A Better Newspaper

US and Iran Trade Strikes as Hormuz Status Is Disputed — But Oil Keeps Moving

The US struck Iran for a third time in a week, prompting retaliatory attacks on at least five Arab nations, as Washington and Tehran issued conflicting declarations over whether the Strait of Hormuz remains open. Despite Iran declaring the waterway closed, a maritime advisory group said the strait's southern route remained open to shipping on Sunday. Bloomberg analysts note that technical talks on Iran's nuclear program continue, but disputes over Hormuz, sanctions, and implementation of the memorandum of understanding keep negotiations at an impasse, with intermittent strikes likely to persist.

Context: Roughly a fifth of global oil passes through Hormuz, so the gap between Iran's rhetorical 'closure' and the operational reality is the number that matters for energy markets. One Bloomberg reporter cautioned that Tehran could emerge with greater strategic leverage if it retains effective control over the waterway — a slow-burn risk premium rather than a supply shock, for now. Prediction markets still price a US ground invasion of Iran as unlikely.

Polymarket: US invades Iran before 2027 18%

Modi Courts Indo-Pacific Partners as China and US Reshape the Region

Prime Minister Narendra Modi used a three-nation tour to deepen India's Indo-Pacific partnerships, spanning missiles, uranium, critical minerals, and sports diplomacy. The push comes as regional powers strengthen ties amid China's growing influence and uncertainty over US engagement.

Context: India is positioning itself as the anchor of a 'China-plus-one' supply chain realignment and a critical minerals alternative to Chinese dominance — themes that directly bear on manufacturing and sourcing decisions for US firms diversifying away from China. The hedging against 'uncertain US engagement' is the tell: allies are increasingly building around, not only with, Washington.

McCaul Presses for Deeper US-Ukraine Defense Cooperation

Rep. Michael McCaul said Ukraine is gaining momentum on the battlefield but urgently needs more Patriot interceptors and stronger air defenses to counter Russia's escalating missile attacks. He urged deeper US-Ukraine defense cooperation, including licensed Patriot production and joint drone development, arguing that sustained military pressure offers the best path to negotiations with Russia.

The Governance Layer Gets a Rulebook: Agent Risk-Tiering Frameworks Arrive

Researchers introduced the TrustX Agent Risk Classification Framework (ARC), a structured, repeatable instrument for classifying seven types of internally built agentic AI systems. It uses a twelve-dimension scoring rubric plus a five-level autonomy model to produce a three-tier governance output with mapped controls. A parallel paper, CogniConsole, argues that LLM reliability is driven less by model capability than by 'inference-time control,' showing that adding structural scaffolding around a model systematically reduces output variance and failure rates.

Context: This is the academic scaffolding for the enterprise AI control plane category Nutanix and Dell are already positioning around. When practitioners can point to a named rubric mapped to controls, it becomes the language of compliance sign-offs, vendor RFPs, and eventually EU AI Act audits — the underbuilt niche is turnkey products that operationalize exactly this. Watch who commercializes the framework before it becomes a de facto standard.

The 'Patchwork Problem': Why AI-Generated Code Passes Tests and Still Breaks in Production

A new paper formalizes what it calls the 'patchwork problem' in LLM-generated code: output that compiles, passes tests, and looks correct but fails in deployment because of structural incoherence — endpoints referencing undeclared config keys, imports targeting nonexistent packages, or new routes silently omitting the authentication guard applied to every sibling endpoint. The authors note standard CI toolchains rarely surface these failures and model structural coherence as consistency invariants over a graph representation of the codebase.

Context: As agentic coding tools (Codex, Claude Code) proliferate, this is the emerging liability surface — the missing auth guard is a security incident waiting to happen, and it slips past the exact tests companies rely on. The commercial gap: a validation layer that catches globally incoherent-but-locally-valid code is a natural bolt-on for enterprises deploying AI coding at scale, and a defensible niche independent of the model vendors.

