaks, and clients need a DHCP renew.
• The host's own free tool applies these settings in one click. House Arrest is a new, open-source part of his UniFi toolkit, runs in Docker against a UniFi API key, and offers per-device presets (internet only, LAN only, no internet, quarantine). It shows the JSON before applying and rolls back with a click. He says he "vibecoded" it with Claude Code and warns it is not for beginners.

2026-10-13 · open the original ↗

• CoreWeave's bond stress is the author's early blow-up signal. Its bonds moved from 14% to 15.1% yield in a day. The author's model said in April that by September it would be clear whether, and who, would blow up, and picked CoreWeave as the leading candidate; the author is waiting to see what follows.
• China's crisis is fiscal, not a banking crisis. The land-sale model that funded local governments for two decades has broken: land revenue fell 31.5% in H1 2026 after almost 15% last year, a fourth straight annual decline. Government debt went from 60% of GDP in 2019 to 107% in 2026, projected 124% in 2030, or 154% on an augmented measure.
• Beijing is shifting local losses to the centre and hunting revenue. Cited steps: a 10 trillion yuan bond in 2024 for off-book "hidden" debt (estimated to cover about a quarter of it), a 500 billion yuan special issue in 2025 to recapitalise big banks, about 670 bank closures since, and a restored 6% VAT on interest from new government and financial bonds. The author expects all governments to follow.
• Stress shows as austerity and repression, not default. With debt held domestically, the author argues it appears as local service cuts, arrears to suppliers, heavier taxes on savers and foreign-owned firms, lower rates and short-term issuance, as in the UK, France, the US and Japan. With GDP growth at 4.3%, lowest since 2022, and weak consumption, exports are the only working engine; the author reads Xi's Washington visit as buying two to four years of exports to fix the home economy.
• The official reading is orderly consolidation; specific data will decide. The finance ministry points to the budget up 4.7%, tax receipts up 5.3%, and land spending falling with land revenue. To watch: monthly land revenue, special bond issuance (2.07 trillion yuan in H1 2026, about 47% of quota), withdrawal of foreign bond investor tax exemptions (now to end-2027), removals from the LGFV list, and central transfers in the 2027 budget.

2026-10-13 · open the original ↗

• China is letting the yuan strengthen despite domestic deflation. He frames this as what Washington has demanded for decades, having accused Beijing of weakening its currency to make exports cheaper. He calls it a dangerous game: prolonged deflation, he argues, is worse than inflation because producers cut wages or lay off workers nationwide, and a stronger currency deepens it.
• Surging exports give Beijing room to absorb a stronger currency. He cites Chinese exports up 25% year-on-year in August even while the domestic economy struggles, and reads this as letting China afford a stronger yuan to bring in more revenue.
• The main motive, in his reading, is paying for oil. He says China gets little oil from Iran since the war but still buys heavily from Russia, which sells to China only in yuan, so a stronger yuan buys more. With 85% of world oil traded in dollars, which have also been strengthening, he argues a stronger yuan protects China's purchasing power for a dollar-priced commodity.
• Beijing is choosing economic pain over an energy shortage. He concludes China would rather deepen its population's hardship than risk running short of energy, even though it has already cut imports. He asserts the United States is "100%" responsible for this catch-22, without setting out why in the clip.

2026-10-13 · open the original ↗

• Annual S&P price targets are a sales function, not a forecast. The show's framing: since 2012 consensus targets missed in 12 of 14 years, by 6% on average, and the S&P already sits about 3% above the January 2026 strategist average. The hosts argue banks can't publish bearish calls without hurting their capital-raising business, so a sell rating arrives only after a stock is "obliterated". They add that a target below 10% would count as bearish, and that short-term options traders should look at IV rank instead.
• A same-strike naked put beats a covered call on capital. The hosts call the two positions identical in delta, and say the dividend is priced into both. One host adds a behavioural caveat: covered-call holders tend to keep managing the position, while put sellers tend to drop it. In an IRA the suggested substitute is a wide put spread with a cheap long leg. For SpaceX, a $150/$115 spread for about $6.50 uses roughly $3,500 instead of $15,000.
• A put ratio spread is a bullish high-probability trade. The hosts reject the "modestly bearish" framing. With the long put at the expected move (about the 84% level) and the shorts near the 90–95% level, they put the chance of profit at roughly 88–94%. Their rule is to read direction from delta at entry and take small profits early, such as 25–50 cents on a 50-cent or $1 credit.
• Nvidia's financing of an OpenAI data center is circular. The show reports a guarantee of up to $100–105 billion (both figures are given) for a site in Ohio. Both hosts call it suppliers funding customers who must then buy their chips, a "loan-sharking" use of cash that would otherwise go to buybacks or dividends. They see red flags but nothing illegal, and worry that markets near record highs leave no room for error.
• The hosts split on whether a 25% AI-stock drop is near-certain. One host gives it a 100% chance in 2027, arguing that Micron, AMD, Nvidia and SpaceX each fell more than 25% at least twice in 2026. The other answers that Nvidia's one-year expected move is about $60 on a $225 stock (about 25%), so a full move in one direction is no certainty. That host would hedge by spending 3–5% of premium collected on shorter-dated puts. A $200,000 bet is offered on air.

2026-10-13 · open the original ↗

• Russia is reportedly paying for Iranian crude with weapons, not cash. An oilprice.com report describes an oil-for-arms arrangement: heavily discounted Iranian oil moves into Russia in exchange for missiles, air defense systems, military hardware and satellite and surveillance technology. He adds that Russia has already been reported to give Iran satellite and intelligence tracking on US military targets.
• Iranian crude is being laundered through Russia's petroleum system. Per the same report, Iranian oil is blended with Kazakh crude to disguise its origin. Once it enters refining and pipelines it carries no label, so he infers Russian diesel is likely made partly from Iranian oil, while conceding there is no direct proof the exact barrels reach US-bound diesel.
• The US Russian-diesel license sends American money into that circle. Washington has lifted restrictions and issued a license for the sale, delivery and import of Russian diesel; he says President Trump is excited about 300,000 metric tons, which he calls a drop in the bucket. He reads the exception as driven by high diesel prices and an approaching election.
• He calls this a contradiction that prices troops' lives in cents. The US sanctioned Russian energy to avoid funding Putin and Iranian oil to avoid funding Tehran, and now pays both, he argues, for savings of perhaps 5 to 20 cents a gallon. He says Russian-supplied equipment has been used to kill American troops in Jordan, and calls the policy geopolitical insanity, not energy security.
• Iran's attacks on shipping are going unanswered, he says. He reports Iran hit six tankers in the last 24 hours and that the US has not responded in nearly a month, while still urging allies to stand against Russia. He wants officials to say where every barrel feeding Russia's diesel exports comes from and what the money finances.

2026-10-13 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-13 · open the original ↗

• No text was available to summarize. The transcript for this 16-second short came through empty (0 words, and the .srt is empty too), so nothing here reflects what the video says about the dollar versus gold.

2026-10-13 · open the original ↗

• US crude exports reflect refinery design, not surplus. As of this 13 October 2026 video, the host puts US crude output at about 14 million bpd, with 10.4 million kept at home and 3.6 million exported, while US refineries process about 16 million bpd. He says the large Gulf Coast refineries were engineered decades ago for heavy sour crude imported from Canada and Latin America, so the light sweet West Texas crude is partly exported and heavy grades are imported.
• High US diesel prices come from a global shortage, not domestic scarcity. The host gives US diesel output as over 5 million bpd against domestic consumption of about 3.8 million, with the surplus exported. Because diesel trades globally, he says US prices are being pulled up by three supply losses: Russia's diesel export ban after drone strikes damaged its plants, the Middle East war knocking out several major export refineries, and the Strait of Hormuz bottleneck making it hard to ship refined fuel out of the Persian Gulf.
• Quoting diesel per barrel lets analysts measure the refining margin. Diesel is sold by the gallon in the US and by the metric ton in Europe and Asia, and analysts convert it to barrels to compute the crack spread. His worked example: wholesale diesel at $5 a gallon is over $200 for a 42-gallon barrel, so with crude at $100 the refiner's margin is roughly $100.
• Building new refineries is too slow and risky to fix prices. A new refinery takes years and billions of dollars; in the host's example, a $4 billion small refinery is underwater if the refining margin collapses to $20 four years from now. His own view is that only physical fixes will lower fuel prices: damaged refineries returning, export bans ending, shipping routes normalizing, or global demand falling.

