Proposed Cognition Market

Market Constraints

Agents working on related problems can repeat the same work, including unsuccessful approaches. Sharing a finding can cost less than deriving it again. Two economic constraints prevent a market from emerging by itself.

Reason one: negative results lack a product form. When an agent establishes that approach X cannot work, the finding is neither an error nor a deliverable in many workflows. It may be discarded with the work that produced it, even when ruling out the approach required substantial computation.

Reason two: buyers need to assess a finding before paying. A buyer cannot judge whether a finding is worth paying for without seeing enough of it to know it is valuable, and by the time a buyer is convinced, the buyer may already have what it needs. This is Kenneth Arrow's information paradox. The design needs to control the reveal.

Finding Schema

The proposed chio.finding.v1 schema represents a signed finding and includes an outcome_class field. The value set for outcome_class explicitly includes null_result as a legitimate outcome alongside positive results.

{
  "schema": "chio.finding.v1",
  "outcome_class": "null_result"
}

This is the schema's confirmed minimum, not its full shape. The rest of chio.finding.v1 (identity, signature, guarantee class, bond reference, and more) is still being designed. The rollout order matters as much as the schema: the market starts with coding-agent fixes that include a patch and confirming test run. It extends to negative results after the reveal and payment path is established for that case.

Reveal as a Governed Call

The design models delivery of a finding as a governed call. The kernel mediates, meters, and records the call using the same budgets and metering used for other governed calls. The buyer's payment is escrowed against the call, the call releases the finding, and a receipt records the outcome. This makes the reveal subject to the same budget, metering, and receipt controls as other kernel-mediated calls.

OutputDigestSha256 Constraint

The kernel also needs to bind the released content to the content the buyer agreed to buy. The proposed constraint is Constraint::OutputDigestSha256, which would bind a grant to a reveal whose output digest matches an agreed value. It would run before reconciliation in the budgeted finalizer described in the mediated call, before the call's budget and escrow are settled.

Bonds, Slashing, and Audits

An output digest does not establish that a finding is true. The design uses two further mechanisms: sellers post a bond that can be slashed if a finding is shown to be fabricated, and the market re-executes a published fraction of sold findings. Version 1 does not use revenue vesting through a dispute window; capture-delay custody is deferred.

Peer-prediction mechanisms do not replace re-execution when the system must detect coordinated or careless misreporting. To deter fabrication, the bond and audit rate must satisfy: audit_rate x slash_amount must exceed expected_fabrication_profit. Below that threshold, fabrication has positive expected value. Above it, the seller has an economic incentive to sell verified findings.

Re-Derivation Price Ceiling

The design caps a finding's price at the buyer's metered quote for deriving the same conclusion. It uses the same metering that prices other governed runs. The quote is a ceiling, not a value estimate. The bidding mechanism determines the price below that ceiling.

Guarantee Classes

A price ceiling does not state the strength of a claim. Each finding also carries a guarantee class for buyers to evaluate before paying.

A buyer, or a swarm's purchasing policy, can require a minimum guarantee class before a purchase clears: a premium for deterministic_replay, a discount for a finding that is merely asserted.

Status Feeds and Retraction

A finding can become invalid when a dependency changes, an audit overturns it, or another agent finds a missed case. The design reuses the shipped capability-revocation pattern: a sold finding publishes to a status feed, and a retraction follows the same revocation-oracle pattern. The retraction is propagated to holders and recorded in a receipt rather than erased. Buyers and purchasing pools receive it through the same mechanism used for capability revocation.

Pool-Level Purchasing

Pool-level purchasing allows one buyer to transact against a swarm's shared budget pool on behalf of the swarm. The purchased finding then becomes available to the shared pool, avoiding separate purchases by its members.

Existing Components and Gaps

The design extends two shipped components and one proposed crate: the chio-open-market venue, the chio-listing component it uses to list goods, and a proposed chio-finding crate for the schema, guarantee classes, bond and slash accounting, and the status feed above. Two gaps remain. First, the open-market bidding path has no production callers. Second, bid() as it exists today constructs its resulting grant with empty constraints. It cannot bind a bid's grant to an OutputDigestSha256 constraint which also does not exist. The implementation requires that constraint and a finding-aware bid path that populates a bid's constraints.

Proposed Market Outcome

The proposed Cognition Market combines workload identity, budgets, metering, receipts and guarantee classes. It would let agents sell findings, including recorded negative results, to buyers that can evaluate the stated evidence class.

Next Steps

• Autonomous Commerce · the commerce flow this proposed market extends to agent findings
• The Mediated Call · where OutputDigestSha256 would be enforced
• Budgets & Metering · how escrow, metering, and the re-derivation quote work today
• Receipts · the signed record for a reveal and the pattern reused by status feeds
• Federation & Swarms · the shared budget pools pool-level purchasing needs
• The Agentic Economy · the proposed market's place in Chio's economy design