The long end is now a war-risk premium sitting on top of a governance premium, and this weekend is where the two meet. The 30-year Treasury closed Friday at 5.27% and the 10-year at 4.75%, both fresh multi-decade highs, while the 2-year fell to 4.28% — a bear-steepening of the curve, meaning long yields rising faster than short ones. That shape says the market is not pricing a hotter economy at the front end; it is pricing a longer-dated risk it cannot hedge cheaply. Two candidates now stack on top of each other: a Federal Reserve committee that Chair Kevin Warsh has freed to vote its own views (thesis 4), and, as of Friday evening, reporting that the President has ordered a bombing campaign against Iran's ENERGY infrastructure that could begin this weekend (thesis 3). The 2s10s curve at +0.47% has steepened 12 basis points in two sessions.
The front end is still not restrictive, which is the part most commentary gets wrong. The effective fed funds rate is 3.63% against headline CPI of 3.73%, so the real policy rate — the interest rate after subtracting inflation, and therefore the actual price of money — is MINUS 0.1%. Policy is not tight; it is roughly free. Meanwhile the 2-year at 4.23% trades 60 basis points ABOVE the funds rate, which is the bond market's mechanical way of saying the next move is a hike rather than a cut. That single spread is why the hike case survives soft inflation prints.
Inflation's level and its momentum still point in opposite directions, and the hawks are voting the level. Headline CPI runs 3.73% year over year, but the 3-month annualized pace is 2.78% and sticky-price CPI — the slow-moving basket that reflects wages rather than commodities — is 2.81%. Momentum is nearly back to the 2% target; the level is not. Expectations are calm: 10-year breakevens, the average inflation rate the bond market expects over the next decade, sit at 2.28%, and 5-year-5-year forwards at 2.30%. The 10-year REAL yield at 2.41% — the nominal yield minus expected inflation, the true cost of borrowing — remains near the highest in two decades. Real yields at extremes against ordinary expected inflation is what deliberate tightening looks like, not a loss of credibility (thesis 5).
Labour is the disconfirmer, and it moved the wrong way for the bears. Initial jobless claims fell to a 202,750 four-week average from 207,500 a week ago; almost nobody is being laid off. The 3-month payroll average runs 111,000 a month, unemployment is 4.2%, and the Sahm rule indicator, which flags recessions when unemployment rises half a point off its cycle low, sits at 0.07 against a 0.50 trigger. There is no labour-market evidence for a recession, which is precisely what keeps the hike case alive. A negative payroll print is the single number that kills it, and the next one lands Friday.
The consumer is genuinely splitting, and this run the split narrowed rather than widened. The buffer is still gone: the personal savings rate is 2.7%, a cycle low, against real income growth that is roughly flat, so consumption growth is funded by drawdown by definition — spending is holding up (retail sales +6.7% nominal, durable goods +7.6%) because households are running the buffer down, not because they got richer. Gasoline at $4.10 is a direct tax on the use decision that precedes any big-ticket purchase, and it is the channel most exposed to what happens in the Gulf this weekend. But two repair signals are now live rather than one: University of Michigan sentiment jumped to 49.5 from 44.8, a second consecutive improvement off a record low, and credit-card delinquencies at 2.92% continue to improve. This desk requires three consecutive months before treating sentiment repair as real. The stressed-not-defaulting read holds, and the household view (thesis 8) stays at 70% because the gas-price channel arming offsets the sentiment repair, not because nothing happened.
Housing tightens the same screw through the long end. The 30-year mortgage sits at 6.66%, up 8 basis points on the week and at a one-year high, and because mortgages price off the 10-year Treasury plus a spread — currently a wide 1.98% — a long end breaking out on governance and war risk pushes the mortgage AWAY from the 5.5% entry gate structurally rather than cyclically. The gate is now 116 basis points away. Case-Shiller national house prices rise 1.11% against 3.73% inflation, so real house prices are falling about 2.6% a year: waiting costs a levered buyer nothing while saving them negative day-one carry. Building permits at 1.374M lead housing starts at 1.427M, so supply is softening ahead. Commercial real-estate loan delinquencies at 1.56% remain the channel to watch for genuine distress (thesis 9).
Credit still refuses to confirm any stress, and that is the strongest argument against every bear on this board. High-yield spreads — the extra yield junk-rated borrowers pay over Treasuries, and the earliest broad warning this desk tracks — sit at 2.84%, near record tights, with investment grade at 0.80%. The Chicago Fed's financial conditions index is -0.55 (negative means loose). The VIX closed at 15.99, down 6.4%, after a week that felt considerably worse than that. Whatever is happening in the long end is not yet a credit event. The one liquidity tell that did deteriorate: net liquidity — the Fed's balance sheet less the Treasury's cash account and reverse repos — fell to $5.82 trillion from $5.92 trillion, a $100B drain in two days, which is Warsh's stated balance-sheet withdrawal showing up in the plumbing.
