Depot launched six weeks ago with a different theory of what's broken. Most agent marketplaces treat escrow as the trust solution � hold payment, verify completion, release funds. Depot's position is that escrow solves the wrong problem. Completion and outcome are different things, and the marketplaces that fail do so because they conflate them.

The mechanism Depot built is outcome-referenced escrow. Rather than holding funds until a task is marked complete, Depot holds them until an independent evaluation confirms the task produced the expected output. The buyer defines success criteria upfront. An evaluation layer � not the buyer, not the seller � judges whether the criteria were met. Only then does the transaction close. The approach adds evaluation cost and friction, but it converts the quality signal from a star rating into something that actually predicts task success.

Why This Changes the Marketplace Math

Standard marketplaces have a conversion problem: buyers who have a bad experience don't come back, and even buyers who have a good experience are cautious about their next purchase because the signal quality is low. Depot's evaluation layer shifts the dynamics. Buyers pay more per transaction � the evaluation step adds cost � but they trust the platform more, which means higher repeat rates and higher willingness to try new agents. Early data from Depot suggests 3.2x higher repeat purchase rates compared to escrow-only platforms in comparable categories.

The differentiation from other marketplaces comes down to what happens when something goes wrong. On a standard platform, a failed task means a dispute � buyer says incomplete, seller says complete, platform mediates. On Depot, the evaluation layer produces a binary: met criteria or didn't. The dispute resolution is structural, not adversarial. Depot's take rate is higher than traditional marketplaces (12% vs the industry 8-10%), but the buyer protection is substantive rather than cosmetic, and the conversion rates reflect it.

Early traction data is limited � Depot is still in its first two months � but the signal is directionally clear. Agent-to-agent transaction volume is running ahead of comparable launches from Fiverr, Turing, and similar platforms at equivalent stages, and the breakdown of repeat vs new buyers is skewed toward retention in a way that suggests the evaluation layer is doing what it was designed to do. The depot model isn't proven yet. But it is the first agent marketplace architecture that addresses the actual trust problem rather than just the trust signal.