Europe's Sovereign Model Play: Soofi S Undercuts on Cost, Not Just Politics

A large German-led consortium released Soofi S 30B-A3B, a sovereign, open-source Mixture-of-Experts hybrid Mamba-Transformer model for German and English. Its design activates only 3B of 30B parameters per token and keeps the inference cache near-constant as context grows, which the authors say gives it a throughput advantage for long-context, high-concurrency deployment. They report it matches dense 14–27B models on English and German benchmarks and beats every European sovereign baseline in their comparison.

Context: The strategic point isn't nationalism — it's that a sovereign model with a genuine inference-cost advantage is a real procurement option for EU regulated buyers who need data residency and AI Act defensibility. The MoE-plus-Mamba architecture choice is a tell: efficiency for high-concurrency serving is where sovereign efforts can actually compete rather than trying to out-scale US frontier labs.

The Models Have a Geopolitical Thumb on the Scale

In an endorsement experiment, four LLMs evaluated identical international economic and security policies after each was randomly attributed to the US, EU, China, or Russia. In the numeric-only condition, GPT-5, Claude Sonnet, and Gemini rated China- and Russia-endorsed policies substantially lower than identical policies attributed to the US or EU; DeepSeek was the main exception.

Context: For anyone deploying LLMs in policy analysis, due diligence, competitive intelligence, or investment memos, this is a documented, source-dependent bias that survives even when only the sponsor's identity changes. It's a live litigation and fairness-audit exposure — and a reason regulated users will increasingly demand model provenance and bias disclosures as part of procurement.

Auditing AI for Malicious Capability Without Triggering It

MIT researchers developed an auditing technique that tests generative AI models for malicious capabilities — such as producing illegal content — without prompting them to actually generate the harmful outputs. The method is aimed at assessing dangerous capabilities for safety and compliance purposes.

Context: Non-generative capability assessment is exactly the tooling regulators and enterprise procurement teams need to evaluate dual-use models (recall Anthropic withholding Claude Mythos over its vulnerability-discovery power) without the legal risk of generating the forbidden material themselves. Expect this class of audit method to get baked into model-release compliance and third-party certification services.

Generative AI Boosts Test Scores, Not Deep Skill — With a Delayed Twist

A randomized, in-person experiment found that undergraduates given access to off-the-shelf generative AI while learning an unfamiliar topic scored 0.27 standard deviations higher on immediate knowledge tests, gains that persisted a week later. Essay quality changed little while students had AI access but improved in style and relevance a week later when they wrote unaided, with the delayed gains larger among 'augmentation' users.

Context: The relevant read for any founder building AI into training, upskilling, or knowledge-work products: measurable learning gains are real but the higher-order benefit shows up later and depends on how the tool is used (augmentation vs. substitution). That distinction is the design principle that separates products that build durable capability from ones that create dependency.

Korea's Rally Is Real — and the Stocks Are Cheaper Than Ever

Bloomberg reports South Korea's stock rally has left investors with an unusual setup: equities are trading at record-low valuations despite world-beating performance. Earnings and index gains have outrun the price multiples investors are willing to pay.

Context: The 'Korea discount' — chronic underpricing tied to chaebol governance and weak minority-shareholder protection — has been a structural feature for decades. A world-beating rally that still leaves multiples at record lows is the arbitrage tell: either governance reform re-rates the whole market higher, or the discount is permanent. For a value hunter, this is the classic 'cheap for a reason vs. cheap because unloved' question, and the answer hinges on Seoul's corporate-reform follow-through.

SpaceX's IPO Makes Founder Control the Public-Market Test Case

Bloomberg reports SpaceX's record-setting IPO gave public investors access to the company but not equal voting rights: Elon Musk holds just over 40% of equity while controlling more than 80% of the vote. IPO adviser Lise Buyer argues dual-class shares shield founders from short-term pressure, while Harvard's Lucian Bebchuk warns SpaceX's structure raises risks around accountability, succession, conflicts of interest, and shareholder value. Danish pension fund AkademikerPension has rejected the stock, citing 'catastrophic governance.'

Context: The signal for markets: superstar-founder IPOs are ratcheting the equity-to-vote gap ever wider, and index inclusion forces passive money to hold governance structures it would never actively choose. Watch whether major index providers or large pensions start carving out or discounting extreme dual-class names — that's where a governance-arbitrage or activist-adjacent opportunity emerges.

