Portfolio Review

A weekly pass over everything currently owned and everything being watched, re-priced against today's market.

A stored valuation is a claim with an expiry date.

Once a week, everything gets re-checked: what is owned, what is listed, what is being watched. Each is re-priced against the current market and the deltas are reported.

How it is built

It reuses the same pricing engine as the buying and selling AI agents, so there is exactly one definition of what something is worth. Every valuation it writes carries its basis, which comparables, how many, from where. A number without a basis is treated as unverified rather than as zero.

What it got wrong

It re-priced things it had no new information about. Early versions recomputed everything every week, which meant a stored valuation could drift purely because the comparable set had been reshuffled. The rule now is that a re-price requires new evidence, not just a new week.

Appending one comparable silently moved other valuations. The comparables are shared, so adding one for provenance could change the price of unrelated items through the matcher. Every append is now diffed against the whole set.

The lesson

Shared state between AI agents is where the subtle bugs live. Two AI agents reading one pricing table is correct; one AI agent writing to it without checking what else moved is how you get a quiet, confident, wrong answer.