Sell Monitor

Reads my own live listings every day, prices each against the market, and flags only the ones that are mispriced.

The best comparable price is one of your own past sales.

The sell-side twin of the buy-side sweep. It reads what I currently have listed, comps each piece against the live market, computes margin, and tells me which ones to reprice, relist or drop.

How it is built

It shares a pricing engine with the buy-side AI agent, one module, called by both, so a correction to how costs are modelled fixes buying and selling at the same time. Keeping two copies in sync was never going to work.

The comparable-price hierarchy, in order of trust:

1. My own realised sales of the same reference and size
2. Live asking prices, discounted for the gap between asking and selling
3. Market-data estimates

That order took a while to earn. Asking prices look authoritative because they’re abundant, but they’re what sellers hope for, not what buyers paid.

What it got wrong

It compared the wrong sizes. A realised sale is only a comparable if the reference and the case size match. One mismatch quietly inflated the valuation of a much smaller watch, and because the source was high-trust, nothing downstream questioned it.

It double-counted a discount. The engine applied a cheap-end adjustment and a liquidity adjustment to the same piece. Each was defensible alone; together they killed a trade that was actually fine.

The lesson

When a number is assembled from several corrections, the corrections need an owner. Ours were added at different times by different reasoning and nobody had ever written down which ones were allowed to stack.