2026-10-13 · open the original ↗

• Stick to fundamentals instead of forecasting where AI goes. Taylor says nobody knows where AI is heading, and over-studying it produces analysis paralysis or public doom-spiralling; clean data and understanding stakeholders' actual problems stay true regardless. The host adds his own fundamental: plan for better models, since frontier or open-weight models six months out may do what a team spent personnel and compute fine-tuning for.
• Leaders should commit to tools rather than chase each release. Taylor argues constant switching between models causes team thrash, and that in her personal use the gap between Sonnet and Fable comes down to edge cases while her workflows have not changed dramatically. The host disagrees in part, saying Fable regularly blows him away and that around Opus 4.6 in February or March he felt his technical skills stopped being a moat.
• Workslop breaks the rule that the sender carries the burden. Taylor says the social contract is that what you send is your thinking or has been reviewed by you; unreviewed AI output shifts the work to the reader. Answering an audience question, she says using AI to stress-test strategy or drafts is fine because "you own the output", but verbose, padded output still reads as slop.
• A "what this isn't" section gives away AI-written documents. Taylor says she has only recently started seeing disclaimers of what a document is not, and attributes them to the negative context users give a model while prompting ("don't go there"), which the model then passes on to a reader who does not need it. She distinguishes this from a legitimate in-scope/out-of-scope statement.
• Women are being promoted onto a "glass cliff". Taylor reports seeing more women in data and leadership roles, but several she knows were brought in during crises, where failure is likely or they are expected to handle emotional fallout such as downsizing. She argues women should be elevated when things are going well, on expertise.

2026-10-13 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-13 · open the original ↗

• Supply is measurable; demand is storytelling. Davidson's capital-cycle framework says the more capital chasing an industry, the lower its returns, so Hosking Partners counts fabs, ships and mines rather than forecasting demand. His argument is that a supply-constrained industry benefits from any demand surprise, whatever form it takes, and that few competitors invest this way.
• US hyperscalers have entered a negative capital cycle. Davidson says they have gone from $200–300 billion a year of free cash flow to free-cash-flow negative, from buybacks to issuance, and from about 1% of investment-grade issuance in 2023 to a fifth of it. He compares the scale to the mid-1800s US railroad boom and notes capex estimates have more than doubled from the $3–4 trillion his "AI paradox" piece flagged a year ago. His inference is that data centres are commoditised factories: China processes three-quarters of AI tokens at about a fifth of the cost, which he expects to force US capacity into price competition.
• The memory-chip trade is near its top, so they're selling. The position in Micron, Samsung and SK Hynix rested on consolidation from about 20–22 players in the 1990s to three, plus rising fab costs, not on any AI demand forecast; the average holding period is just under a decade. Over roughly the last six months they have sold aggressively because enterprise value is now three to four times replacement cost, and Davidson expects new capacity from Chinese IPOs and US reshoring.
• Platinum-group metals are the ideal supply-constrained setup. Davidson says platinum is about 20 times rarer than gold, mined 2 km deep, mostly in South Africa, by about four dominant companies. Fear that EVs will kill auto-catalyst demand has halved industry capex from about $600 an ounce a decade ago to $300, and he argues that constrained supply matters more than the unknowable demand outcome.
• Japan is a country-level capital cycle shrinking the denominator. Davidson says corporate Japan is raising return on capital by cutting invested capital, unwinding cross-holdings, excess cash, duplicate head offices and idle plant, which most companies avoid because of the social friction. Japan is the fund's largest country position. They play it by co-investing alongside activists who target low-valuation companies under the stock-exchange rules, and he expects the trend to have a long way to run.

2026-10-12 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-12 · open the original ↗

• Position size, not pick quality, decides who survives a losing streak. The setup gives two traders identical picks and a five-trade losing streak: risking 1% per trade leaves 95% of capital, needing a 5% gain to recover, while risking 10% leaves 59%, needing 69%. The hosts' slogan is "size kills" — genius fails only by getting too big.
• Risk 1–3% per trade, because an "edge worth sizing" is only perceived. One host argues nothing really bad can happen at 1–3% of the account per trade, since five-trade losing and winning streaks are statistically certain. The same host rejects the premise that a real edge justifies sizing up, calling it "just the perception of an edge." No data beyond the opening example is offered.

2026-10-12 · open the original ↗

• The energy shock is very inflationary but has not yet shown up. One of the hosts says they have been telling people the pass-through will take 6–7 months, noting diesel at record highs, administration talk of restricting exports and a G7 release of oil and diesel, while inflation prints have not yet matched 2022's.
• China's quiet reserve drawdown kept prices below expectations. The economist answering says prices rose far less than expected after the Strait of Hormuz shut in February, seemingly because China stopped buying on world markets and drew down its stockpiles, a "global public good". Nobody knows how large those reserves are, so escalation, such as Iran hitting Saudi pipelines to the western ports, could push prices much higher.
• Data-center demand keeps adding pressure on energy. The economist names the data-center build-out as a second factor, with increasing demands on energy and no sign of it slowing.
• The November election raises the stakes on high energy prices. The economist says politicians least want high energy prices heading into US elections, because gas prices are what motivates voters.
• Pass-through to inflation will be slow, uneven and longer-lived. The economist expects it to depend on Asian manufacturers, now often in Vietnam or Indonesia rather than China, eventually passing energy costs on in goods prices. Long supply chains and an erratic strait mean the causes of each month's figures will only be clear in hindsight; the host who asked says they have abandoned forecasting it.

2026-10-12 · open the original ↗

• Lossdog pitches a new Correlation Map feature. As the company describes it, the tool takes any ticker and shows how it moves alongside more than 9,000 stocks, ETFs, funds and major cryptocurrencies, from what it says are over 229 million correlations across six time frames. It is also said to flag which relationships are new, persistent or recurring. This is a product announcement; no example results are shown.
• New and upgraded tools are flagged in the platform's navigation. Stars in the left-side navigation mark recent additions, and the company points viewers to a platform-features playlist on YouTube for tutorials. The piece is entirely promotional and makes no market argument.

2026-10-12 · open the original ↗

• The US blockade of Iran is probably leaking through ship-to-ship transfers. He plays an intercepted radio warning in which a US aircraft tells a vessel that received Iranian cargo by ship-to-ship transfer to turn towards Chabahar. He reads it as proof that someone in the Gulf is helping Iran, and infers that some cargoes have got out. Any leakage, he argues, extends Iran's staying power by weeks and drains US crude, gasoline and diesel inventories for longer.
• Russian diesel will not drive US prices down "in record numbers". Treasury's General License 135 authorized imports of Russian diesel, which Trump said would cut prices in record numbers. He expects pennies at most, because Ukrainian strikes have damaged as much as half of Russia's diesel-producing refineries, Russia had to ban diesel exports, and Ukraine hit its largest refinery days ago. He does not believe that capacity has been restored.
• Russian cargoes bound for the US deserve provenance scrutiny for Iranian supply. He stresses there is no evidence yet, and no Russian diesel has arrived. He says a tanker reported as bringing diesel actually carries fertilizer. His concern is indirect: Iranian product entering Russia's supply chain, or Russia sourcing elsewhere such as Belarus, would free Russian barrels for export to the US.
• Sanctions on fungible fuel only add a middleman. He argues that under every administration, Republican or Democrat, allowed suppliers buy from sanctioned countries and resell to the US. Here that would mean paying both Russia and Iran to ease a shortage that wars involving both helped create. He calls that a self-defeating policy, and says where the fuel was refined and who got paid is what matters.