Commodities carry the weekend's asymmetry. Brent closed $90.15, up 3.76%, and WTI $87.18 — still 2.0% BELOW the $89 confirmation gate this desk pre-committed to before adding oil exposure, which is why the below-support trigger row stays TRIGGERED even as the geopolitical risk rises. The US strategic petroleum reserve stands at 308 million barrels, the lowest since March 1983, after an 18th consecutive weekly decline totalling 108 million barrels, or 26%; commercial crude excluding the reserve is 405 million barrels, the lowest since October 2018 and 7% below the five-year seasonal average. There is no inventory cushion to absorb a supply event, and a supply event is what Friday's reporting describes. Gold at $4,107 holds the $4,000 floor while down roughly 26% from its February high, resilient given where real yields are (thesis 5). The broad dollar index at 120.71 stays inside its 117.45-to-122.70 range, so the debasement is still being paid in gold and bitcoin rather than in the currency. Bitcoin at $62,528 (live, Saturday afternoon) now sits 0.4% BELOW the $62,791 repair line, having been 0.3% above it at the morning read; the deciding weekly close prints Sunday evening, with the valuation model at -1.05 standard deviations below trend and the Fear and Greed index — a sentiment gauge where 0 is maximum fear and 100 maximum greed — at 27, "Fear."
One structural point the front-month oil price hides: this is a duration problem, not a spot problem. The barrels at risk from a strike on Iranian production and export infrastructure are largely barrels that already are not reaching the market, because the Strait of Hormuz is already disrupted. The difference between the two scenarios is repair time, and it is large: a closed strait that reopens resolves in roughly three months, while destroyed production and export infrastructure takes twelve to eighteen months to restore (Philip Pilkington, economist, on the Nawfal panel; the empirical anchor is the Houthi strike on a single Saudi refinery, which needed four to five months of repair and produced a 6% spike that round-tripped). The practical consequence for reading this board: the $89 WTI confirmation gate is a front-month number, and front-month prices are the part of the curve a strategic-reserve release or a wave of speculative short-selling can most easily suppress. If a campaign actually lands, the honest tell will be the shape of the futures curve twelve to eighteen months out rather than the spot print, and the front month may confirm late or not at all. That is a reason to watch the curve, not a reason to move the gate.
What to watch next: whether a strike lands before Asian markets open Sunday evening New York time. CBS reported that officials discussed wrapping the campaign up before markets open Monday, which makes this a gap-risk weekend rather than a slow-burn week. If energy infrastructure is struck at scale, WTI clears the $89 gate on the open and the $107 shock gate becomes the live question; if the reporting proves to be negotiating leverage and no strike lands, the same inventory math means the downside is shallow. After that: JOLTS job openings Tuesday (7.25M expected against 7.59M prior), ISM services and the MBA mortgage rate Wednesday, unit labour costs Thursday (2.7% expected from 1.8%), and Friday's payrolls, consensus 91,000 against 57,000 prior with unemployment expected to tick up to 4.3% — the named disconfirmer for the entire hike case.
This is a gap-risk weekend, and that is the whole briefing. Friday evening the Wall Street Journal reported that the President has ORDERED a multi-day bombing campaign against Iran's energy infrastructure, beginning as soon as this weekend; CBS reported the same preparation and added that officials discussed wrapping it up before markets open Monday; Axios reported it as still under consideration with no final order given; CNN described a two-week campaign against missile sites rather than energy. The four accounts disagree on the verb and the target, which is what pre-strike signalling normally looks like. What does not disagree is the inventory position underneath it: the US strategic petroleum reserve stands at 308 million barrels, the lowest since March 1983, after eighteen consecutive weekly draws totalling 108 million barrels, and commercial crude excluding the reserve is at its lowest since October 2018. There is no cushion. Oil (thesis 3) rises from 52% to 57%. Separately, the Bitcoin self-custody story turned into a real loss event: a five-year-old firmware flaw let an attacker drain 594 bitcoin, about $38 million, from roughly 500 Coldcard wallets in a twenty-five-minute sweep, and a contested protocol change reaches its forced-signalling deadline on August 7 with miner support at 2.4% against a 55% requirement. Bitcoin's repair (thesis 7) falls from 65% to 63%. Everything else held: the AI building boom stays 68%, memory chips 71%, the Fed's hike lean 68%, rotation out of the giants 72%, the household squeeze 70%, and levered residential real estate 70%. An afternoon re-run added four more transcripts and moved nothing; see the section immediately below for why that is a finding rather than a gap.