Primary Dealers Go Record Net-Short Treasuries — Nobody's Sure Why

Bloomberg reports that for the first time in recent history, primary dealers have gone net short on Treasury debt they used to hold billions of dollars of, sparking a search for explanations among market participants.

Context: Dealers flipping net-short the debt they're mandated to backstop is a structural stress signal — it can reflect hedging around basis trades, balance-sheet constraints, or a positioning bet against duration into the Fed's next moves. Prediction markets still lean heavily toward multiple 2026 cuts, so if dealers are shorting into an easing cycle, the tension is worth watching: it hints someone thinks the long end reprices regardless of what the front end does.

Polymarket: Fed makes multiple rate cuts in 2026 78% · Fed cuts at July 2026 meeting 0%

Stocks Priced for 'Sunshine and Rainbows' Meet the Earnings Test

Bloomberg reports Wall Street is anticipating a near-record earnings season, but investors face the question of whether results will be strong enough to sustain the bull run given how much optimism is already in prices.

Context: When a market is priced for perfection, the risk is asymmetric: beats are already discounted, misses get punished. Pair this with dealers shorting Treasuries and it reads as a market where the crowd is long risk and short duration simultaneously — a positioning that unwinds violently if earnings disappoint.

Aldi's $9bn Urban Push: Can Hard-Discount Crack Manhattan?

BBC News reports German discounter Aldi is deploying a $9bn US expansion targeting dense urban hubs like Manhattan, leaning on private-label products such as $4 almond butter to undercut incumbents. The piece frames the open question of whether its discount model can match Walmart's scale in American grocery.

Context: The opportunity read is in the model, not the stock (Aldi is private): hard-discount, ~90% private-label, small-footprint stores are moving into high-rent urban markets that legacy grocers priced as impossible for discounters. If Aldi proves unit economics work in Manhattan, it validates a replicable template — tight SKU counts, vertical private label, dense urban micro-formats — for anyone building in adjacent retail categories.

DOJ Subpoenas NYT Reporters Over Air Force One Story

The Committee to Protect Journalists is demanding the Trump administration withdraw subpoenas targeting several New York Times reporters after the paper's coverage of security concerns involving the Qatari-donated replacement Air Force One. The subpoenas were issued by Jay Clayton, the U.S. attorney in Manhattan.

Context: Clayton, the former SEC chair, using the subpoena power against reporters is notable given the DOJ's own recent history of narrowing when it will compel journalist testimony; watch for a motion to quash that could test the current reporter's-privilege posture.

A New Drug Target for Parkinson's: Blocking the Proteins That Spread the Disease

Yale scientists identified two neuron surface proteins that appear to help spread the toxic protein linked to Parkinson's disease through the brain. Blocking these proteins in mice dramatically reduced disease progression, offering a potential new therapeutic target.

Context: Parkinson's therapeutics have historically managed symptoms rather than the disease itself. A mechanism that halts cell-to-cell spread of pathology — if it survives the leap from mice to humans — is the kind of disease-modifying target that reshapes a multibillion-dollar market and attracts serious pharma dealmaking.

Tau's Double Life: The Alzheimer's Villain Is Also Essential for Memory

Researchers found that tau — long known as a driver of Alzheimer's — is actually essential for turning new experiences into lasting memories by helping organize the brain's memory-storing cells. The mouse study showed how abnormal tau may contribute to Alzheimer's by disrupting both the formation of new memories and the recall of existing ones.

Context: This complicates the drug developer's playbook. If tau has a legitimate day job, the many anti-tau therapies in the pipeline may need to suppress the pathological form without knocking out the normal one — a narrower, harder target than simply clearing the protein.

A Common Antidepressant Class May Accelerate Heart Valve Disease in Some Patients

Columbia scientists found evidence that serotonin may speed the progression of degenerative mitral regurgitation, a common heart valve disease. Patients taking SSRI antidepressants who also carry a specific genetic variant may develop severe valve damage sooner, potentially requiring surgery at a younger age.

Context: SSRIs are among the most prescribed drugs on earth. A gene-drug interaction that flags a subset of patients for accelerated valve damage is exactly the kind of finding that drives pharmacogenomic testing markets and, eventually, prescribing-label scrutiny.