2026-10-12 · open the original ↗

• Democratic donors are neglecting competitive governor races. Citing Politico's Jonathan Martin, the post reports small-dollar money has flowed to candidates like James Talarico and Adam Hamilton while Republicans are on track to outspend Democrats on TV roughly 6-to-1 in Florida, over 3-to-1 in Texas, 2.5-to-1 in Georgia and over 2-to-1 in Wisconsin, with narrower gaps in Ohio and Nevada; the quoted reporting says GOP strategists concede they would lose most of these races without that money. The author's stated stakes for the 36 races on November 3: who oversees certification of the 2028 presidential result, plus reproductive rights, healthcare, education and redistricting (spending ratios are a snapshot as of 22 days out).
• Arizona and Nevada put 2028 certification in election deniers' reach. In Arizona, incumbent Katie Hobbs, who has spent four years vetoing the Republican legislature's bills, faces Andy Biggs, who pushed the 2020 "Big Lie" and still has not acknowledged Biden's win; the author calls it a must-win. In Nevada, incumbent Joe Lombardo (who flipped the state in 2022 after COVID hit its hospitality economy) said in 2021 he didn't know if the 2020 results were accurate and has kept company with a "fake elector"; challenger Attorney General Aaron Ford runs on $1.2 billion in opioid settlements, cost of living, and a pledge that data centers won't raise Nevadans' costs "one penny."
• Florida's open seat is the Democrats' best opening in years. The author argues conditions are favorable: the last open race in 2018 went to Ron DeSantis by 0.4 points (about 32,000 votes), the 2026 environment is likely worse for Republicans, and Democrat David Jolly, a moderate former congressman running on affordability, health and education, faces Byron Donalds, described as struggling all year. Beyond Donalds' Trump ties, the author points to population growth of 1.7 million driving up property taxes and rents; Jolly says the race is winnable if donors pay attention.
• Oregon is unexpectedly close despite being Democratic territory. It is a rematch of Tina Kotek against Christine Drazan, whom she beat four years ago on reproductive rights, but abortion is less prominent now. Kotek is weighed down by what The Oregonian calls four simultaneous problems — housing, schools, mental health and economic sluggishness — plus Portland crime headlines, though the author notes violent crime is down.
• Wisconsin is a must-hold for the 2028 path. Democrat David Crowley, backed by outgoing Governor Tony Evers and running on apprenticeships, small-business formation and healthcare access, led Republican Tom Tiffany by just three points in the latest Marquette poll (a pre-election snapshot). The author stresses Tiffany is outspending him 2-to-1 in one of the more Republican-leaning Rust Belt swing states.

2026-10-12 · open the original ↗

• The S&P 500's consolidation is about to resolve in a big move. He says the index has consolidated since ~13 August with locally higher highs and higher lows, and expects the compressed price action to break within days or next week. He leans upside because no individual stocks have truly crashed and longer consolidations tend to produce bigger moves, but allows it could be a "huge fake out".
• Rainy.talks' three AI-infrastructure picks pass his screen. The TikTok creator named Amphenol (high-speed AI cables and connectors), nVent (liquid cooling) and Eaton (grid-to-server power). He finds all three in well-developed uptrends with locally higher highs and lows, and growing EPS and revenue annually — the conditions he wants. nVent placed second in last week's midweek movers.
• nVent is his favourite of the three; Amphenol his least. His objection to Amphenol is share count, about 2.47 billion shares outstanding. He also likes that these names lack the bloated market cap he sees as Nvidia's problem.
• He trades for probability of running now, not five-year holds. He explicitly declines the creator's five-year framing, saying no one has a crystal ball. He contrasts these picks with typical influencer stocks sitting at 52-week lows in downtrends, where dollar-cost averaging is in his view the worst thing a portfolio can do.

2026-10-12 · open the original ↗

• Position sizing matters more than picks, and edge is only perceived. The setup: two traders take the same five-trade losing streak. Risking 1% a trade leaves 95% of capital (5% to recover); risking 10% leaves 59% (69% to recover). The hosts agree that "size kills" and that 1–3% per trade means nothing really bad can happen. Their only concession is that premium sellers get paid more for the same delta when volatility is high.
• Tokenized stocks' extra volatility is an illiquidity problem, not new risk. On the IMF finding of about 1.5 times the volatility of the underlying shares, one host argues the cause is overnight spreads as wide as $2 in Apple against about $0.05 listed. That host predicts no added systemic risk and calls tokenized derivatives the real prize. The other sees no point for US investors in US names, only for assets they can't reach now.
• A tariff-refund dollar does not deserve a new-sales multiple. S&P Q3 earnings growth is tracking near 30% on about 12% revenue growth. About $100B of an estimated $166B in tariffs has been refunded since the February Supreme Court ruling. Both hosts say how earnings are made matters. They split on whether the market will see through it, with one citing Micron as about 200 points below its post-earnings level.
• No sector should be shielded from AI; regulation belongs federally. One host calls the entry-level hit real, citing friends of their children in private equity. The other says they see only job creation and that graduates found work at 10–11% unemployment. Both oppose state rules like California's SB 951 (60-day notice). One would accept 60 days' notice, citing Meta and Google allegedly over-hiring and then running mass layoffs.
• Long straddles pay fair value; try them small and further out. The hosts argue liquid options are priced fairly by high-frequency market makers. The option market is usually right: Micron was priced for an $80 move and closed unchanged, crushing premium buyers. They suggest a small straddle in a later expiration, so that if the stock doesn't move it doesn't collapse overnight. One host says the expected moves are now too narrow to be short premium.

2026-10-12 · open the original ↗

• Checkmk's free edition can monitor nearly anything, with tinkering. The video is sponsored by Checkmk. Chuck installs the Community Edition package on an Ubuntu 24.04 VM, checks it with omd version, then runs sudo omd create monitoring and sudo omd start monitoring and browses to the server's IP at /monitoring. He warns the free version needs manual work that the paid editions automate, and that the server must sit where it can reach everything it watches.
• An installed agent plus service discovery shows far more than ping. On a Rocky Linux server he fetches the RPM agent with curl -O, installs it with dnf install -y, opens the firewall port, then runs service discovery: plugins find services to monitor, including a critical NFS service he hadn't known was down. Data is collected every minute and kept up to 4 years, with default thresholds for alerting. Network gear is found by a folder-level network scan using SNMP v2/v3 community credentials.
• Docker containers and Proxmox VMs need extra steps in Community Edition. Docker requires python3-docker, the mk_docker.py agent plugin and its config on the host, then a folder with no IP, no agent and "always use piggyback data", with container names imported from a CSV headed host_name. Proxmox uses an API "special agent" rule with a Proxmox user, nothing installed. A VM's host name must match Proxmox's to link the two.
• Plugins, active checks and local scripts let it monitor anything scriptable. He installs a community UniFi special agent from exchange.checkmk.com (mkp add, then mkp enable unifi 3.0.0, run as the site user), "at your own risk". His AI built a Plex check run as an integrated Nagios plugin; Home Assistant gets the same treatment. Hermes agent status comes from a local script that the Checkmk agent itself executes.
• Alerts can open tickets for AI agents, which can also run Checkmk. A Slack notification rule caught a stopped Hermes gateway in about a minute and reported its recovery. A second rule sent the same failure to Paperclip as a ticket, and an AI agent restarted the gateway. An automation-secret admin user let his main AI add his remaining hosts and build dashboards. Chuck says Checkmk Ultimate's free trial covers up to 750 services on one site.

2026-10-12 · open the original ↗

• Oil markets price barrels, not airports. Houthi strikes on King Khalid International Airport in Riyadh killed at least 12 and injured hundreds — the deadliest attack in a Gulf Arab state since the U.S.-Israeli war with Iran began in February 2026 — yet Brent barely moved from around $106/bbl. The author argues the thing to watch is Yanbu, the Red Sea end of the East-West pipeline meant to bypass Hormuz.
• The bypass cannot export its full rated capacity. The pipeline is rated at 7 million barrels per day, but about 2 million go to west-coast refineries, and the Yanbu port can load only about 4 to 4.5 million barrels per day onto tankers. The paper capacity overstates what can actually leave the country by sea.
• Houthis only needed to make the route dangerous, not win. After announcing a blockade of Saudi Red Sea oil in July 2026, they hit the Yanbu and Jizan refineries, and August pipeline flows fell to 2 million barrels per day, the lowest since January. The author draws the parallel to Ukraine against Russia and Iran against the U.S. in Hormuz: asymmetric pressure beats matching strength.
• Saudi workarounds only partly offset the double disruption. Eastern Gulf terminals and ship-to-ship transfers near Oman fill some of the gap, while Brent rose from ~$80/bbl to ~$106/bbl since August as both chokepoints tightened. The energy minister reports flows back to ~5.8 million barrels per day, but fighting on the Yemeni coast escalated just as they recovered.
• The Yemeni coastline matters more than the Riyadh airport. Most Saudi crude goes to Asia and must exit through Bab el-Mandeb, so the bypass "insurance" pays out only if that second strait stays open; retaking the coast with the Yemeni government would protect it. The author reads the deep strikes as timed to Houthi coastal losses, and repeats an earlier argument that commercial and legal factors can decide whether a strait is truly open more than the fighting does.

