Long-form publications

Amazon Marketplace is one of the most complex pricing environments in commerce. This paper argues that the difficulty is not primarily a pricing problem, and that treating it as one is why so much marketplace automation erodes margin while working exactly as designed.

Version 2.0 develops two properties the first version left implicit. State relativity: the commercial meaning of a signal is not a property of the signal but of the configuration the firm is in — the same price premium is associated with a materially different featured-offer outcome depending on delivery position, and a fixed monetary undercut is a large move on a cheap product and an imperceptible one on an expensive one. Decision dynamism: a prescription is a time-bounded object with an entry state, a horizon, an observable market response and a revision rule, in an environment where both the market and the deciding policy are non-stationary.

The unit of analysis is accordingly the full chain — state, action, market response, settled commercial outcome — and the paper distinguishes an observation from a decision, an accepted execution, a market response and a settled result. Every quantitative statement was re-derived directly from the operational database at a stated cut-off, with its unit of observation.

It also reports what the evidence does not support. A re-derived response surface is presented together with the selection bias that limits it; a change in the observation instrument inside the analysis window is documented rather than smoothed over; and the question of whether this approach performs better than conventional reactive repricing is posed as an open research question, not answered. Claims from Version 1.0 that could not be reproduced are listed as withdrawn in Appendix C.