Four more transcripts landed after the morning run, three of them on the Iran question, and no probability moved on any of them. That is the finding, not an absence of one. The new material is commentary on the same Friday-evening reporting the morning run already priced, and this desk does not pay twice for one news cycle. It is also genuinely two-sided, which is the more useful part:
Oil (thesis 3) therefore holds at 57%, deliberately. The upgrade case (Weichert's 80%, tankers already burning, a maximalist Iranian demand, an open cyber front) and the downgrade case (a CENTCOM commander saying he lacks the ships, an Israeli official telling CBS Israel has not been asked to join, a documented false-start record) very nearly cancel. Bitcoin (thesis 7) holds at 63%, with one live change worth knowing: spot has slipped to $62,528, which is now 0.4% BELOW the $62,791 weekly-close repair line rather than 0.3% above it as at the morning run. Bitcoin's weekly close prints Sunday evening, so that line resolves within roughly thirty hours. Theses 8 and 9 were re-scored against the same macro file (no new prints land on a Saturday) and hold at 70% each.
Oil (thesis 3) rises from 52% to 57% for a spike by September. This moved on a category change, not another headline. Every prior escalation in this war was struck infrastructure inside the Gulf states, tanker attacks, or missile exchanges with military targets. Friday's reporting is the first in which the target set is Iran's own energy production and export capacity and the decision is attributed to the President rather than to planners. The Wall Street Journal's account says ordered; Axios's says considering, no final order; CBS's adds the pre-Monday-open timing constraint; CNN's names missile sites and a two-week window instead. Treat "ordered" as unconfirmed and treat "prepared" as certain. The reason this is worth five points rather than one is what sits underneath it. US crude inventories have no slack: the strategic reserve at 308 million barrels is the lowest since March 1983 and has fallen for eighteen straight weeks, down 26% over the streak, and both speakers on the Mario Nawfal panel flagged 300 million barrels as the level below which the remaining oil is a quality and usability problem rather than a usable buffer, which at the current draw rate is roughly a week away. Commercial stocks ex-reserve at 405 million barrels are 7% below the five-year seasonal average. Add the positioning: speculative gross SHORT positions in West Texas Intermediate sit near a five-year extreme around 228,000 contracts, so the marginal seller has already sold and a supply event forces them to buy back. What caps this at 57 rather than 65: WTI at $87.18 is still 2.0% BELOW the $89 gate this desk pre-committed to before adding oil exposure, so the tape has not confirmed; the war has produced a mutual pause roughly every three weeks and every prior spike has resolved; Dr Anas Alhajji's demand-destruction arithmetic still stands; and Axios's version of the story says no final order exists.
Bitcoin's repair (thesis 7) falls from 65% to 63%. Three separate negatives landed on the same weekend and none of them is about the price. First, self-custody produced a real, quantified loss: a flaw introduced in Coldcard firmware 4.0.0 in March 2021 caused affected devices to skip their hardware randomness generator and fall back to predictable software key generation, giving Mk3 devices roughly 40 bits of entropy instead of the intended 128. An attacker drained 594 bitcoin, about $38 million, from roughly 500 single-signature wallets between 01:31 and 01:56 UTC on July 30 — a twenty-five-minute sweep of coins that in some cases had sat untouched for years. The affected window is firmware 4.0.0 through 5.0.3 on Mk2 and Mk3 hardware; wallets whose seed was generated by dice roll, imported from elsewhere, or protected by a passphrase are unaffected. Second, the contested BIP-110 protocol change reaches its forced-signalling block around August 7 with miner support at roughly 2.4% against the 55% required, and Adam Back and Jameson Lopp — two of the most cited technical voices in the network — are publicly warning that the activation parameters risk splitting the chain. Third, price: bitcoin at $63,004 sits 0.3% above the $62,791 weekly-close repair line, and there is no cushion left at all. What keeps this at 63 and not lower is that the valuation model is squarely in its buying band at -1.05 standard deviations below trend and roughly 0.54x modelled fair value, the Fear and Greed sentiment index reads 27 ("Fear"), and the credit stack behind the MicroStrategy complex is being actively defended: management committed $1B to $3B to return the STRC preferred to its $100 par value and maintained the 12% dividend.
Everything else held, and two holds are deliberate rather than passive. The household squeeze (thesis 8) stays at 70% on genuinely mixed data. Against the thesis: University of Michigan consumer sentiment jumped to 49.5 from 44.8, a second consecutive improvement off a record low, and credit-card delinquencies at 2.92% continue to improve, so two of the three consecutive repair months this desk requires are now on the board. For the thesis: the savings rate at 2.7% is a cycle low, spending growth on flat real income is buffer drawdown by definition, and gasoline at $4.10 is now directly exposed to whatever happens in the Gulf this weekend. Those offset; the probability does not move on offsetting evidence. Levered residential real estate (thesis 9) stays at 70% with the 30-year mortgage at 6.66%, a one-year high, and the entry gate at 5.5% now 116 basis points away and receding.
