Image Pipeline Experiment

An experiment built to answer one question: at what archive size does build-time image processing stop being the right default?

Method

The same 240-image archive is processed three ways: build-time resizing, on-demand transformation with cache warming, and a hybrid that only processes recent images locally. Build time, first-request latency and total bytes transferred are recorded for each.

Current status

Results are still being collected. The early signal is that the hybrid wins on build time and loses on complexity, which is the expected trade and the reason this remains an experiment rather than a recommendation.

Read Lab Note 018 for the running notes.