2026-10-11 · open the original ↗

• Washington is conditioning Ukraine's weapons supply on halting refinery strikes. Citing the Financial Times, backed by Reuters, he reports the Trump administration is threatening to block the mechanism Ukraine uses to buy Europe-funded weapons and to restrict intelligence support unless Ukraine stops hitting Russian oil refineries. He sets this against Russia having struck over 150 Ukrainian energy infrastructure sites.
• The announced energy ceasefire collapsed within hours. Trump said Russia and Ukraine had agreed to stop hitting energy targets; Zelensky said it was the first he had heard of it but Ukraine would stop if Russia did. Russia then struck the CHP-5 thermal plant, leaving it dark. He calls the announcement a fabrication.
• He reads the pressure on Ukraine as driven by US fuel prices. In his view Trump looks bad because diesel and gasoline are expensive, refuses to attribute that to Iran, and wants to lay the cost on Zelensky. He argues Trump's base would trade Ukrainian lives to save 25 cents a gallon on diesel.
• The sequence shows Putin getting relief while Ukraine gets punished. He traces it: Ukrainian refinery strikes hurt Russia's fuel industry, Putin seeks relief, Trump strikes a deal opening the door to Russian diesel restricted by sanctions and the Lindsey Graham-named law, then the fake ceasefire, then threats against Ukraine. He reads this as Putin again playing Trump.
• Blocking the sales costs the US money as well as credibility. He stresses these are weapons Europe buys from US manufacturers, not gifts — tens of billions of dollars spent in the United States. Cutting them off, he argues, loses American sales and allied confidence, degrades Ukraine's military, and gives Russia breathing room, which he says fails any traditional "America First" test.

2026-10-11 · open the original ↗

• Attacks on Saudi Arabia have moved to the capital itself. The host reports that on Saturday (the weekend of October 11, 2026) a missile hit a packed terminal at Riyadh's main airport, killing 12 and injuring more than 300 — the third hit on that airport in a week; the Houthis now say every Saudi airport is a target, and Trump says the US may join Saudi Arabia in striking back. His read: this is not a port or a tanker but the capital's airport, so the risk is moving closer to the center of the oil world.
• The Trump–Putin diesel deal has not stopped Ukraine's strikes. Hours after the deal was announced, Ukrainian drones hit Russian fuel sites again: a large fire at an oil loading terminal in Rostov-on-Don and a refinery/oil plant near Samara. Zelensky offered to stop hitting refineries if Russia stops hitting Ukraine's power grid; as of the video no one had accepted, and US–Ukraine talks in Miami ended early.
• Tanker attacks around Hormuz are continuing, not winding down. Reported over the weekend: a tanker leaving the Gulf hit by an unknown, unclaimed projectile late Saturday and set on fire (crew safe); a second tanker hit and damaged Sunday (crew safe); a crude carrier hit at anchor off the Emirates on Friday; the IRGC saying another tanker hit a mine; and the US disabling a commercial ship in the Gulf of Oman after it tried to leave an Iranian port. The host counts at least a dozen tanker hits since the start of October and at least 16 in September, and calls it "a new normal" rather than an ended crisis.
• Gulf Coast output lost to the hurricane still has to restart. The host gives Thursday's figure as the one that matters: about 1.3 million barrels a day shut in before the storm made landfall on the Florida panhandle, which he puts at 63% of all Gulf Coast oil. Restart requires crews to fly back out and inspect every platform for damage first, and he says it is not yet known how long that will take.
• Calm screen prices mask a disrupted physical market, in the host's view. His bottom line ties the four items together as paper versus reality — a diesel agreement on paper against drones in the air, tankers on fire, an attacked and shut airport, and oil fields waiting to restart — and he argues the price may look temporarily calm while the physical world is not.

2026-10-11 · open the original ↗

• Oracle's real leverage sits largely off the balance sheet. Notes and borrowings rose from ~$75B in late 2024 to $125.3B at 31 Aug 2026 (10-Q), plus $43.8B of on-balance-sheet leases and $288B of not-yet-started data-center leases (15–19 year terms, starting FY27–FY29) that rating agencies will count as debt as sites go live. FY26 free cash flow was -$23.7B; S&P projects about -$42B for FY27 on capex guided up to $95B, and Morgan Stanley projects ~$290B of long-term debt by 2028.
• The market already prices Oracle near junk. 5-year CDS closed at a record 261bp on 8 Oct (57bp in Sept 2025, ~144bp at start of 2026) versus ~55bp for the IG index; 2056 bonds trade near 82 cents (yield above 8.3%) and Oracle's own 10-Q puts its $125.0B of notes at $105.7B fair value. At 40% recovery the spread implies ~4.3% annual / ~20% five-year default probability, which the briefing stresses is a protection price inflated by hedging, not a forecast.
• Banks are stuck holding Oracle-linked project loans they cannot sell. The $38B Texas/Wisconsin Stargate package led by JPMorgan and MUFG took over six months to distribute, partly at a discount, and the $18B New Mexico Project Jupiter loan was quoted at 89–91 cents in September. Banks are using SRTs, discounted sales, sub-participations and Oracle CDS, but much of the risk moves to private credit and insurers financed by the same banks; the widely cited ~$30B JPM/MS exposure is unverified because no bank discloses single-borrower exposure.
• Uncertainty about Oracle's total need drives the spread, not debt size. Listed drivers include negative cash flow, the $288B lease overhang, OpenAI at about half of backlog (per S&P), and lenders buying Oracle CDS as a proxy for SPV loans; Oracle is reportedly ~$6.5B of a record $12.5B net CDS exposure on major tech names. Evidence offered: the February 2026 fixed $45–50B plan, half equity, cut CDS ~35bp despite including $25B of new bonds. Mitigants cited are $664B backlog, $36.4B cash and 30% Q1 revenue growth.
• The rating cliff is two notches away, not one. S&P is BBB- (cut 9 July 2026) but Moody's (Baa2, negative) and Fitch (BBB) are two notches above junk, and index rules use the middle rating, so forced IG selling needs two agencies at junk. A Moody's cut to Baa3 is the main trigger to watch; a full fallen-angel case would force $100B+ of index selling and hit SRT investors. Other watch items: Jupiter loan below 89c, CDS through 300bp, JPM/MS Q3 10-Qs and Oracle's December results.

2026-10-11 · open the original ↗

• Gulf data centers and undersea cables are now explicit military targets. He reports an Iranian warning that, with the US and Israel resuming the war, cables, data centers and AI facilities in the region will be destroyed, and stresses how large the Amazon and other data-center footprint is in Qatar, Kuwait and Saudi Arabia. He says he has warned for three years that underwater critical infrastructure would become part of military action.
• The reported size of banks' stuck Oracle debt is unverifiable. Articles say JPMorgan and Morgan Stanley have held ~$30B of Oracle-linked debt unsold for six months after underwriting; he argues no one can know the figure because filings never disclose single-borrower exposure — the Fed and Treasury know, but the Street would "drive them into the ground." If the figure were right and marked down ~15%, he estimates $4.5–5B of losses on bank books.
• Oracle's obligations are far larger than headline debt. He adds $125B of debt and $43.8B of on-balance-sheet leases to $288B of off-balance-sheet lease commitments, putting the total near $330B against $67B of equity. Q1 capex of $28.5B against $23.1B operating cash flow, which itself included $11.4B of customer prepayments he says must be repaid if capacity isn't built, plus capex rising to $95B, means continued heavy raising ($40B more this year).
• A downgrade to junk would force immediate selling across IG mandates. He frames the rating cliff as a key narrative, arguing from 2007–09 that rating agencies are "bought and paid for," including on private credit. If the market pushes them to junk, IG-only ETFs, mutual funds and corporate and municipal bond managers would have to sell on day one.
• Oracle's record CDS reflects SPV hedging more than default risk. At 261bp (vs ~55bp IG index) the implied ~20% five-year default probability is a protection price, not a forecast; his reading is that builders and lenders on SPV data centers backed by Oracle's rent promises have no direct hedge, so they buy Oracle CDS, pushing spreads up. His conclusion: Oracle has a lot of work to do but is "not as bad as it seems," and the CDS is not indicative.