An oil supply shock is being negotiated in public, against zero inventory cushion, over a weekend when the market cannot reprice. The strategic petroleum reserve at 308 million barrels is the lowest since March 1983 — that is a 43-year low, and it is the buffer that absorbed every supply disruption of the past two decades. It has fallen for eighteen consecutive weeks. Commercial stocks are at a 2018 low and 7% under seasonal. Into that, four major outlets published within roughly an hour of each other that the United States is preparing, considering, or has ordered a campaign against Iranian energy infrastructure, with Israel possibly joining for the first time in weeks, and Iran responding through its state agency that it will strike energy infrastructure in Israel and the Gulf states if it happens.
The operationally useful part is the asymmetry, not the forecast. If a strike lands at scale before Asian markets open Sunday evening, WTI clears the $89 confirmation gate on the open rather than during a session, and the $107 shock gate becomes the live question rather than a distant one. If the reporting turns out to be negotiating leverage — which is what the WSJ's own framing implies, since the stated objective is forcing Iran back to the table — the downside is shallow, because the same empty inventories that make a spike violent also put a floor under the price. That is a favourable payoff shape for exposure already held and a poor one for adding into the weekend at a level the desk's own gate says is not confirmed.
Keep the disagreement. Brandon Weichert, the national security writer on the Nawfal panel, argues there is no off-ramp and no viable targeting plan, because Iran has spent 47 years dispersing its infrastructure underground. Nawfal himself puts the strike at "more likely to happen than not." Against them, this desk's own oil framework — Marco Papic of BCA Research, via Jim Bianco — says the President is hawkish at $70 oil and deal-seeking near $100, and Brent is already $90.15. On that reading, the reporting IS the policy: the threat is worth more than the strike, and the closer Brent gets to $100 the more the administration's own incentive flips toward a deal.
| Instrument | Level (Fri Jul 31 close) | Change | What it means |
|---|---|---|---|
| Brent crude | $90.15 | +3.76% | Rallying through a strike pause; no inventory cushion under it |
| WTI crude | $87.18 | — | Still 2.0% BELOW the $89 add-gate; oil sleeve stays at target, no adds |
| 30-year Treasury | 5.27% | +6bp | Multi-decade high; a governance premium with a war premium stacking on |
| 10-year Treasury | 4.75% | +7bp | Through the 4.70% breakout line; every long-duration asset here reprices off it |
| 2-year Treasury | 4.28% | +5bp | 60bp above the 3.63% funds rate: the market says the next move is a hike |
| S&P 500 | 7,489.72 | +0.70% | 1.7% off highs; the tape is not confirming the bond market's alarm |
| Nasdaq Composite | 25,373.85 | +1.00% | AI complex bouncing after the forced seller cleared |
| VIX | 15.99 | -6.44% | Volatility priced as if none of the above is happening |
| Gold | $4,107 | -1.29% | Holds the $4,000 floor, 27% below February's high, 200-day at ~$4,588 |
| Bitcoin | $62,528 (live, 3:10pm ET Sat) | -0.47% | Now 0.4% BELOW the $62,791 repair line; Sunday's weekly close decides it; model in the buy band at -1.05 sigma |
| Amazon | $271.58 | +15.32% | AWS growth accelerating to 37% with margins up; the single largest AI-capex validation of the week |
| MSTR | $93.28 | -4.56% | Near its 52-week low of $82; the leveraged bitcoin expression is under stress |
| STRC preferred | $89.46 | -0.04% | 10.5% below the $100 par management has committed $1-3B to restore |
| High-yield spread | 2.84% | — | Near record tights; credit still refuses to confirm any of this |
### Oil and the Gulf
Wall Street Journal (unnamed US officials): the President has ordered a heavy strike campaign against Iranian energy infrastructure, potentially this weekend and lasting a few days, with the stated objective of forcing Iran back to ceasefire terms. Implication: the target set has moved from military and Gulf-state infrastructure to Iran's own export and production capacity, which is the first version of this war that directly removes barrels.
CBS News (unnamed officials): the US and Israel are preparing the most extensive bombing campaign yet against energy-related targets, potentially beginning over the weekend, with officials discussing wrapping it up before markets open Monday. Implication: the timing constraint is explicitly about market impact, which tells you the planners themselves expect a price reaction large enough to manage.
Axios (Barak Ravid, US officials): the President is CONSIDERING strikes on Iranian infrastructure including energy within the next few days and has not given a final order. Implication: this is the single most important disagreement in the set, and it is why the probability moved five points rather than fifteen.
ZeroHedge inventory data, cited on the Nawfal panel: the strategic petroleum reserve fell 3.8 million barrels last week to 308 million, the lowest since March 1983 and the eighteenth consecutive weekly decline, down 108 million barrels or 26% over the streak; commercial crude fell 7.2 million barrels to 405 million, the lowest since October 2018 and 7% below the five-year seasonal average. Implication: the shock absorber is gone, and both panellists put the usability floor at 300 million barrels, roughly a week away at the current draw rate.
