Our forecasting track record

What changed, in plain words

• Hormuz: five cargo ships crossed the Strait of Hormuz on Saturday and none on Sunday, against 31 the previous weekend. The ship-tracking data, from Kpler via Reuters, follows the UAE saying three Abu Dhabi National Oil Company vessels were attacked in transit last week. — The world's most important oil channel went from busy to empty in a week.
• Iran talks: the 60-day US-Iran arrangement signed on 17 June lapsed on 16 August with nothing to replace it. A senior White House official called the talks "static"; Iran's foreign minister said Tehran has not decided to restart negotiations and that the memorandum was "the end of the war", not a ceasefire that could be extended. — The diplomatic route out of the blockage is not moving.
• Oil: Brent crude traded near 89.28 dollars a barrel, flat on the day but over 5% higher on the week. Prices have given back almost all of their early-August fall, after the tanker attacks and a strike on a Saudi Aramco refinery. — Fuel is expensive again — the brief discount has gone.
• The US shopper: American retail and food-service spending fell 0.6% in July, the first decline in nine months, against forecasts of a small rise. Cars fell 1.8% and online shopping fell 2.2%; clothing rose 1.9%. Spending is still 5.0% above a year ago. — Households pulled back, though they still spend more than last year.
• AI debt: Bank of America estimated Broadcom's chip-financing vehicle, built with Apollo and Blackstone, could carry 370 billion dollars of senior debt by mid-2029. That debt sits in the vehicle, not at Broadcom, which discloses up to 29 billion dollars of its own exposure; the bank cut its credit view and the shares fell about 6% on Friday. — The AI buildout keeps moving onto borrowed money held outside company accounts.
• How people feel: the University of Michigan's August sentiment survey fell to 51.0 from 55.2, well below the 54.5 forecast, with the sharpest drops among older, lower-income and non-college households. Only 8% of people expect their income to grow faster than prices over the next year. — Nine in ten Americans expect prices to outrun their pay.
• The Fed: market-implied odds of a September interest-rate INCREASE fell to roughly 30% from about 50% a month ago after the weak spending and sentiment data. Different venues put the number anywhere from about 22% to 32%, so treat it as a range, not a figure. — Weaker demand data made an imminent rate rise look less likely.
• Memory chips: TrendForce now expects the contract price of ordinary memory chips to rise another 13-18% this quarter and flash storage 10-15%, after roughly doubling in the first quarter, as manufacturers move capacity to the high-margin chips that AI servers need. — The chips inside phones and laptops keep getting dearer, which eventually shows up in retail prices.
• Gold: gold rose toward 4,400 dollars an ounce, quoted at 4,395 and up 0.45% on the day, extending Friday's gain as expectations of a near-term rate increase faded. — Gold keeps climbing as confidence in the currency's purchasing power softens.
• Shares: US indices slipped on Friday but still finished a third straight winning week — the S&P 500 fell 0.17% to 7,786, the Nasdaq Composite fell 0.28% to 26,729 and the Dow fell 0.20% to 53,732. — Share markets are still near records and have not reacted to the shipping news.

Headline

The discrete event of the day is physical, not financial. Commercial transits through the Strait of Hormuz went to five vessels on Saturday and zero registered for Sunday, against 31 the prior weekend (Kpler shiptracking via Reuters, EVT-0383). The trigger was the UAE's disclosure that three ADNOC-operated vessels were attacked in transit last week, compounded by a strike on a Saudi Aramco refinery. Roughly a fifth of global oil and gas moves through this waterway.

The diplomatic channel closed alongside it. The 60-day Islamabad Memorandum of Understanding signed 2026-06-17 lapsed on 2026-08-16 with no successor; a senior White House official characterised talks as "static", and FM Araghchi's position hardened into a definitional argument — that the MoU was "the end of the war", not a ceasefire, and therefore cannot be extended (EVT-0384). Tehran is reportedly pursuing a separate Oman track on Hormuz traffic, which is the one live thread. This bears directly on P-0074 (announced interim reopening by 2026-08-20, p=0.55 when logged): that probability now looks too high on the evidence, and the reviewer should expect it to resolve false absent an Oman surprise this week.

The price response is the part worth flagging as a possible narrative trap. Brent sits near 89.28 with WTI at 81.74, up over 5% on the week but roughly unchanged today (EVT-0385) — a physical shutdown of the world's most important chokepoint that is NOT producing a disorderly price move. Reuters reporting notes Middle Eastern producers moving barrels through the waterway covertly, which is the most plausible reconciliation. Either the market is correctly reading the closure as porous and temporary, or the price is lagging the physical fact. We cannot distinguish those today, and it would be a mistake to write the second story as though it were established.

On the demand side the US consumer produced two independent measurements pointing the same way on 2026-08-14: advance retail and food-service sales −0.6% m/m to USD 763.6bn, the first decline in nine months against a +0.1/+0.2% consensus (EVT-0386), and preliminary August sentiment at 51.0 from 55.2, with one-year inflation expectations at 4.3% and only 8% of respondents expecting income to outpace prices (EVT-0387). These are treated as ONE Readings row deliberately: they are hard and soft measurements of the same underlying condition, released the same morning, and splitting them would fake corroboration in the global rollup. Note the sampling caveat — the advance retail estimate carries ±0.4pp of error, so −0.6% is real but not precise.