2026-10-11 · open the original ↗

• The Strait of Hormuz is a war zone, not a working shipping lane. He shows a burning very large crude carrier and a photo from its stern in which he counts five to seven separate ship fires, taken the day before. He reads this as making official claims of 18–20 million barrels a day getting through implausible.
• Oil should open sharply higher, though he won't bet on it. He expects international markets opening that evening "should be up significantly" after the weekend. He also says markets are so manipulated, and the news so filtered, that he would not be surprised if they barely move, as has happened after earlier violence.
• The weekend marked the worst escalation of the war so far. He reports four oil and gas processing facilities in eastern Saudi Arabia, near Bahrain, struck the day before. A ballistic missile hit the passenger terminal at King Khaled airport in Riyadh, killing 12, one of them American, and wounding over 300. He says US coverage ignores this.
• Blaming Ukraine for high diesel prices misreads the timing. His chart shows diesel flat through the period Ukraine was striking Russian refineries, then surging after the US attacked Iran and the conflict became an infrastructure war. He argues Russia had already cut diesel exports, so the world could absorb that loss, while conceding both contribute.
• He expects Congress to kill Trump's diesel deal with Putin. He says the Ronald Reagan Peace Through Strength Act, introduced by Republican congressman Brian Fitzpatrick, has bipartisan support likely enough to override a veto. He also argues Trump authorized the Ukrainian refinery strikes, sharing intelligence, so Ukraine is not "defying" him.

2026-10-11 · open the original ↗

• Record stocks and multidecade-high bond yields coexisted, with fiscal slippage underneath. US equity indices hit new records on narrow tech momentum while several advanced economies saw multidecade-high government bond yields. Brent stayed above $100 a barrel on the US-Iran stalemate, and the US deficit for the fiscal year just ended came in around 6% of GDP despite growth of 3–4% (Atlanta Fed GDPNow) and a strong labour market.
• France, not the periphery, became the euro area's bond stress point. France traded at wider spreads than Italy, a historic inversion, with contagion unusually running from the core to the periphery. He suggests this may trigger a delayed political awakening, tested this week in parliamentary budget discussions and positioning ahead of next year's presidential election.
• US midterm politics is now steering foreign policy. He reads an unorthodox US-Russia diesel supply deal as driven by a domestic need to keep prices manageable amid declining energy inventories ahead of next month's midterms. Many countries, especially in Europe, were left puzzled at best.
• The Fed signalled more hikes, but not in October. The minutes and several officials acknowledged a pickup in already robust activity and pointed to likely additional rate hikes, while signalling consensus that a move is off the table for October. He attributes bond-yield pressure to volatile short-term Fed expectations and the shape of the US yield curve.
• A hot CPI print would upend the expected October pause. Consensus for September US CPI is core at 0.2% MoM (from 0.3%) and headline at 0.6% (from 0.4%), or 2.5% and 3.6% YoY. He warns a hotter print would fuel fixed-income volatility, with rate risk already leaking into credit risk. Also ahead: US PPI, Eurozone PMIs, UK data before the October 28 budget, Japanese cross-border securities flows, and the IMF's World Economic Outlook at the Annual Meetings in Thailand.

2026-10-10 · open the original ↗

• The Trump–Putin diesel deal is staged and capped, not immediate. As reported on 2026-10-10, Russia will release up to 4.8 million tons (about 36 million barrels) of diesel in waves: 300,000 tons immediately, 500,000 in November, then a million later. The US Treasury issued a sanctions waiver through April 2027, while Russia's own domestic export ban stays in place through October 31.
• The headline volume shrinks to a trickle once converted to flows. At roughly 7.5 barrels per ton, the opening 300,000 tons is 2.25 million barrels, or about 75,000 barrels a day averaged over a month. By the host's math, the full near-term package of 13 million barrels is less than half a day of global diesel demand and about 4 days of total US consumption.
• America's diesel problem is not a production shortfall. The host puts US refiners' diesel output at 5–5.2 million barrels per day against domestic demand of roughly 3.8 million, with about 1.5 million barrels per day already exported. His reading is that an empty global market is pulling every spare barrel overseas.
• The global gap is about 1.6 million barrels per day versus pre-war. Russia exported 800,000–850,000 barrels per day of diesel in 2025, but by September 2026 that had sunk to almost nothing because of drone strikes and the export ban. Shut Gulf refineries and product cargoes unable to transit Hormuz make up the rest of the hole.
• The host expects no pump-price crash from the deal. In his view, importing diesel into the US only reshuffles existing tanks without fixing the global deficit, since Middle East refineries are war-damaged, few product tankers are sailing through Hormuz, and China is only now restarting halted exports. He calls the deal a diplomatic headline and a temporary sanctions waiver, because Russia cannot export what its drone-damaged refineries cannot produce.

2026-10-10 · open the original ↗

• The diesel deal was Putin trading a fuel promise for sanctions relief. On October 1 Putin said Russia would supply no diesel to global markets until sanctions were lifted; eight days later Trump announced the deal and Treasury issued a broad six-month authorization for Russian diesel transactions. He reads diesel sales to the US as irrelevant to Putin's aims.
• Russia's diesel shortfall comes from its own export ban, not US sanctions. Russia banned exports itself as refineries came under attack, then deputy prime minister Alexander Novak called the domestic market "balanced". He cites the IEA for refining throughput falling from 6.5 to 3.8 million barrels a day, a drop of 2.7 million barrels a day.
• He believes Ukraine's claim of half of refining knocked out. His reasoning: Russia used to export about half its diesel, and a ban on all of it implies roughly half of production was lost. Continued large diesel imports from Belarus, he argues, show Russia is not making enough itself. Ukraine's figure is about half; another agency put it at 38%.
• Lifting sanctions does not repair refineries or create export surpluses. He argues sanctions relief only eases trade, so it was not the only way to get more diesel into the market. Besides the relief, he says Putin got Trump's own party to object publicly and got a law passed three weeks earlier reversed.
• The real prize was a wedge between Washington and Kyiv. The deal was announced while Ukrainian representatives were in Miami for peace talks, and Zelensky said his team had been used as a front. Trump had backed Ukraine's refinery strikes with intelligence earlier in the year but now urges them to stop. Zelensky had offered a mutual halt on energy strikes, and the license puts no such condition on Russia.

2026-10-10 · open the original ↗

• Dysprosium now trades as two commodities, separated by the licence. Oxide sells for roughly US$212–218/kg inside China (17 Sep) against US$3,250–3,500/kg ex-China; Benchmark puts Europe at 16.3x the China price, up from 6.8x at end-June. Licensed exports run about 30–40% over domestic (H1 avg US$312/kg), and the note warns that price services quoting ex-China near US$234–248 do not describe deliverable licence-free material.
• Western supply is constrained by feed, not separation plants. Separation is cheap (Energy Fuels' Utah expansion ~US$104M; Lynas Malaysia ~A$180M), but heavy-rich ionic clays are under 1% of deposits outside China and Southeast Asia. Listed projects reach roughly 700–900 t/yr Dy+Tb nameplate by 2028–29 against 2,800–3,500 t global output, perhaps a quarter on paper; the ex-China cost curve is viable only while the political premium lasts.
• China is forgoing rent to hold position, keeping a flood option. It keeps domestic prices low, meters exports case by case rather than banning, holds the 10 November 2026 expiry of its October 2025 controls as a recurring option, and entity-listed MP Materials and USA Rare Earth from 22 June 2026. The author projects (framework, not forecast) China holds the spread wide through 2027, then floods in 2028 to strand Western plants, as with Molycorp in 2015.
• Washington is answering a price problem with balance-sheet tools. The US$400M Defense Department stake in MP and loans to Energy Fuels still fail if China collapses the price; the author argues a state "cannot own its way out of a price that someone else sets." The US$110/kg NdPr floor and the July 2026 order ending non-allied magnet waivers from 1 January 2027 work, but no floor exists for dysprosium or terbium.
• Support should shift to floors, stockpiles and feedstock diplomacy. Proposed: a Dy/Tb floor set above the ex-China cost curve, a stockpile that buys when China floods, and securing clay feed in Brazil, Chile, Madagascar and Namibia, keeping equity as governance rather than insurance. Registered predictions: China extends or narrowly reinstates on 10 November rather than fully re-imposing; no Dy/Tb floor before Western heavy circuits hit nameplate; Europe/China multiple stays above 5x at end-2027.