Brandon Weichert (national security writer, 19FortyFive): there is no off-ramp and no viable targeting plan, because Iran has spent 47 years dispersing its infrastructure underground, and US interceptor stocks are depleted after roughly 40 days of striking Iranian missile launchers with no result. Implication: if he is right, this is a campaign that raises the risk premium without removing the threat, which is the worst combination for oil consumers.
Mario Nawfal (host): "more likely to happen than not, for sure," having called the prior "wipe out a civilization" threat a bluff with confidence. His two escalation red lines are energy infrastructure at scale, which he calls suicidal, and boots on the ground. Implication: a commentator who publicly faded the last threat is not fading this one.
Philip Pilkington (economist, on the Nawfal panel, afternoon tape): crude "should be $160 a barrel at least" at today's conditions, before any campaign against Iranian energy infrastructure. His mechanism is that Washington has held the price inside an artificial $70-90 band using three tools — strategic-reserve releases, large speculative short sales that lean on the tape, and the ability to de-escalate militarily whenever the price ran. Iran striking pre-emptively rather than only in retaliation has removed the third tool, so this weekend is the live test of whether the other two still work. Implication: he expects "a mediocre oil price rise" next week followed by renewed short-selling on Monday's open, and he says plainly he does not know whether the suppression holds. Note what this does to the bullish reading of the crowded short position: at roughly 228,000 contracts of speculative gross shorts near a five-year extreme, the standard read is dry powder for a squeeze, while Pilkington's read is that those shorts are policy-adjacent and will re-engage into strength. Same number, opposite conclusion. That is the sharpest conflict in today's oil set.
Pilkington, on duration, which is the part with real portfolio content: if the problem is only the strait being closed, it resolves in roughly three months; if Iran's production and export infrastructure is actually destroyed, restoring both the infrastructure and normalised transit takes twelve to eighteen months. The empirical anchor he uses is the Houthi precedent: one strike on one Saudi refinery required four to five months of repair, spiked crude about 6%, and then round-tripped lower. Implication: the near-dated barrels are already not reaching the market, so a strike does comparatively little to spot and a great deal to the back of the futures curve. If this thesis is right, the expression is calendar structure rather than front-month price, and the $89 gate this desk watches may understate what is happening further out.
Brandon Weichert (national security writer, 19FortyFive), afternoon tape: 80% odds on energy infrastructure struck at scale in a two-week campaign, high conviction that Iran does not return to the table ("I don't think there's a deal to be had"), and the judgement that the US military is the most strained it has been since the interwar 1920s with no latent industrial capacity to draw on, implying several years of logistical fallout even if the war ended today. Implication: his own capacity argument is in tension with his own probability. A force that cannot sustain the campaign is less likely to start it at the scale he forecasts, and that tension is why this desk did not follow the 80%.
Marco Papic (BCA Research), via Jim Bianco: oil is the independent variable driving the war rather than the dependent one — a hawkish President at $70 crude and a deal-seeking one near $100. Implication: Brent at $90.15 is already in the range where the administration's own incentives argue against execution, which is the strongest structural counter to the strike case.
### Bitcoin, self-custody and the protocol
Coinkite disclosure, via CoinDesk and Bitcoin.com: firmware 4.0.0, shipped March 2021, caused affected Coldcard devices to skip their hardware random number generator and fall back to software generation seeded by non-secret chip data, giving Mk3 devices roughly 40 bits of entropy against the intended 128. An attacker drained 594 bitcoin, about $38 million, from roughly 500 single-signature wallets between 01:31 and 01:56 UTC on July 30. Every drained wallet was single-signature and held more than 0.15 BTC. Implication: the loss is small against a $1.2 trillion asset, but it is the first mass self-custody failure with a named vendor, a countable dollar figure and a five-year exposure window, and it hands the custody argument to the custodians at the worst moment for retail confidence.
Nick (BitGo, $100B+ under custody), on the True North emergency stream: the bug sat in open, auditable code for five years because "we almost assume someone did the verifying already," and his conclusion is that individuals belong with a bankruptcy-remote qualified custodian. Implication: note the commercial interest, and note that he is nonetheless describing a real failure of the community's own verification norm.
Jesse (Amboss) and the True North host, same stream: self-custody survives this if you use multi-signature across multiple hardware vendors and generate entropy physically, by dice or card shuffle. Their instruction set is update firmware, generate a new seed, send a test transaction, then migrate, and never type a seed phrase into a website that offers to check your exposure. Implication: the operationally correct response is a procedure, not a product change.
Simon Dixon (founder, BnkToTheFuture; seventh-largest creditor in the Celsius bankruptcy) dissents directly: Celsius creditors recovered about 25% of their bitcoin, and handing over keys makes you a subordinate claimant. He also puts the affected firmware window at 2021 to 2023 rather than open-ended, while disclaiming his own accuracy. Implication: use the conservative framing — any seed generated on affected firmware after March 2021 — and keep the custody disagreement rather than resolving it, because the two sides have opposite commercial interests and Dixon has the direct loss experience.
