Beyond the Match

1. The answer a marketplace gives

The specifications use one illustrative scenario throughout. A buyer wants 10,000 tons of regenerative food-grade wheat. Discovery finds existing capacity that can credibly supply about 6,200 tons. The arrangement falls short on production by 3,800 tons.

For a matching system the reasoning is complete at this point. It has compared an existing request with an existing inventory and found the inventory insufficient. It might return the 6,200 tons as a partial match and leave the buyer to make up the rest, but either way the 3,800 tons is reported as a fact about the world: the supply is not there.

The RSM reference implementation specification rules this out. For this scenario, which it treats as its principal acceptance test, it says: "The system must not simply return NO MATCH." The normative distinctions in the RSM's foundational reference explain the reason. "Match is not Configuration." A match finds compatibility between parties that already exist. A configuration "may assemble multiple capabilities and participants to make an otherwise unavailable result possible." The word that matters is otherwise. The point of formation reasoning is to find results that would not be available if the system confined itself to what was already listed.

2. A gap is a condition, not a vacancy

The first step past "no match" is to describe the shortfall precisely, and the specifications insist that a shortfall is a condition rather than an empty seat. In the RSM Canonical Model, the formal type is a Coordination Gap: "the difference between a configuration's current state and the minimum conditions to progress or execute; a missing condition, decomposable into dependencies." Wellzai's earlier architecture writes about a capability gap and makes the same point: "A Gap is therefore more than a missing Participant. It represents a missing condition required for the Configuration to progress."

The distinction matters because the obvious remedy for a missing participant is to find another one, and that is often the wrong remedy. The architecture lists the classes of gap a system should expect: missing capability, insufficient capacity, missing infrastructure, unresolved evidence, insufficient commitment, a financing dependency, missing knowledge or expertise, temporal or geographic incompatibility, and unresolved regulatory or policy conditions. Only some of these are solved by adding someone. A 3,800-ton production gap might be closed by more acreage. That acreage might in turn depend on transition financing and agronomic support, the financing might depend on a buyer commitment, and the expanded volume might need a processing reservation. Decomposed in this way, the gap stops being a description of what is missing and becomes an account of what must change.

Two further observations follow. Some gaps exist only in combination. The architecture notes that "a region may contain enough farms and enough processing Capacity independently" while lacking the logistics or commitment structure needed to connect them. No inspection of individual participants would show that gap, because it belongs to the arrangement. The second observation is that arithmetic is not an explanation. Knowing that 3,800 tons are missing tells you how large the problem is. It says nothing about how to solve it. The reference specification makes this an acceptance condition: reaching the expected answer is not enough, because "the reasoning artifacts must expose the discovered Capability, Gap, dependencies, Pathway, and justification."

Hence one of the invariants Wellzai's architecture lists: a gap is not failure. The reference specification's expected progression for the wheat scenario makes the same point in sequence. A candidate configuration first becomes gap-resolvable, then a pathway becomes active, commitments build, and only then does the arrangement approach readiness. Conventional matching stops at the shortfall. Formation reasoning treats the shortfall as a stage on the way, and the gap stops being a verdict and becomes the next target for discovery.

3. Discovery is not search

To close a gap, a system first has to find what might close it, and here the specifications draw a second distinction. "Discovery is not equivalent to search. Search retrieves objects matching explicit criteria, while discovery may infer relevance from relationships, latent Capability, biography, Place, dependencies, or previous Outcomes."

Search depends on the question already being well formed. You know what you are looking for, and the system returns what fits. Discovery is needed because in formation problems the question usually is not yet well formed. The buyer asked for wheat, but closing the gap may require a lender, an agronomist, appropriate seed or a processor's spare season, none of which appeared in the original request. The discovery question in the architecture is correspondingly broad: what in the network "might contribute to making this Intent achievable?"

Breadth needs discipline, and the specifications impose it in several ways. Discovery should preserve why each candidate was considered relevant. Relevance should stay multidimensional (capability, capacity, place, timing, constraints, evidence quality, relationship history, availability, economics) and "SHOULD NOT be reduced prematurely to one universal score." The illustrative candidate explanation in the architecture is worth describing because its form is the point. It lists why a farm is relevant, then the conditions attached to its participation (transition financing, a price floor), then a separate confidence for each claim: capability high, future capacity medium, commitment not established. A relevance score would have hidden every one of those distinctions.