Why Gold Never Tarnishes — and Why That Matters for Catalysis

Researchers discovered that atoms on gold surfaces reorganize into patterns that block oxygen from reacting with the metal, suppressing oxidation by up to a trillion-fold. Beyond explaining why gold jewelry stays bright for generations, the finding could help scientists design more powerful gold-based catalysts for manufacturing and clean energy.

Context: Gold catalysts already appear in emissions control and chemical synthesis. Understanding the atomic self-protection mechanism gives materials engineers a lever to tune surface reactivity deliberately — the difference between an interesting fact and an industrial design tool.

1971 Mercedes-Benz 280SE 3.5 Cabriolet — Blue-Chip W111 Rarity

One of roughly 1,230 examples produced between 1969 and 1971, refinished in blue over gray with a 3.5-liter V8 and four-speed automatic. Equipment includes power windows, air conditioning, and disc brakes; a 2025 valve adjustment saw the fuel pump and various ignition components replaced. Offered with owner's manual, tool kit, and a clean California title in the name of the owner's LLC.

Context: The W111 3.5 cabriolet is one of the most coveted postwar Mercedes bodies, with strong examples routinely trading well into six figures — production scarcity plus timeless styling makes it a genuine store of value, not a depreciating toy.

1970 TVR Vixen 2500 — Possibly the First Built, Only Factory Fuel-Injected Example

This TVR Vixen 2500 is believed to be the first of approximately 385 examples produced and the only unit fitted from the factory with Lucas mechanical fuel injection. Chassis LVX1745/6T was reportedly used by TVR's US importer Gerry Sagerman as a promotional vehicle before relocating to California for refurbishment about a decade ago. Finished in yellow over black with a 2.5L Triumph inline-six, four-speed manual with overdrive, adjustable Spax coilovers, and front disc brakes.

Context: Provenance like this — plausible first-of-series plus a genuinely unique factory spec — is what separates a quirky British special from a collectible. For a fraction of the cost of mainstream classics, this is a car with a documentable story nobody else can claim.

2023 Toyota GR Corolla Morizo Edition — 1 of 200, Instant Collectible

One of 200 Morizo Editions produced for the US market, bought new by the current owner and showing 28k miles. Finished in Smoke over black with red accents, powered by a turbocharged 1.6-liter inline-three driving all four wheels through a six-speed manual, front and rear Torsen limited-slip differentials, and the GR-FOUR AWD system. Equipped with 18" forged BBS wheels and a forged carbon-fiber roof.

Context: The two-seat, lightweight Morizo was the halo of the GR Corolla line and capped at 200 US cars — a modern homologation-style special that's likely to appreciate as the manual, analog performance era ends. The mileage keeps it usable rather than museum-frozen.

18k-Mile 2008 BMW M3 Coupe 6-Speed — The Last V8 M3, Barely Driven

An E92 M3 Coupe delivered new to Vern Eide BMW in Sioux Falls, registered in South Dakota with its original owner until 2026, now showing 18k miles. Finished in Space Gray over Fox Red Novillo leather with the Premium and Technology Packages, powered by the 4.0-liter V8 with a six-speed manual and limited-slip differential. Offered with window sticker, accident-free Carfax, and a clean South Dakota title.

Context: The E92 is the only V8 M3 ever built, and clean low-mile manual examples have been steadily climbing as buyers recognize it as a future blue-chip. 18k miles on the odometer plus single-owner history is the exact combination the market rewards.

1973 Porsche 911T Project, No Reserve — Air-Cooled Entry With Papers

A non-running 1973 911T coupe in Silver Metallic over black, with a 2.4-liter flat-six, Bosch K-Jetronic injection, a 915 five-speed transaxle, sunroof, and 15" Fuchs wheels. Offered on dealer consignment at no reserve with a Porsche Certificate of Authenticity and a salvage Utah title.

Context: No-reserve air-cooled 911s with a COA are how patient buyers still get into the long-hood market below finished-car money — but the salvage title is the identifiable catch that will cap the ceiling and demands a hard look at the bodywork before bidding.