BIP-110 status (bip110monitor, CoinDesk, Bitcoin.com): the proposal, a one-year consensus restriction on arbitrary data embedding aimed at Ordinals inscriptions and large OP_RETURN payloads, requires 55% miner signalling within a difficulty period, versus the 95% threshold used in past soft forks. Signalling is running near 2.4%. Failure to reach the threshold pushes it into a MANDATORY signalling phase at block 961,632, projected around August 7, which rejects any block that does not signal. Adam Back and Jameson Lopp call the parameters reckless and warn of a chain split; Bitcoin Core has not endorsed it; Michael Saylor has pushed back publicly. Implication: this is a dated binary inside six days that the bitcoin view already lists as a stressor, and it is now the nearest catalyst on the calendar.
Avi Felman and Jonah Van Bourg (1000x, ex-BlockTower and ex-Goldman/DRW/Cumberland): bitcoin is in "no man's land" until the AI trade cools or rates fall, and rates are not falling; unlike memory chips, which can return to highs without the forced seller, they argue bitcoin cannot go straight back to highs without Saylor being liquidated first. Their custody advice is a major hardware wallet with the seed in a bank vault, or a reputable exchange with a passkey for non-technical holders. Implication: this is a new and specific bear mechanism inside the complex, and it directly contradicts Dixon.
Simon Dixon, on the same question: bitcoin ROSE while AI equities crashed this week, and he declares the bitcoin-to-AI-equity correlation thesis dead. Implication: single-sourced and contradicted by the price action into Friday, but logged, because if it holds it removes the main channel through which an AI unwind reaches this book.
### AI, memory and the forced seller
Avi Felman and Jonah Van Bourg (1000x): Leopold Aschenbrenner's Situational Awareness fund ran roughly $25-30B of capital against roughly $120B of market exposure, about 4x leverage, in names that move 120% annualized, holding roughly 25% of one company and about 5% of several others — a violation of the rule of thumb that says never own more than 10% of a name's average daily volume. Bidders for the liquidated public book were Jane Street, Millennium and Citadel; Citadel paid the best price and they estimate Ken Griffin is already up $3-5B. The LP letter says the fund continues as a hybrid public-private vehicle running the public book on a fully paid-for basis, still up 80% year to date despite a 67% down month, a figure Van Bourg openly disbelieves. Implication: this closes the diagnosis this desk has run since Wednesday — a leverage flush, not a thesis break — with primary detail rather than inference.
Ben (technology analyst, Mizuho Securities, on Bloomberg Technology): Amazon's AI story is "a game changer," and the specific facts behind that phrase matter more than the phrase. Amazon Web Services revenue growth ACCELERATED to 37%, and Amazon disclosed for the first time a $25B annual run rate in renting compute on its own in-house chips rather than Nvidia's. AWS margins are rising despite negative free cash flow over the trailing twelve months, which directly refutes the bear case that AI capital spending would compress cloud margins. His reasoning on the in-house silicon has three legs: supply, because Nvidia chips are hard to obtain; cost, because Amazon can undercut on price; and the margin it no longer surrenders to Nvidia. He notes Amazon trades in the low-to-mid 20s times earnings, which he frames as mispriced against Costco and Walmart. Amazon rose 15.32% to $271.58 on Friday. Implication: this is the strongest fundamental support the AI building-boom view (thesis 1) has received, and it is a disclosed number rather than a forecast — but the morning run already raised that thesis from 65% to 68% on this same earnings print, so it is confirmation, not a second move. Thesis 1 holds at 68%. His contrasting call is that Apple's 9% fall is the unwind of a crowded hiding place rather than an earnings problem, and he would not buy it, because services (the high-margin engine) missed and Tim Cook has admitted under-ordering processors for the iPhone 18 cycle.
Same source, on positioning: they are now buying Micron and SanDisk, adding to both, avoiding SK Hynix, Samsung and Korean equities entirely, and reallocating into Intel on national-security grounds. Their fundamental claim is that memory is physically undersupplied by at least 40%, probably 50%, for the next twelve months, with module prices expected to TRIPLE over that window. They expect a retest of the lows first and two to five weeks before the displaced $25B returns. Implication: this is the strongest bull mechanism the memory view (thesis 2) has faced, and it comes from a desk that was out of the trade for two months and is reversing. It is a named dissent, not a reason to move: Forward Guidance's read on the same tape is that the destroyed levered buyer base means months of chop, and Micron fell 5.9% to $823 on Friday. Thesis 2 holds at 71%.
### Tesla and the vision-only question
Cern Basher and co-host (Tesla-focused investors): the Optimus factory at Giga Texas is being built to 1 million units of annual capacity with steel going vertical, one of three major factories under construction on that site and ten across Tesla and SpaceX combined; the Cortex 2 training cluster is roughly 50% activated; cumulative FSD miles on Tesla's own counter approach 12 billion. Driverless robotaxi service is live in four cities including Miami, pending in Tampa and Orlando, and running with safety drivers in San Francisco. Implication: the buildout is not the constraint.