2026-10-10 · open the original ↗

• Licence access, not cost, sets what the West pays for dysprosium. Reading from the "Two Markets, Two Models" briefing, he puts China's price at US$210–250/kg, approved foreign buyers at a 30–40% premium (about US$310–350), and unlicensed buyers at US$3,250–3,500. Europe pays 16.3x the China price, up from 6.8x in mid-June; he adds that Japan is getting nothing.
• The West is fighting the wrong battle with its funding model. China funds the sector as a strategic asset needing no return, while Western firms must earn one on government equity and loans that leave taxpayers exposed to the risk that actually kills the asset: a price China sets.
• Western feed is too dilute to compete on cost. China's ionic clays hold about ten light units per heavy, and he says Western feed is far more dilute. He guesses Western costs at roughly US$800–1,000/kg against China's ~US$200, and says announced capacity overstates heavy output. He notes the US lifted sanctions on Myanmar's military government days earlier and implies a supply motive; that link is his speculation.
• China can strand Western plants by flooding the market. He expects Beijing, once plants come online around 2028, to release supply and crush prices, as it did to Molycorp. Entity listings already deny MP Materials and USA Rare Earth Chinese-made plant equipment, and he says Chinese rare-earth staff are barred from contact with the West. He closes by asking whether the Western approach should change.

2026-10-10 · open the original ↗

• Insider-risk controls cannot tell whether an authorized decision is correct. The author accepts Satya Nadella's essay "Models as Insider Risks in the Super Intelligence Era" (restrict privileges, monitor, keep the ability to intervene, separate a model's intelligence from authority to act), but argues security and decision correctness are different problems. In the author's hypothetical, a credit-portfolio agent acts entirely within its permissions with a full audit trail yet recommends a reallocation that breaches concentration tolerance, perhaps through a misread constraint, a bad correlation assumption or ignoring stressed losses; an audit trail records what it did, not whether it should have, and a kill switch helps only if something notices.
• Verification must be engineered, not handed to another LLM. The author's proposal: the LLM proposes, but acceptance criteria sit outside its control and are checked by calculation, authoritative records, validated statistical models and optimization (for feasibility and better alternatives); "there is no reason to ask another LLM whether a numerical constraint has been satisfied when we can calculate it directly." Where Nadella stresses transparent reasoning, the author prefers independently reproducible evidence, since a chain of thought may not faithfully explain an answer. The concession: the checks themselves can faithfully implement the wrong constraint, so they need testing by people who understand the maths, data and business.
• Automating junior work erodes the expertise that independent validation needs. Drawing on their Detroit engineering and model-validation career, the author argues judgment is built by doing the work, and validators need technical competence, not just a separate reporting line. The predicted path: fewer junior hires, survivors reviewing AI output instead of building analyses, seniors retiring, until validation becomes procedural; the author imagines AI validating an AI-built model, writing the report, and a committee approving it from AI summaries. The risk is not that AI work is wrong but that firms lose the ability to tell excellent work from "plausible nonsense".
• Productivity gains are measurable; expertise erosion is not, so it compounds. Firms consume expertise built under the old system without replenishing it, and as verification gets harder they automate it further, a reinforcing cycle of dependence the author says is not inevitable, since AI could instead speed up learning if work is organized deliberately. The illustration: an analyst spending three days on a model's odd behaviour looks less productive than one getting a polished AI explanation in three minutes, yet learns more. The author sees the same risk scaled up as a "governance paradox": ever more policies, dashboards and AI monitoring AI around systems fewer people understand.
• The author prescribes four practical responses to keep institutional competence. (1) Separate generation of candidate actions from evaluation and authorization, with criteria the proposing AI cannot change, and reserve humans for judgment, unresolved uncertainty or out-of-limit cases, since blanket human approval does not scale. (2) Redesign junior roles to build judgment: use AI to explore alternatives but still understand the calculations, reproduce results independently and investigate disagreements with established methods, with real mentoring. (3) Keep in-house technical capability to challenge vendor and foundation-model systems, because relying on vendors to vouch for reliability "outsources" the ability to judge. (4) Measure competence alongside productivity through periodic challenge exercises: unfamiliar problems, planted errors, reproduction by alternative methods.

2026-10-10 · open the original ↗

• Cartel-linked gold has reportedly entered the US Mint's supply chain. She cites a New York Times investigation from last April: Colombian cartels running illegal mines laundered raw gold through middlemen, foreign pawn shops and third-party refiners until, refined to .999 fine, it was virtually untraceable. Follow-up reporting, she says, found the Mint's refiners also took metal from pawn shops in Mexico and Peru, a destructive Honduran mine, and a Chinese-run operation in the DR Congo.
• The Treasury has refused to certify the coin's gold as domestic. Federal law requires bullion coins such as the American Gold Eagle and the new Trump coin to be struck from US-mined gold. Congress demanded a guarantee in June for the 250th-anniversary release; Treasury and Mint officials declined any legal certification or mine-to-coin tracking, arguing that cutting off foreign refiners would create bottlenecks making it nearly impossible to meet demand.
• She concludes the Trump coin will be struck from illicit gold. The 24-karat proof coin launches November 5th; her claim that it is made from illicit foreign gold is her own inference from the sourcing reports and the Treasury's refusal, not a finding she cites. Her stated position is that she will buy no gold from the US Mint until it guarantees compliance with the sourcing law.

2026-10-10 · open the original ↗

• The Russian diesel deal contradicts months of punishing others' purchases. He notes Trump threatened countries, and imposed tariffs on India, for buying Russian energy that funds the war on Ukraine, then on October 9 announced a personally negotiated deal to put Russian diesel back on US and global markets. In his reading, only one thing changed: US diesel above $6 a gallon (he recalls a peak near $6.52) with an election coming.
• The license undercuts sanctions law Trump signed three weeks earlier. On September 18 Trump signed the Lindsey Graham Sanctioning Russia and Iran Act of 2026; Section 112 allows tariffs up to 500% on Russian imports including petroleum products, and Section 113 targets countries buying Russian crude or gas. Treasury's General License 135 now authorizes Russian diesel transactions, including US imports, for six months to April with no tonnage ceiling. He concedes the act has waiver provisions but wants congressional scrutiny, and notes Republican Rep. Don Bacon has criticized the deal.
• Ukraine's refinery strategy now collides with US fuel-price policy. He says the announcement ties further supply to the condition of Russian refineries, roughly half of which sources say are offline after Ukrainian strikes. He raises, without claiming it is a condition of the deal, whether Ukraine could be pressed to stop those strikes; Zelensky has condemned the decision and called Trump weak. European allies that paid to cut Russian petroleum are left without a unified sanctions front, and the US cannot override their sanctions.
• Russia could backfill exported diesel with Iranian supply. With its domestic refining under pressure, he suggests Russia may replace exported diesel from Iran or other Middle Eastern sources, and possibly already takes some from Belarus. He acknowledges no verified evidence of this and that Iran has its own shortages, but argues Iran would sell at full price if it could, and wants the administration to explain how it will monitor indirect trade.
• The volumes are too small to move pump prices. Trump announced 300,000 metric tons for October and 500,000 for November; he puts October at about 2.25 million barrels, which on a daily basis is about 5% of US diesel exports, so the US exports 20 times as much each day. Diesel futures fell on the announcement and came straight back once the terms were understood, and he expects no relief at the pump for the credibility spent.

