This week’s reading list is about the distance between a number and the thing that number is supposed to describe. Markets have spent August being reminded that a statutory loss is not a cash outflow, that a record profit can be a price outcome rather than an operating one, and that the measure everyone quotes is rarely the measure that binds. Several of the readings look at what happens when investors mistake conviction for accuracy, or a backtest for a live position. Others examine the same problem in physical form: a data centre lease that runs four years against debt that runs to 2049, a racehorse bred so precisely for speed that the breed lost the diversity it would need to keep improving, an ice business that looks like a commodity until you notice that nobody can ship it more than a few hundred kilometres. There is also a piece on how quickly the economics of the open web have been rewritten beneath everyone’s feet, and one on where subjective experience might actually sit in the brain, because not every good question has a ticker attached. The useful lesson for investors is that the answer to one question is not the answer to all of them, and a locally successful outcome is not a complete one. The best investors are not trying to predict every headline. They are trying to work out which clock is running, and which one runs out first.

📚 The Situational Awareness Fund Blow-up: Collateral Damage from Investment Conviction: Aswath Damodaran’s post-mortem on the Aschenbrenner implosion, and the best thing written on the episode. He is less interested in the AI thesis than in what conviction actually is, where it comes from, and why it so reliably ends in position sizes and leverage that the underlying idea cannot support. His three lessons, on leverage truncating your time horizon, on momentum as the uninvited passenger in every strategy, and on humble money beating smart money, are worth the read on their own. (Musings on Markets)

📚 The Clocks on One Building: Four-Year Leases, Sixteen-Year Guarantees, Debt to 2049: Shanaka Anslem Perera reconstructs the financing architecture behind Meta’s AI data centre ventures from the public filings, and finds not one duration but several running side by side. A four-year initial lease, options out to twenty years, a residual value guarantee declining over sixteen, and amortising notes maturing in 2049. The point is not that something is wrong. It is that credit recovery, accounting control, physical succession and workload continuity are four different outcomes, and success on one is routinely reported as success on all of them. Long, but the most useful thing we have read on AI infrastructure risk. (Shanaka Anslem Perera)

📚 How Hard Can Quant Trading Really Be? I Tried It to Find Out: Lauren Leek, a political economy PhD with Python and free data, builds three quant strategies on a laptop and reports the results honestly, including the machine learning model that looks spectacular until you ask about beta, drawdowns and execution assumptions. Her conclusion is the right one: the professionals’ edge is not foresight, it is data, infrastructure and the discipline to build systems that do not break when conditions change. The comment from Dan Davies underneath is almost worth the price of admission by itself. (Lauren’s data Substack)

📚 How Trump’s Next Tariffs Could Drive Companies Back to China: Chad Bown at the Peterson Institute traces what actually happened to smartphone and clothing supply chains under the 2025 tariffs, and finds that what mattered was never the level of the tariff but the differential between countries. The looming irony is that the next round, paired with lower rates on nonsensitive Chinese goods, could push sourcing back toward exactly the place the policy was designed to move it away from. A clean example of second order effects doing the opposite of what the first order intended. (Peterson Institute)

📚 The Ultimate Horse: Sophie Fessl on the thoroughbred paradox. Breeders have spent three centuries selecting for speed, and race times have not meaningfully improved since about 1910. Roughly 95 per cent of living thoroughbreds descend from a single eighteenth century stallion, ten heavily crossed ancestors account for more than 80 per cent of inbreeding in the Australian thoroughbred population, and the inbreeding itself correlates with shorter and less lucrative careers. Optimise hard enough for one variable and you eventually breed away the diversity that would let you keep improving. Portfolio managers may find that uncomfortably familiar. (Works in Progress)

📚 Inside the Billion-Dollar Industry That’s Keeping Your Beer Cold: Lauren Larson on Reddy Ice, packaged ice and one of the more instructive business models you will read about this year. Ice cannot be shipped economically beyond a certain radius and must be kept cold on arrival, which means the industry stays stubbornly local no matter how large the operator gets. The chief financial officer’s line, that the business is not recession proof but recession resistant, is a better summary of defensive earnings than most sell side notes manage. Note also the roll-up strategy, the consolidation of a competitor this year, and what ice sales reveal about when people are and are not gathering. (Texas Monthly)

📚 What Was the Internet? Five writers on how quickly the economics of the open web have been rewritten as AI answers replace links. Siva Vaidhyanathan’s contribution has the numbers that matter for anyone modelling a media, marketplace or search dependent business: click-through on top ranking pages down sharply where AI summaries appear, and zero-click rates across search now estimated at roughly 60 to 68 per cent of queries. Whatever your view of the essays’ politics, the structural change they describe is a real revenue event for a lot of listed companies. (Boston Review)

📚 The Tantalizing Possibility of Locating Consciousness in the Brain: Daniel Freeman, an MIT scientist, on what artificial intelligence has done to a long-standing assumption in neuroscience. If machines can now reason, plan and joke without anything resembling a biological neuron, then perhaps cognition and consciousness were never the same problem, and subjective experience may sit in far older and more anatomically discrete structures deep in the brain. No investment application whatsoever. We include it anyway. (NOEMA)