The constraint, same source: a New Jersey bill authorises autonomous vehicles but requires "a redundant safety system," which in practice mandates lidar-class hardware and excludes vision-only. Tesla has stood up a constituent campaign citing 578 New Jersey traffic deaths last year and 94% of serious crashes attributable to human error. Against it, the Innermost Loop digest reports NHTSA fast-tracking the first-ever national AV performance standards explicitly to replace the state patchwork, plus 2,500 Zoox robotaxis a year. Implication: both are true, and federal preemption is now the load-bearing variable for Tesla's vision-only cost advantage. TSLA at $311.21 is this book's largest single-name position at $466,815, so this is the regulatory risk that matters most to the portfolio.
Live net asset value $3,022,146, with Friday costing $18,427, or 61 basis points. Best contributor QURE +$5,400; worst BITB -$19,594.
Exposure by cluster. Digital assets (BITB, BSOL, COIN) $807,358, 26.7%. AI equity, which here is effectively TSLA alone, $466,815, 15.5%. The STRC preferred reports separately at $393,624, 13.0%. Biotech $330,850, 11.0%. Oil and gas (FCG, PSCE) $72,082, 2.4% against the 5% sheet target. Cash 11.5%.
Biggest gap: oil, and it is a deliberate one. The energy sleeve sits at 2.4% against a 5% target precisely because the desk pre-committed to a $89 WTI confirmation gate before adding, and WTI closed at $87.18. That discipline cost nothing this week. It costs something if a strike lands over the weekend and the gate is cleared by a gap rather than a session, because the add price will not be $89. That is the honest cost of a rule, and the rule is still right: buying an unconfirmed war premium into a weekend is a different trade from buying a confirmed one.
Sharpest conflict in the book: bitcoin exposure against bitcoin's own calendar. Digital assets plus the STRC preferred are 39.7% of net asset value, and STRC — although it reports on its own line by design — is analytically credit in the bitcoin complex, so a bitcoin drawdown reprices both legs together. The S1 stress scenario, bitcoin down 53% to $30,000 with STRC repriced to $75, costs -16.9% of net asset value. Against that concentration, the next six days contain a protocol event with chain-split warnings from Adam Back and Jameson Lopp, a fresh self-custody failure with $38 million gone, and a price that has now slipped 0.4% BELOW the line the desk itself named as the repair level, with the deciding weekly close due Sunday evening. Set against it: the valuation model says buy, sentiment reads Fear at 27, and Strategy is committing $1-3B to defending the preferred at par. Both things are true at once, which is what a concentrated position feels like when it is working as designed.
Model verdicts in plain language. The bitcoin model reads BUY at -1.05 standard deviations below its long-run trend, with price at roughly 0.54x modelled fair value and the Fear and Greed sentiment gauge at 27 ("Fear") — scheduled accumulation is justified, a larger one-time purchase is not until the model reaches -1.3 sigma. The full stress book: an AI unwind scenario (Nasdaq -25%, crypto -30% in sympathy, TSLA -40%, power names -30%) costs -15.8%; oil breaking to $70 costs -0.6%, which is the mathematical statement of how small the energy sleeve is; the combined worst case is -27.9%. Effective diversification is 2.72 independent bets across eight sleeves at an average correlation of 0.23. Year-to-date the book is -24.9% on the proxy measure, with a money-weighted return since inception of 25.3%, Sharpe 0.81, volatility 51.3% and maximum drawdown 40.8%. Three open items, two of them overdue.
1. Position size is the risk; being right is not a defence against being forced to sell. (Avi Felman and Jonah Van Bourg, 1000x.) Leopold Aschenbrenner's call on AI infrastructure was, by their own account, prescient and is still working — and he lost 67% in a month anyway, because roughly $25-30B of capital carried about $120B of exposure in names that move 120% annualized, and he owned roughly 25% of one of them. Their rule set afterwards is explicit: no more than 2x leverage on a long-term thesis, bottom-drawer positions, never more than 10% of a name's average daily trading volume. Apply: before adding to any position, ask what fraction of it you could exit in a single day at a price you would accept. If the answer is "not much," the thesis is irrelevant, because you will not be the one choosing when to sell.
2. In a forced sale, the winning bid comes from the longest holding period, not the best valuation model. (Jonah Van Bourg, 1000x.) Citadel outbid Millennium and Jane Street for Aschenbrenner's book not because it valued the stocks differently but because Millennium's pod-level risk limits and Jane Street's short holding periods both cap what they can pay for an illiquid equity block they may need to sit on. Griffin's own stated framing is that "the next form of alpha is three-to-five-year alpha." Apply: in any distressed situation — a liquidation, a divorce sale, a dealership taking a trade nobody wants — your edge is duration tolerance rather than analysis. Know before you bid how long you can hold, because that number, not your valuation, is what you are actually bringing to the auction.