2026-10-10 · open the original ↗

• Headline-driven swings are shaking retail out of good positions. He describes the week as a mid-week scare followed by a rise into the close, and repeats his standing advice to ignore the news where the portfolio is concerned.
• Many reader-requested stocks currently rate as buys. Of 27 tickers ranked, 13 were marked "buys" for strong established trends, good fundamentals and high relative strength, many with imminent bullish chart setups; one ranked top. The available text does not name the tickers.
• He bought a Midweek Movers pick just starting its breakout. His rationale is the series' record: he reports its picks have averaged a 25.9% gain each this year. The stock is not named in the visible text.
• The quiet week reads to him as consolidation before another leg up. He takes the absence of anything alarming as a positive sign that the market is still coiling, which is his interpretation rather than a stated forecast with levels or timing.

2026-10-10 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-10 · open the original ↗

• This short makes no market or policy argument. The 16-second clip's captions are eight words, a greeting called out to someone and an exclamation ("Oh my goodness"), with no prices, volumes, names of officials or claims about energy markets to report.

2026-10-09 · open the original ↗

• The Russian diesel deal adds nothing to the global market. Responding to a clip of Trump thanking Putin for "massive amounts" of diesel, he argues the US has no diesel shortage — the world does — and since diesel and gasoline are priced globally, US inventories merely reflect global ones. Redirecting existing Russian barrels to the US adds no new supply.
• The volume is trivial against what the US already exports. He puts the Russian diesel at roughly a twelfth of what the US exports daily, shown on a bubble chart against monthly US diesel consumption, and predicts it will do "less than nothing" to lower diesel prices.

2026-10-09 · open the original ↗

• Split tabular prediction into a learned comparison rule and a linear solve. The author and Aijun Zhang (paper: "Learning Spectral Geometry for Kernel Regression", SSRN) learn a kernel that sets how rows are compared, then fit a regularized kernel ridge regression given that kernel; a tree ensemble fits both at once, and its splits mean nothing on a table with a different schema. The stated motive is reuse: learn the geometry once and need only a linear solve on a new task. They concede the paper tests this only within single datasets, and cross-table transfer (which needs a schema-agnostic representation and a cross-task objective) is still to be shown.
• Each learned component handles scale, direction or decay of similarity. Each numerical input becomes K sine-cosine pairs with trained, ordered frequencies (drawn from Bochner's theorem, but as a learned representation rather than random or spectral-mixture features). A positive-definite metric B weights the pairs, and its cross-feature terms let joint changes (location with income) count differently and make the kernel nonstationary. A Laplace kernel with learned bandwidth T was chosen over the Gaussian for three reasons: it is positive semidefinite in any dimension (Schoenberg), it suits irregular response surfaces, and it is nonadditive, so interactions come without specified interaction terms.
• A canonical parameterization makes the fitted geometry readable and comparable. Free amplitudes, metric and bandwidth are redundant: rescalings leave every prediction unchanged, so feature weights have no defined value. Fixing the metric to unit trace via a normalized matrix exponential removes this, and pair and feature "masses" can then be read off. In the California Housing model, latitude takes about 45% of the trace and longitude 31%, put mostly on their highest frequency, while income and house age sit on low frequencies. Sensitivity scales with frequency squared, so location is resolved finely and income is treated as slowly varying.
• On California Housing, the exact ridge solve mattered as much as the geometry. Geometry is trained episodically (fit ridge on a support set, score on a disjoint query set, backpropagate through the solve), then a final exact ridge fit is done on all training rows. Across three 80/20 splits the refined model's RMSE is $820 below tuned CatBoost (1.90%), lower on every split ($1,107, $1,001, $351). Swapping a 2,048-landmark Nyström approximation for the exact solve cut about $1,700, roughly two fifths of the total gain; refining the geometry cut about $2,372 more and full-pool refitting about $78.
• Results on mixed and binary data are mixed, with CatBoost winning Adult. Categorical fields enter as fixed orthonormal contrasts under one shared metric, and binary scores are sigmoid-calibrated, as are CatBoost's. The spectral model has lower RMSE on Bike Sharing on all three runs and AUC higher by 0.003–0.006 on Taiwan with log loss within 0.0011, but CatBoost is more accurate on every Adult measure. Code, checkpoints and predictions are at github.com/asudjianto-xml/Tabular-Spectral-Kernel (v0.1.0), with separate verify, replay and full-training reproduction levels.

2026-10-09 · open the original ↗

• The Trump–Putin diesel deal is too small to move prices. As announced on October 9, Russia is to supply 300,000 tons of diesel for all of October, 500,000 tons for November and a million tons at some later point. He calls October's volume about 18 minutes' worth, essentially one cargo, and about 5% of daily US exports; November's is about 30 minutes' worth, roughly 7–8% of a day's exports.
• The deal runs backwards on supply: Russia is short, the US is not. He says Russia is in a massive diesel shortage while the US is exporting record amounts of diesel, which makes the US buying from Russia strange on its face.
• Quoting the deal in tons may be meant to obscure its size. He suspects the announcement used tons and metric tons to confuse a US audience that measures fuel in barrels — his own reading, not something stated by the administration.
• Delivery itself is doubtful. He notes Russia has largely banned diesel exports and that Ukraine struck Russia's largest refinery the day before, and argues Zelensky could "crater" the deal with a single drone. Either way, he predicts it will not make US diesel one penny cheaper.

2026-10-09 · open the original ↗

• The "governments will force money home" narrative is settled. He says 42 countries have now instituted multiple measures to make capital come back, so the question is answered; he replaces it with a new narrative tracking how punitive governments everywhere, not only the US and China, will get in trapping money at home.
• Corporate debt is now competing with, and outranking, sovereign debt. As one data point, he cites 38% of France's corporate bonds being rated less risky than French government debt. In the US, he argues, hyperscaler debt competes directly with Treasuries and now yields more; he asks how many US corporations would rank above the sovereign had they not borrowed "like drunken sailors" for AI build-outs.
• Italian bond yields are low only because of the EU guarantee. He counts up to 17 subgroups now forming within Europe and says no one is in control; his inference is that without an EU backstop, Italian bonds would trade on their own credit at around 15%. He illustrates national distrust with a crisis-era bank that had to be sold in three national pieces.
• Artificially drawn "construct countries" are likely to break apart. Discussing a commentator's map predicting partitions of Iran, Syria, Iraq and Yemen, he disagrees with many specifics (including on Israel) but agrees such borders will shift. He endorses the forecast that the next five to seven years will be fascinating and the following seven to ten terrible to witness.
• Private credit and private equity marks cannot be trusted. He exited the asset class about three and a half years ago after then-SEC chair Gensler's proposed quarterly third-party valuation rule was struck down by a court in New Orleans; managers charge 2–3% fees on self-reported values. He cites a Semafor report that Kellermeyer Bergensons Services loans held by a KKR private credit affiliate were marked from near 100 cents to 10 cents while its equity was written to zero, and notes an SEC warning to auditors.

2026-10-09 · open the original ↗

• Hurricane Isaias has knocked most Gulf oil output offline. As of October 9 it was a category 3 storm with 120 mph winds, confirmed by Hurricane Hunter planes, with landfall expected that night or early Saturday between Gulf Shores, Alabama, and Destin, Florida. As of Thursday, October 8, almost two-thirds of Gulf oil production (about 1.3 million barrels per day) and 57% of Gulf natural gas production were shut in.
• US-Iran talks briefly pushed oil down before hurricane risk returned. Trump said on October 8 that the US is having productive discussions with Iran and ruled out any attack before the November 3 midterms; Iran says it is reviewing the US response to its proposal to reopen the Strait of Hormuz within 7 days. On the news WTI fell from $93 to $91 and Brent to about $103, before hurricane pricing came back.
• Ukrainian drone strikes on Russian refineries are feeding US diesel prices. That week Ukraine hit one of Russia's largest refineries (440,000 barrels a day, 2,500 km from Ukraine, its second hit this year) plus the Salavat refinery and three missile sites, and Russia struck Ukrainian energy sites in return. US diesel was at a record average of $6.28 a gallon as of October 9, up 70% since the Iran war began; the host's reading is that drones and Middle East missiles both show up in what drivers pay at the pump.
• OPEC's flat November targets mask a large physical shortfall. The seven big members pumped about 25 million barrels a day in August, 5 million below pre-war levels, with Russia and Saudi Arabia most of the miss, and Gulf exports running at about 60-80% of normal. The host's read: targets are paper, barrels are physical, and the gap between them is "the whole story of 2026".
• Hormuz restrictions shrank global LNG trade in August. A gas exporters' group reports world LNG trade fell almost 9% year on year in August as Persian Gulf supply was choked; Asia's imports fell hard while Europe barely grew, with American cargoes swinging back to Europe as the Asia premium shrinks. The host's bottom line is that screen prices and pump prices diverge, because physical prices are set by the storm track, drone range and shipping.

