-
Machine learning · Measurement
There was a random number hiding inside my measurement
I reported that a descriptor loses its value when computed cheaply. The collapse was a random draw my own solver made, and the check meant to catch it passed.
-
Machine learning · Measurement
Averaging threw the answer away before the model ever saw it
The number my model predicts is an average, and averaging deletes what tells two shapes apart. On corrected data the recovery is small: +0.02 at worst.
-
Supply chain · Measurement
The bouquets got better and sales fell
Shelf life went from half a week to 14–21 days and sales fell anyway. Customers could see the size of a bouquet. They could not see the freshness.
-
Machine learning · Measurement
The metric was measuring my model, not the problem
A transferability score held at 0.54 across two independent budgets, then collapsed to 0.14 when I changed the model class. It was tracking the algorithm.
-
Machine learning · Measurement
The training set moves the error 390x. Nothing predicts it.
Across 231 training-set choices the worst case moved by a factor of 390. I tested nine cheap ways to pick a good one. None of them held up.
-
Coding theory · Supply chain
A scratched label still scans, and that is not luck
A check digit tells you a barcode scan was wrong. Reed–Solomon parity tells you what it should have been. That difference decides what a damaged label costs.
-
Supply chain · Measurement
You cannot count the sale you did not make
Sales data records what you sold, not what people wanted. The two differ exactly on the days they matter most, and the gap teaches a forecast to shrink.
-
Machine learning · Supply chain
My model scored perfect and that was the problem
A perfect held-out score measured one thing: whether the model could repeat what it had seen. Of 30 cross-family tests, 20 fell below a crude baseline.
-
Vision science · Display engineering
Why a screen cannot measure vernier acuity
A screen pixel at 60 cm subtends about 90 arcseconds; clinical vernier thresholds are 2–5. Antialiasing still measures below one pixel. Here is the arithmetic.
More to come, from the same places the work already is: holding a thousand live WebSocket connections in one Cloudflare Durable Object at no infrastructure cost, running one SQLite schema and one dialect across a native desktop build and WebAssembly in the browser, and the point at which a forecasting or optimisation model quietly stops being trustworthy.