The Brief

Trump floated a 20% toll on ships transiting the Strait of Hormuz on Monday, framed as reimbursement for US security provision, then dropped it within a day after pushback from Gulf allies and from businesses and markets questioning both its workability and its effect on energy costs. The retreat came as Centcom completed a seventh wave of strikes on Iranian missile, drone, naval and coastal-defence targets, and with Brent crude up 15% over the week following Tehran’s 12 July announcement that it was closing the Strait.

The toll collided with the administration’s own arithmetic. Data released Tuesday showed inflation fell last month on the back of cheaper oil and petrol — a decline the ceasefire produced by reopening the Strait and raising expectations of freer Gulf supply. A transit levy would have taxed the precise chokepoint that delivered the disinflation, with the cost landing in freight economics and, on a short lag, at the American pump. That is why the reversal took a day rather than a fortnight: the transmission line from Hormuz throughput to petrol prices to the November midterms is short enough that the price of the policy was legible immediately.

The asymmetry sits in who could impose the cost, not who bears it. Gulf allies and market participants together killed a US policy in 24 hours — a demonstration that Washington’s leverage over the waterway is bounded by the fact that it also needs oil moving through it cheaply. Shippers and energy buyers get the near-term reprieve; the constraint that saved them is the same one now visible to Tehran, which can read the reversal as evidence that pressure applied to Strait throughput is expensive to the White House on a fixed political timetable.

Gradlines’ position is that the toll’s death priced the constraint in public, and the constraint tightens from here. Brent’s 15% move has already reversed the disinflation the ceasefire bought, and every week closer to November raises what any further Hormuz-linked escalation costs the administration in the one variable it cannot hedge. Expect the pattern to repeat: measures that touch the Strait’s economics get proposed and withdrawn faster as the election approaches, which makes the retreat interval — not the strike tempo — the thing to watch.

China’s economy grew 4.3% year-on-year in the second quarter, down from 5.0% in the first and under the floor of Beijing’s 4.5%–5.0% full-year target, as fixed-asset investment contracted 5.7% year-on-year over January–June — state-sector investment included, at −2.3% — while exports jumped 27% and June retail sales rose just 1.0%.

Composition is the story, not the miss: first-half growth of 4.7% sits inside the target band, which is why analysts expect the end-July Politburo to stop short of major steps given debt concerns. What has changed is that state-sector investment — the lever Beijing pulls when the cycle turns — is now shrinking alongside private capex, because the local governments that channel it are under orders to cut costs, leaving exports as the only load-bearing source of demand.

Property and construction-linked firms take the clearest hit, with property investment down 18% over the half and home prices easing, as does anything dependent on local-government spending. But exporters riding the 27% are not the safe side of this: the figure is inflated by US retailers racing to stock Black Friday and Christmas shelves before tariffs rise, with the universal 10% levy expiring on 24 July and the USTR proposing 12.5%. That demand has been pulled forward, not created.

Macquarie’s Larry Hu argues Beijing moves on domestic demand only when exports fail — which makes the end-July Politburo the wrong meeting to watch. The frontloading now flattering the export line is borrowed from the fourth quarter, so Q4 is when the conditional gets tested: shipments normalising into a higher tariff, a domestic engine already contracting, and no counter-cyclical buffer running. That is the point at which Beijing’s reluctance starts to cost it the growth target.

Technology secretary Liz Kendall announced that from next spring, 16- and 17-year-olds in the UK will have a midnight–6am block on certain social-media apps switched on by default, alongside default deactivation of autoplay and personalised content feeds — all overridable, the latter in a couple of clicks. The apps covered have not been named. The measure extends last month’s under-16 ban, which named Snapchat, TikTok, YouTube, Instagram, Facebook and X, and the government declined to restrict VPNs, citing its own finding that 7–10% of children use them to bypass age checks and concerns about free speech.

The evidence and the policy were built to different specifications. DSIT commissioned Savanta in May to test three restriction models; curfews came out most enforceable, and the sleep benefit was strongest and most consistent when Snapchat, TikTok, Instagram, Facebook, X and Reddit were blocked — from 9pm to 7am, with YouTube, Teams, Google Classroom and messaging left open. What ships is four hours shorter, starts at midnight, names no apps, and can be switched off. The reported gains — earlier bedtimes, better concentration, less stress, and cumulative in participants’ own accounts — were observed under a ten-hour window that captured the evening. The policy captures the hours after it.

With no app list, no platform can scope the work. The build differs by an order of magnitude depending on whether “certain apps” means the six Savanta tested, the six named in the under-16 ban, or something wider — and those lists are not identical. Reddit sits in the trial set and not in the ban set. Platforms carry that ambiguity until the naming lands next spring, which caps what any of them can put into product plans now; the default-off treatment of autoplay and personalised feeds is a design change with a known shape, but no defined perimeter. The cost is deferred, not avoided.

Gradlines’ position is that the gap between trial conditions and shipped conditions makes underperformance against the sleep outcomes the likeliest result, and that this is what matters commercially. A default measured at ten hours and delivered at six, with an override, is unlikely to reproduce the effects it was sold on. When the evaluation arrives, thin results will read as evidence that defaults are insufficient rather than that this default was mistimed — which is the argument that carries a statutory version. The compliance question for platforms is not next spring’s build. It is whether next spring’s build becomes the baseline for a mandatory one.