Our forecasting track record

What changed, in plain words

• Inflation: US consumer prices rose just 0.1% in July, easing the annual rate to 3.4% from 3.5%. The narrower "core" measure that strips out food and fuel eased to 2.5%, close to the Federal Reserve's 2% goal. — The measure the central bank watches most is nearly back to normal.
• Your paycheck: The average US hourly wage now buys 0.2% less than it did a year ago. Weekly pay is up just 0.1% in real terms, and only because people worked slightly longer hours. — Pay rises have not kept up with prices over the past year.
• AI chips: Supermicro reported record orders and doubled its profit margin — and burned $6.81bn of cash over the year. Sales were $11.1bn against $5.8bn a year earlier, but inventory rose from $4.7bn to $12.9bn, funded by $9.5bn of new borrowing and share sales. — The order book looks superb, but cash went out faster than it came in.
• Memory chips: Micron, Samsung and SK Hynix shares sit more than 20% below recent highs even as their chip prices surge. Sequential price gains were strong in absolute terms but still short of what analysts had penciled in. — The chip industry's most profitable corner is being marked down while its prices rise.
• Long-term rates: The US 30-year borrowing rate held at 5.24%, near multi-decade highs, despite the cool inflation news. The pressure sits in the return lenders demand above inflation, not in inflation expectations. — Long-term borrowing costs are not falling even as inflation cools.
• Fuel: Energy costs are still 14.7% higher than a year ago and petrol 24.6% higher, which is the whole reason the headline rate (3.4%) sits so far above the core rate (2.5%). — One item — fuel — explains almost all of the gap between the two inflation numbers.
• The Fed: Traders trimmed the chance of a US interest-rate *rise* in September to about 42%. Note the direction: the live debate is whether rates go up, not down. — The next move in US rates is still more likely to be up than down.
• Crypto: Bitcoin traded near $64,200, about 49% below its October 2025 peak of $126,080, on a day US share indexes sat close to record highs. — The most speculative corner of markets is still deeply depressed while shares are not.

Headline — a benign print that did not reach the household, and an order book that did not reach the cash box

The July CPI (BLS, released 08:30 ET on 2026-08-12, EVT-0366) is a clean disinflation print at the level the Federal Reserve cares about: core +0.2% m/m, 2.5% yoy, down from 2.6%. Headline was +0.1% m/m and 3.4% yoy. Shelter contributed roughly two-thirds of the monthly rise at +0.1%, and the energy index fell 1.5% on the month as gasoline dropped 2.9%.

The honest reading is in the 0.9-point wedge between headline (3.4%) and core (2.5%). That wedge is almost entirely energy: the energy index is +14.7% over twelve months, gasoline +24.6%, fuel oil +39.1%. Core goods — the tariff-exposed line — rose 0.2% m/m, its firmest since February, but remain only +0.8% yoy, so Section 301 passthrough is still not visible at the index level. This is a print where the measure the central bank targets improved and the measure a household actually pays did not.

BLS made that explicit in the same release window (EVT-0367): real average hourly earnings for all employees fell 0.1% on the month and are down 0.2% over twelve months. Real weekly earnings are up just 0.1% yoy, and only because the average workweek lengthened 0.3%. Twelve months of nominal wage gains have been fully consumed by prices. For a system whose stated purpose is protecting savings in real terms, that is the load-bearing fact of the day, not the 3.4%.

The long end refused to celebrate. DGS30 sat at 5.24% on 08-11 (EVT-0369), a few basis points off its 5.27% period high, with the 30-year inflation-protected real yield at 3.00% while the 10-year inflation breakeven is only 2.26%. That decomposition matters: the pressure at the long end is in required real compensation, not in expected inflation. Cooling CPI does not relieve it, and every risk sensor is priced for the opposite world — 2s10s at +0.48pp, high-yield spreads at 2.72%, VIX at 15.28, the S&P within a whisker of records. Open claim P-0076 already carries this divergence; nothing today resolves it.

Supermicro's fiscal Q4 (8-K Ex-99.1, 2026-08-11, EVT-0368) is the day's dominant AI-buildout development and it is genuinely two-sided. Net sales of $11.12bn against $5.76bn, GAAP gross margin of 17.5% against 9.5%, diluted EPS $1.62, more than $60bn of new orders, record backlog, and an FY27 guide of $65–72bn — that is a strong confirmation of the compute-demand leg, and the stock rose about 19% on 08-12. Read the cash-flow statement and the picture changes: FY26 net cash used in operating activities was negative $6.81bn (versus +$1.66bn in FY25), driven by an $8.88bn inventory build and a $3.92bn receivables build, funded by $9.48bn of financing inflows including $4.23bn of mandatory convertible preferred and $1.41bn of common issuance. Inventories nearly tripled to $12.90bn. The demand is currently an inventory-and-receivables position financed in the capital markets, and the filing itself is preliminary, unaudited, and discloses an ongoing independent Board review of transactions connected to export-control issues. That is why the day's escalated disconfirmer (P-0084) is on the inventory-conversion axis rather than on memory pricing, which open claims already cover.