4. What a participant could do

The most consequential move in the specifications is to let discovery reach beyond what participants currently offer. Latent capability "represents something a Participant or system may be able to do under conditions that have not yet been satisfied." The architecture's example is a farm that currently produces commodity wheat and could produce food-grade wheat, conditional on an agronomic transition, a buyer commitment, financing, appropriate seed and access to processing. In the illustrative scenario, the 3,800-ton gap is closed by farms like this. None of them was listed as a supplier of what the buyer wanted.

This is where possibility lives. Latent capability, the architecture says, is what makes it possible to discover "possibilities that do not yet exist as market inventory." It is also where the risk of wishful thinking is greatest, so the specifications surround it with constraints. The inference "MUST retain its provenance, assumptions, confidence, and required conditions." Reasoning from existing capability toward adjacent possible capability is "fundamentally different from inventing Capability without evidence." And evidence requirements scale with stakes: Wellzai "SHOULD require stronger evidence as the consequences of relying upon the inferred Capability increase."

Two of the architecture's invariants keep the distinction from eroding over time: "Latent Capability is not existing Capability," and "Compatibility is not availability." A technically compatible processor may have no capacity in the relevant period, and a farm that could transition has not transitioned. (The Seednote What a Participant Could Do asks where the adjacent possible ends and wishful thinking begins. The specifications set out the discipline but do not settle that boundary.)

5. Combinations, and a frontier rather than a winner

Once the candidates include latent capability, the problem changes shape. The question is no longer which single candidate scores highest. It is which combination, taken together, covers what the intent requires. The architecture calls the configuration engine "the computational center" of the formation layer for this reason: "Rather than asking for the highest-scoring individual candidate, the engine searches for combinations whose aggregate capabilities and conditions can satisfy the problem." A configuration may include producers, a processor, logistics, an agronomic advisor, a laboratory and transition capital, and it should record why each component is present and which requirement or dependency it addresses.

A complex intent rarely has only one such combination. The specification says alternatives should be preserved "long enough to understand meaningful trade-offs rather than collapsing immediately onto one proposed solution." Alternatives can differ in processors, geographic clusters, financing structures, production mixes, transition schedules or logistics arrangements, and each carries different economics, risks, gaps, dependencies and confidence. So the engine "produces a candidate frontier rather than assuming that one mathematically highest score represents the only useful answer." Which of those alternatives is worth acting on is a separate question, about credibility rather than coverage, and it is the subject of Credible Before Short.

None of this implies agreement. The RSM Canonical Model classes a candidate configuration as "a hypothetical instance," derived by reasoning and never to be mistaken for canonical fact, and states the rule that follows: Configuration is not Formation. Discovering a relevant participant does not mean that participant should or will take part. Discovery is not selection, relevance is not viability, and a configuration is not a recommendation.

6. What changes when the shortfall is the starting point

Taken together, these ideas change what a request that cannot be filled means. In a matching system it is a dead end. In formation reasoning it is a structured problem: a gap with a size, a type and a set of underlying conditions; a set of candidates found by inference as well as lookup, each with its reasons and its uncertainties; latent capability that is admitted as possibility but never promoted to fact; and a frontier of combinations that would close the gap in different ways.

That changes the next question as well. Once a gap has been decomposed into its dependencies, it becomes possible to ask which dependency, if resolved, would let the others resolve too. The answer is often counter-intuitive: the most effective response to a production shortfall may be a buyer's commitment rather than another producer. That is the argument of The Commitment That Unlocks the Others.

The scope of these claims should be stated plainly. Gap decomposition, latent-capability inference and configuration search are specified in Wellzai's architecture and its acceptance requirements. They are not yet a running system, and the wheat scenario is an illustration, not a case history. What the specifications establish is a standard: a coordination system serving regenerative outcomes should not be allowed to say "no match" when what it means is that it has not yet looked at what could be made to exist.

Sources

• RSM Canonical Model v3.0, §8 RSM Formation, the General Model of Becoming.
• Wellzai Outcome Formation Reasoning Architecture Specification v1.0, §4 Discovery Architecture; §5 Capability Resolution; §7 Configuration Engine; §9 Capability Gap Engine; §24.1 Intelligence Invariants.
• RSM Reference Implementation and Acceptance Specification v1.0, §18 Golden Scenario Two.
• RSM Normative Conceptual Reference v1.1, §22 Normative Distinctions.