Machine Humanities

What this is

Every work here starts as a question and ends as a book-length corpus: chapters of narrative argument, a glossary, a set of figures drawn from the material itself, and a forecast that commits to a number. The subjects run from Connected-TV auction mechanics to Carlyle, Dickinson, and the Great Awakenings, because the method does not care what it is pointed at.

Alongside the library sit three standing desks: Ad Tech, a trade desk covering the programmatic and Advanced-TV market with a daily wire; The Ghost of Times, a paper of writer-voiced op-eds; and The Forecast Desk, where the predictions are priced, logged, and graded against what actually happened.

How it is made

The pipeline is the point. It is not a chatbot asked for an essay — it is a research process with adversarial stages built into it:

1. Multi-round research. Successive sweeps widen coverage, then a critic names what is still missing and sends the process back out to fill the gaps.
2. Adversarial verification. Claims are re-checked against independent sources by readers whose job is to refute them. Refuted claims are corrected in place, not quietly dropped — several corpora here carry dozens of corrections made this way.
3. Priced forecasts. Every probability is the output of a process — a market price, a counted base rate, or a decomposition — never a number someone liked the look of.
4. Public grading. Calls are frozen when logged and scored when they resolve. Misses stay on the board next to the hits, with Brier scores and a calibration curve.

What it is evidence of

Read commercially, this site is a portfolio of three things: designing agentic systems that do real intellectual work end to end and survive contact with their own errors; deep, current coverage of the programmatic and Connected-TV ad market, from auction plumbing to identity to regulation; and calibrated forecasting with a track record kept in the open rather than a deck of confident claims.

The Track Record is the honest version of that argument. It is public precisely because it includes the losses.