In the 1960s Stanley Milgram posted folders to strangers in Nebraska and asked them to forward the folders, first name basis only, towards a stockbroker in Boston. Sixty-four of the 217 chains that were started arrived, and those passed through a mean of 5.2 intermediaries (Travers and Milgram 1969). Read more

A corpus of 51,864,055 crawled LinkedIn job postings, covering 23,970,734 distinct adverts across six monthly snapshots from February to July 2026, produces statistically significant results almost on demand. With cells of that size, a difference of a tenth of a percentage point clears any conventional threshold. Read more

A rental listing carries an asking price, and when a landlord changes that price the change is recorded. Collect enough of those records and two different questions become answerable. The first is about the market: how fast are rents rising, and where. The second is about behaviour: who decides to change a price, and when. Read more

A question-answering system over a fixed set of documents looks like a solved problem. Give a language model a search tool, some pages, and an instruction to cite what it used. The answers come back fluent, linked and plausible. Plausible is the difficulty. A wrong answer here is not garbled. Read more

slurmpast is an open-source command-line tool that reads what a set of finished jobs actually did, and states what the next submission should ask for. It is the counterpart to a tool introduced here earlier: slurmwatch reads a job while it is running; slurmpast reads the accounting record after it has stopped. Read more

Every job submitted to an HPC cluster begins with a guess. The submission script names how much memory the job may use and how many CPU cores it may occupy; Slurm, the scheduler that hands out the cluster’s machines, reserves exactly that and runs the job. Read more

Abstract. Public discussion of the artificial-intelligence labour market rests on counts of “AI jobs”, but the counting itself is rarely examined. Using 51,864,055 LinkedIn job postings collected monthly from February to July 2026 (23,970,734 of them distinct), I show that three quantities routinely conflated differ by a factor of six: 10. Read more

Abstract. Official measures of rent inflation are accurate and slow. This article builds a unit-level repeat-rent index for the United States from 12.8 million rental listings, using the repeat-sales design of Bailey, Muth and Nourse (1963) with the variance correction of Case and Shiller (1987), and evaluates it against two external benchmarks. Read more

Abstract. Twelve US states require employers to publish a pay range in the job advertisement itself. I measure compliance in a corpus of 51,864,055 LinkedIn job postings collected monthly from February to July 2026, of which 23,970,734 are distinct and 14,250,750 are locatable to a US state. Read more

A companion article described how Argonne 3.5-base was pretrained: the same 2.88-billion-parameter architecture as Argonne 3.0, unchanged component for component, retrained on 88.84 billion tokens of web text and mathematics with a corrected learning-rate schedule, and finishing with a context window of 13,568 tokens that was trained rather than assumed. Read more