The memory leg rhymes (EVT-0372). Contract DRAM prices are surging — DigiTimes reports the increase lifting Samsung and Micron while SK Hynix loses share — yet Micron, Samsung, SK Hynix and the memory ETF all sit more than 20% below recent closing highs. The second-hand interpretive layer, which I have marked partially-verified rather than verified, is that sequential ASP gains of roughly 30% (SK Hynix) and above 40% (Samsung) still undershot sell-side expectations near 39% and 48%. Strong prints inside a de-rating tape is the same pattern in both names.

One cross-asset check worth keeping on the record because it cuts against the "broad melt-up" story: Bitcoin sits near $64,200, roughly 27% down year-to-date and about 49% below its October 2025 high of $126,080 (EVT-0370). Risk appetite is not uniformly euphoric. The most liquidity-sensitive, most speculative sleeve broke months ago and has not recovered, which is a poor fit for a narrative of indiscriminate risk-seeking.

Readings

Fact-check log

Standing theses — re-scored

• AI-buildout fragility (#1) — CONFIRM and DISCONFIRM in the same filing; net: unchanged, better evidenced. Supermicro's $11.12bn quarter, >$60bn of new orders and $65–72bn FY27 guide are the strongest demand confirmation in weeks. Its −$6.81bn operating cash flow, tripled inventories and $9.48bn of financing inflows are the clearest evidence yet that the financing leg, not the demand leg, is where this thesis lives. Escalated as P-0084 on the inventory-conversion axis. — The AI demand is real; whether it is funded by cash or by borrowing is the open question.
  For your money Most diversified savings now carry a meaningful slice of this one story through index funds. The question worth knowing the answer to is how large that slice is, not what to do about today's headline. Staying spread out across sectors and not borrowing against a concentrated position remain the ordinary defences.

  What to watch It matters more if the next quarter's cash flow stays negative while inventory grows, or if audited figures diverge from these preliminary ones; less if the backlog converts to cash.
• Real-purchasing-power erosion — CONFIRM, and this is now a first-class thesis. BLS reported real average hourly earnings −0.2% yoy on the same morning as a friendlier CPI. The operator's refinement (protecting savings = protecting real purchasing power) is exactly what this print measures, and no open prediction covered it until today. Escalated as P-0083. — A year of pay rises has been fully eaten by prices.
  For your money This is the slowest and most certain way savings lose value, and it affects cash balances as much as anything else. It rewards a spread-out, unleveraged position and a cash buffer sized for real bills rather than any single decision today.

  What to watch It matters more if the yearly figure stays negative for several consecutive months, or if fuel turns higher again; less if energy keeps easing and pay growth holds.
• Sovereign / long-rates stress (#6) — CONFIRM on the level, not on the direction. The 30-year at 5.24% with a 3.00% real yield and only a 2.26% ten-year inflation breakeven says the long end is demanding real compensation, and a cool CPI did not move it. Already carried by open claim P-0076; deliberately not re-logged. — Long-term borrowing costs are high for reasons inflation data cannot fix.
  For your money Sticky long rates keep mortgages and business loans dear while finally paying savers something above inflation. It argues for spreading savings across different maturities rather than concentrating in either very long bonds or in assets that only work if rates fall.

  What to watch It matters more if the long end rises on good inflation news again, or if government debt sales need visibly higher yields to clear; less if it converges back toward the cooler inflation picture.
• Equity concentration / breadth (#7) — MIXED, with a genuine disconfirmer of the euphoria reading. The S&P sits near records with 91.7% of sector sensors above their 200-day average, yet Bitcoin is ~49% below its October 2025 high and memory equities are >20% off theirs. Indiscriminate risk-seeking is the wrong description of this tape. Open claim P-0070 covers the breadth question. — Shares are near records, but several risky corners are not — this is selective, not blanket, optimism.
• Energy / critical commodities (#3) — CONFIRM on the household channel, easing on the market channel. Brent around $89 is well below July's ~$98 spike and the energy CPI fell 1.5% on the month, but energy is still +14.7% and gasoline +24.6% over twelve months. The market has de-escalated; the household bill has not caught up. Open claims P-0051/P-0061/P-0079 already cover this axis. — Oil has calmed down; last year's rise is still in the price of petrol.
• Tariff passthrough — NO confirmation yet, and that is informative. Core goods CPI rose 0.2% m/m, its firmest since February, but is still only +0.8% yoy — three weeks after the Section 301 forced-labour tariffs took effect on 2026-07-24. Passthrough is not visible at the index level. Open claim P-0058 covers the PCE core-goods version. — The new tariffs have not clearly shown up in shop prices yet.

Research only — not financial advice. The practice portfolio is for learning only; this is a read-only risk briefing meant to help explain risk, never tell anyone what to buy or sell.
Project snapshot — changes waiting for review: 16 · tracked forecasts open: 51 (checking now: 3) · track record so far: 0.118227 · practice portfolio updated: 2026-08-12.
Past briefs & dashboards: market-sentinel-lyart.vercel.app/reports