Farm Fresh Insights

Sign in Subscribe

Latest — 22 Jul 2026

Operating a Retrieval-Augmented Classifier

Consider a design: a query gets embedded, retrieval pulls the nearest labeled examples from a vector store, and an LLM assigns a label from a fixed set. Mechanically this is kNN classification with an LLM head: the store is the training set, retrieval finds the neighbors, and the LLM replaces

More issues

Notes from a fan

A few days ago I was looking at Wirecutter's best-fan guide. The Vornado 630 was \$60 on Amazon, \$80 on Walmart, and \$80 on Vornado's own website. A 33% gap on a fan caught my attention. I wondered how common gaps like this were. So

Science: 20 GPs in a single District

One of the quieter successes of post-Independence India is that the gender gap in education has closed. The trajectory is plain in the historical data. In 1951, fewer than three in ten girls were enrolled at the primary level, against six in ten boys; at upper primary, girls were

Ceci N'est Pas un Proxy

Chaturvedi, Das, and Mahajan (2024) argue that female reservation in the 2015 Gram Panchayat elections raised household toilet provision more in villages with a higher Muslim share. At low Muslim share the estimated reservation effect is essentially zero. In villages where Muslims make up about a quarter of the population,

The Ties That Bound

The previous post described a conservative record-linkage pipeline: accept only high-confidence matches, flag ambiguous cases, and leave the rest unmatched. That is the right approach when false matches are more costly than missed ones, but it means the matched subset is selected on ease of matching, and estimates

Join, or Die

Record linkage is the problem of identifying the same entity across two or more tables that lack a shared unique identifier. You have a name, maybe a district, maybe an age, and you need to figure out which row in table A corresponds to which row in table B. When

Recommended Treatment

The effect of a treatment on an outcome depends on three things: the kind of treatment, the kind of person receiving it, and the context in which it is delivered. A ranking change and a button color change work through different mechanisms. A heavy user and a new user will

No Logbook, No Problem: Recovering LATE When You Can't Observe Compliance

Suppose you run a randomized experiment with imperfect compliance. You assigned treatment, measured outcomes, and now you want the LATE. The Wald estimator says divide the ITT by the compliance rate, $\pi = E[D_i \mid Z_i = 1] - E[D_i \mid Z_i = 0]$, and you're