The AI complex delivered the running theme's material development. BofA's Tom Curcuruto put up to USD 370bn of senior debt at Broadcom's XPV vehicle by mid-2029 at 20GW scale, including ~USD 150bn of 2027 issuance, and cut the credit view from Overweight to Marketweight; AVGO fell ~6% (EVT-0388). The structural point is that the debt sits at the vehicle, formed with Apollo and Blackstone, while Broadcom's disclosed exposure is the USD 29bn backstop of certain customer lease obligations. This is the second consecutive session where the AI story's centre of gravity moved from demand to underwriting — Friday's Goldman estimate that roughly a third of 2026 AI capex is debt-funded, today's specific vehicle. Thesis #1 (AI-credit fragility) is confirmed on the STRUCTURE leg and unconfirmed on the PRICE leg: nothing in the credit gauges has moved (see below), so what we have is a growing measured quantity of leverage, not evidence of stress in it. P-0088 escalates the falsifier.

Memory pricing cuts the other way and is worth stating plainly because it disconfirms the fragility read: TrendForce forecasts conventional DRAM contract prices +13-18% q/q and NAND +10-15%, after +93-98% in Q1 and +58-63% in Q2, with HBM4 costed at USD 31-32/GB to NVIDIA versus USD 17-18 for HBM3e (EVT-0392). That is a demand signal, not a glut signal, and it keeps P-0064's China-glut disconfirmer unfired. It is also the mechanism behind the consumer-electronics cost row that has run in this brief since 2026-08-14.

The stress gauges themselves stayed quiet, and this is where the discipline matters. The deterministic credit-and-curve tool reads calm on both components: the 2s10s slope at +0.51pp on 2026-08-14 with no re-steepening event, and high-yield spreads at 2.71pp on 2026-08-13, well inside the 4.0pp band and with zero widening velocity over the trailing 21 observations. Equities are consistent with that — the S&P 500 at 7,786 after a third straight weekly gain (EVT-0390). So the honest summary is a single discrete physical stress inside an otherwise calm financial system, with the transmission channel (fuel costs, then household budgets) slow rather than immediate.

Readings

Fact-check log

Standing theses — re-scored

• #1 AI-credit fragility — CONFIRMED on structure, UNCONFIRMED on price. The Broadcom XPV estimate (EVT-0388) is the most specific quantification yet of off-balance-sheet AI leverage, and it follows Friday's estimate that a third of 2026 capex is debt-funded. But the price leg is unmoved: high-yield spreads at 2.71pp with zero widening velocity. Growing measured leverage is not the same as stress in it, and conflating them is the narrative trap here. Escalated as P-0088. — The AI boom's borrowing is getting bigger and less visible, but lenders are not charging more for it yet.
  For your money Most savers own these companies indirectly through pensions and index funds. What matters is not that borrowing is large but whether lenders start demanding more for it — that is the signal worth waiting for, and spreading savings so no single story dominates is the ordinary defence.

  What to watch Matters more if lending terms on AI-linked borrowing visibly worsen or a large planned deal is pulled; matters less if this debt keeps being placed easily.
• #6 Energy and critical commodities — CONFIRMED, with an important caveat. The physical supply leg fired hard (EVT-0383): transits to zero, the diplomatic track lapsed (EVT-0384). But the price leg did not follow — Brent unchanged on the day. Coverage audit returns 3 covered / 0 partial / 0 gap; deliberately NO new prediction logged, because P-0051 (Brent above 100), P-0061 (Brent below 75), P-0074 (announced reopening), P-0079 (second chokepoint) and P-0085 (CPI spillover) already cover this theme's axes. A sixth would inflate nominal sample without adding independent information. — The oil route is genuinely shut, but the oil price is not behaving as if it is.
  For your money Fuel is the fastest-moving cost in a household budget, so this is the thesis with the shortest path to your actual bills. The gap between a shut route and a calm price is a reason to watch rather than act, and a cash buffer is the ordinary preparation for months when fuel jumps.

  What to watch Matters more if transits stay near zero into next week or the Oman track collapses; matters less if traffic resumes or a reopening is announced.
• #9 Labour and consumer deterioration — first genuine confirm in weeks. Retail sales −0.6% (first drop in nine months) and sentiment at 51.0 with 8% expecting income to beat prices (EVT-0386, EVT-0387). Counted as ONE signal, not two, since they are hard and soft measurements of the same condition released the same morning. The labour leg proper (claims, payrolls) has NOT confirmed — Friday's claims data was noisy but the four-week average held. P-0089 tests whether July was an air pocket. — People spent less and feel worse, but the job market itself has not turned yet.
  For your money Spending is most of the US economy, so a sustained pullback eventually reaches jobs and company profits. One month is not a trend. Several months of expenses held in cash is the ordinary protection against that sequence, and this isn't a reason to make a big move.

  What to watch Matters more if August spending falls again or unemployment claims start climbing with it; matters less if spending rebounds while jobs data stays steady.
• #4 Compute as a strategic resource — CONFIRMED and unchanged in direction. TrendForce's +13-18% q/q DRAM forecast (EVT-0392) keeps the memory-glut disconfirmer P-0064 unfired and reinforces the demand leg. This cuts AGAINST the fragility read in #1, and both are being tracked rather than reconciled prematurely.
• #7 Equity concentration and breadth — no change. S&P 500 at 7,786 after a third weekly gain (EVT-0390), with equities not reacting to the shipping halt. P-0070 (equal-weight vs cap-weight through 2026-09-04) remains the open test; nothing today bears on it.
• #2 De-dollarization, #5 sovereign rates, #8 AI services — no material new evidence today. Gold's climb (EVT-0391) is consistent with #2 but a single day's move is not evidence; no auction, fiscal or IT-services development crossed the threshold.

Research, not financial advice. Portfolio results are mock capital.
Project snapshot — changes waiting for review: 3 · tracked forecasts open: 51 (checking now: 0) · track record so far: 0.128245 · practice portfolio updated: 2026-08-14.
Past briefs & dashboards: market-sentinel-lyart.vercel.app/reports