2026-10-09 · open the original ↗

• Iran's cheap drone threats are beating America's expensive arsenal. He reports the Air Force pulled a dozen B-1 bombers from RAF Fairford after British authorities foiled an Iranian drone plot, which University of Birmingham professor David Dunn told Time is an unprecedented setback; Iran's IRGC uses "violence-as-a-service," hiring untrained, disposable gig-style operators. He argues the drone war has drained U.S. interceptor stocks, prompted a $95 billion multiyear munitions request from the Trump administration and made U.S. victory "a mathematical impossibility," citing Spencer Ackerman that the U.S. is far from being able to force the Strait of Hormuz open and keep it open.

• Asymmetric innovation gives an army a lead, not a moat. Ukraine, now the leading drone innovator, moved the enemy casualty ratio from 2:1 at the war's start to 8:1 in the first half of this year, with Russian casualties around 30,000 a month against 27,000 monthly recruits. But historian Timothy Snyder told him Russia copies every Ukrainian innovation within months "at scale" and is exploiting an interceptor shortage (which Galloway calls America's fault) to hit Ukrainian cities — the military version of the business second-mover advantage (Microsoft over Netscape, Visa over Diners Club, Facebook over MySpace, Apple generally).

• Great powers that spend more on debt than defense decline. He cites Niall Ferguson's 2025 "Ferguson's Law" (named for 18th-century philosopher Adam Ferguson), noting U.S. interest payments — the fastest-growing budget line — now exceed defense spending, and that sovereign debt crises ended empires from Rome to the Soviet Union; Eisenhower's 1961 farewell address and Admiral Mike Mullen in 2010 (debt as the "most significant threat to national security") made the same warning. His view is that defense cuts are a political third rail because leaders fear looking unmanly, a reflex he blames Defense Secretary Pete Hegseth for feeding.

• Deep Pentagon cuts are necessary, and the options already exist. He calls for $200 billion a year in cuts and frames the real metric as the fiscal gap, which Treasury puts at 4.7% of GDP — closing it by spending cuts alone would mean a 21% cut to every non-interest dollar — against FY2026 defense spending of about $1 trillion ($838 billion appropriations plus $150 billion from the One Big Beautiful Bill) and a presidential request to raise it 45% to $1.5 trillion, above the inflation-adjusted World War II peak. Precedents and options he cites: a 30% defense cut from 1988 to 1997, a 2024 CBO menu of roughly $100 billion a year (18% smaller active-duty force, fewer carriers and fighter squadrons, more allied burden-sharing), a watchdog's ~$40 billion in weapons-program cuts, and Palmer Luckey's claim that ending cost-plus procurement could bring the budget below $500 billion — plus swapping $90 million fighter jets for $30,000 drones.

• Casualty aversion plus defense-industry capture built an expensive glass jaw. He contrasts adversaries' willingness to sacrifice soldiers with the U.S. bias toward safety — at least 22 servicemembers dead in the Iran war, and since 1980 accidents, suicide and homicide have killed 7x as many active-duty troops as enemy fire — and says presidents will approve trillion-dollar systems but not LBJ-scale losses (over 500 killed per week at Vietnam's height). His conclusion is that the current conflicts are an opportunity to build a more lethal force while cutting a trillion dollars, with a strong balance sheet as the real "superweapon."

2026-10-09 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-09 · open the original ↗

• The proposed bond rollback cuts security, not cleanup cost. The Trump administration proposes lowering the federal onshore minimum bonds set under Biden in 2024 from $150,000 per lease and $500,000 statewide to $10,000 and $25,000, cuts of about 93% and 95%. He stresses it is a proposal, not yet in effect, and that the plugging obligation and its cost are unchanged.
• A $25,000 statewide bond would not plug even one well. Statewide bonds cover all of an operator's wells in a state, and he says operators almost always use them. Against the BLM's 2023 average of $71,000 per well to plug and reclaim on federal land ($30,000 to $200,000 in his experience), seven wells cost about $497,000, leaving a $472,000 gap.
• Federal-land exposure alone runs to billions. He cites a 2026 Taxpayers for Common Sense analysis of about 92,000 producing federal wells: roughly $6.5 billion to plug at $71,000 each, shifting about $6.2 billion to taxpayers at a $25,000 average bond. Federal land is just over 20% of US production; a 2020 national estimate put onshore closure costs near $300 billion, with guarantees covering about 1%.
• Operators routinely use bankruptcy and field sales to dump plugging liabilities. In his account, companies distribute cash to shareholders, then file bankruptcy behind LLC protection; majors sell aging fields cheaply to small operators — e.g. a $50 million field carrying $20 million of plugging liability — who defer plugging, produce, then abandon. He calls weakening bonds contradictory after the infrastructure law committed nearly $5 billion to orphaned wells.
• Bonding should match actual plugging obligations, and he expects passage anyway. The administration argues high bonds burden small companies and that the BLM can demand more through risk reviews; he doubts agency staffing can enforce those and says the cut would barely affect production. He argues anyone who can spend $10 million drilling can set aside $50,000–75,000, and predicts the rollback will pass.

2026-10-09 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-10-09 · open the original ↗

• A silver dollar kept its purchasing power; a paper dollar lost it. The speaker's illustration: in 1926 a bakery loaf cost about 10 cents, so any dollar bought 10 loaves. Today the loaf costs $5, so a paper dollar cannot buy one, while a 1926 silver dollar sold to a coin shop still buys about 10. The figures are the speaker's round numbers, with no coin or spot price given.
• Gold has outperformed silver over the same century. By the speaker's account, a 1926 gold dollar coin cashed in today would buy at least 40 loaves, about four times its original purchasing power. The video closes by asking viewers which they would save in, an advocacy framing rather than a stated forecast.

2026-10-09 · open the original ↗

• Cheap code moves the bottleneck from execution to judgment. In the episode 1023 clip, Srinivasan argues that what once took weeks to build mindfully now takes 10 minutes mindlessly, so speed is no moat; intent about what is built, for whom and how it scales still is. She says agents must be evaluated end to end, from user input through every tool call and failure, plus traditional security cases such as DoS, unauthorised logins and prompt injection, because non-determinism compounds at every "joint".
• Attention reads as curved space, not a lookup table. In the episode 1025 clip, Serrano says he came to see attention as words moving along lines towards each other ("word gravity"), as "river" pulls "bank" into the nature region of the embedding. A physicist at a Toronto talk recast it as space-time curvature, with the key, query and value matrices appearing in the geodesics. Serrano calls it a loose, mainly visual analogy, published last November with that physicist and a York University professor.
• Managing people turns out to be training for managing agents. In the episode 1029 clip, Malone says defining tasks and acceptance criteria, switching between independent work streams and "trust but verify" carry over directly to agentic AI. Non-technical managers may know more than they realise, while strong engineering ICs who hate reviewing output and losing flow face the problems managers already have.
• Sub-agents and rising capability drive token use up regardless of price. In the episode 1031 clip, Shah says a Pokémon-style anniversary game starring his dogs consumed billions of tokens in a weekend because agents spawn sub-agents that enlarge the pie of work. He invokes Jevons paradox: cheaper tokens don't matter if usage keeps compounding. The host says METR's autonomy chart has gone "off the charts" since Mythos-class models, because no human baselines exist for tasks of dozens of hours.
• An always-on family agent needs a different product playbook. Kahawala, CEO of Anna, says she has had to throw away a decade of product practice; her three hardest problems are having no UI, users unused to "just ask" agents, and reliability on household data that coding-optimised models aren't trained on. She says most always-on agents get abandoned after about 2 weeks, and that voice mode, where Anna must queue several fanned-out tasks and interject, only became workable when Gemini Live improved about 3 months before recording.