3. A control you have never personally verified, and cannot name someone who verified, is an assumption. (Nick of BitGo, on the True North emergency stream: "we almost assume someone did the verifying already.") The Coldcard entropy flaw sat in open, auditable, widely deployed firmware for five years inside a community whose founding slogan is "don't trust, verify," and it cost 594 bitcoin in twenty-five minutes. Open source is a necessary condition for verification, not the verification itself. Apply: list the controls you depend on but have never tested — the backup that has never been restored, the insurance policy never read past the summary page, the trust document nobody has opened since signing. For each, either test it or write down whose job it was, by name. If neither answer exists, it is not a control.
4. When a forecasting model scores itself against the variable someone is manipulating, the manipulation never gets detected. (Philip Pilkington, economist, on the Nawfal panel.) His argument for why verbal intervention in the oil market keeps working: the Bayesian inference models that price political statements use the oil price itself as the target variable, so if a statement coincides with a flat or lower price, the model records the statement as credible and weights it more heavily next time. The feedback loop is self-confirming, which means adaptation makes it stronger rather than correcting it. Apply: whenever you notice that some actor's announcements reliably "work," check whether the thing you are using to measure their credibility is the same thing they are able to move directly. If it is, you have no evidence at all, only a mirror. This generalises well beyond oil, to guidance-versus-print at any company you own and to any process where the scorekeeper and the scored are connected.
5. Capability is a harder forecasting input than intent, because intentions leak cheaply and logistics do not. (Drawn from the contradiction inside Brandon Weichert's own segment.) He put 80% on a large campaign against Iranian energy infrastructure while simultaneously reporting that the CENTCOM commander told the Pentagon he lacks the naval forces to keep defending Israel, that five Patriot batteries have left Erbil, and that the US military is the most strained it has been since the 1920s. Four major outlets sourced the intent to unnamed officials and disagreed with each other on the verb; the capacity constraint came from a named commander's internal warning. Apply: when forecasting what any organisation will do — a government, a competitor, a counterparty in a negotiation — weight the disclosure about what it can afford or staff far above the reporting about what it plans. Intent is free to signal and is often signalled precisely because it is not affordable.
Pulled today (20 transcripts, roughly 184,000 words). Morning run (16): Forward Guidance (1), Macro Voices #543 with Jim Bianco (1), True North (2: the Strive team's post-earnings Hurdle Rate Ep. 67, and the Coldcard emergency stream), Simon Dixon (3, two of which are the same show), 1000x Network (1), Mario Nawfal (2), Bloomberg Technology (2), Cern Basher (1), a16z (1), The Innermost Loop (1), Wealthion (1).
Afternoon run (4 new): Mario Nawfal, "OIL EXPECTED TO PUMP HARD AS TRUMP PLANS IRAN ATTACKS" with Philip Pilkington (7,315 words); Mario Nawfal, "BREAKING: U.S. ISSUES SECURITY ALERT" with Brandon Weichert (9,175 words); Bloomberg Technology, "Amazon's AI Story Is a 'Game Changer,' Says Mizuho" (1,167 words); Simon Dixon, "AI Unwinds, Bitcoin Wallet Hack & the Middle East Realigns Part One" (11,553 words).
Deduplication note: the Simon Dixon Part One file is a 97% verbatim subset of the longer Dixon episode already ingested this morning ("The AI Sell-Off, Bitcoin Wallet Hacks, Global Realignment"), which contains Part One plus a further 19,000-word interview. No unique content; not counted twice in the synthesis.
Skipped as too thin to be usable: three Jim Bianco clips (155, 234 and 171 words) and one Wealthion clip with Andrew Sarna on gold (74 words). These are short-form uploads, not episodes.
Unavailable, requeued for tomorrow: Mario Nawfal, "TRUMP EXPECTED TO STRIKE IRAN POWER PLANTS WITHIN 48 HOURS" with Professor Rob Pinfold (no transcript published). The Ken Griffin and David Rubenstein Bloomberg TV interview remains unavailable, ninth day in the queue.
Nothing blocked and no fetch errors on either run. No new issues from the paid newsletters (Lyn Alden, Northstar) since yesterday.
Already folded on July 31 and not double-counted: Forward Guidance, Macro Voices, True North Ep. 67, Wealthion, a16z, Bloomberg Technology x2, Innermost Loop. Genuinely new to the morning run: 1000x Network, Simon Dixon, the True North Coldcard stream, Mario Nawfal x2, Cern Basher.
Web sources for the two moving stories: Wall Street Journal, Axios, CBS News, CNN and Washington Post on the Iran strike reporting; CoinDesk, Bitcoin.com, Coinkite disclosure and bip110monitor on the Coldcard exploit and BIP-110 status. Market levels from live FMP quotes at Friday's close; macro series from FRED.
No paid newsletter issues from Lyn Alden or Northstar since the last run.