Zum Inhalt springen

Domain Knowledge and Group Definition

Dieser Inhalt ist noch nicht in deiner Sprache verfügbar.

Conformance language (MUST/SHOULD/MAY) follows BCP 14 [RFC2119]/[RFC8174] as defined in 00-overview.md. The binding source is the FlowMCP Schemas Specification v4.2.0.


A topic group (selection composed of several namespaces) develops conventions, shared vocabularies, and provider-specific quirks that are not visible from any single namespace in isolation. Grading at the group-bound tier (see 06-determinism-and-tier.md) MUST be validated against the group’s Domain-Knowledge content — and that content lives in the selection’s About Resource, not in a separate document.

This is a binding identity in this spec: the Domain-Knowledge = the About Resource of the selection (graded by the about-selection Area, see 05-phases-selection.md and 11-about-convention.md). It is an internal document, written so that it carries the seven mandatory sections of Mandatory Sections below. There is no second file.

Without Domain-Knowledge content for a group, the selection-side grader cannot produce a group-bound grading entry. The schemas in the selection MAY still be graded at the autonomous tier; those grading entries then carry gradingTier = autonomous and maxAttainableGrade = B.

The dimension that consumes the Domain-Knowledge content is domainConformance (carried by the selection-aggregate Area). It checks the members against the document — distinct from about-selection, which checks the document quality. The two checks are independent; there is no circularity.


A “group” — i.e. a selection deserving its own Domain-Knowledge content — is defined by two thresholds over the number of namespaces in the selection:

ThresholdConditionEffect
Soft5 namespacesA selection SHOULD be treated as a group. Selection skills MAY be created. The selection Areas (see 05-phases-selection.md) run with reduced expectations. aggregateGrade = A is NOT regularly attainable at this threshold.
Hard7 namespacesFull group optimisation applies. personaUseCaseFit is fully scaled. aggregateGrade = A is regularly attainable at this threshold.

A selection with fewer than 5 namespaces is explicitly not a group in the narrow sense. A 3-namespace selection MAY still be grouped for convenience (UI grouping, skill registration), but it does NOT trigger the group-bound grading path; entries derived from it carry gradingTier = autonomous.

“The more namespaces a selection contains, the better.”

This statement is the rationale behind the hard threshold of seven. Diversity of sources is the precondition for a meaningful persona-/domain-conformance evaluation: a selection that aggregates only two or three providers cannot reliably tell whether a given convention is a domain-wide standard or a single-provider quirk. The hard threshold encodes this preference structurally.


Mandatory Sections of the Domain-Knowledge Content

Section titled “Mandatory Sections of the Domain-Knowledge Content”

The Domain-Knowledge content (carried by the selection’s About Resource) MUST contain the following seven sections. The document MAY add further sections; the seven below are the binding minimum. Content that lacks any of these sections is INVALID for the purpose of domainConformance grading and scores low on about-selection.

  1. Identity — group identifier, human name, one-paragraph description.
  2. Shared Lists — the list of Shared Lists the group adopts (e.g. the Shared Chain Name List), with a MUST-statement that schemas in the group use the Shared List in place of any provider-specific convention.
  3. Conventions — naming, casing, ordering, aggregation strategy. The conventions the group has agreed upon and which schemas MUST follow.
  4. Forbidden Conventions — explicitly listed provider conventions that schemas MUST NOT adopt. The crypto example below is the canonical illustration.
  5. Use Cases — typical scenarios the group serves; the source of truth for personaUseCaseFit reasoning at the selection level.
  6. Personas Reference and Lens — which of the four generalised base personas (see 12-personas-contract.md) apply to the group, including the group’s Lens definitions (e.g. crypto-trader as a Lens over decision-maker).
  7. Aging Rule — how long the content remains valid for grading purposes. Default: 90 days. Once exceeded, the content MUST be re-reviewed; grading entries that referenced it MAY be marked score = stale per the aging rule in 08-grading-model.md.

The aging rule for the Domain-Knowledge content itself is separate from the aging defaults for individual dimensions (API_DAYS, TOS_DAYS, RETENTION_DAYS — see 08-grading-model.md). The 90-day default for Domain-Knowledge content is a SHOULD; groups MAY override.


The crypto domain is the canonical illustration of the Domain-Knowledge contract, and the Forbidden Conventions section is where the contract bites.

Concrete conflict. A widely-used third-party data provider uses the chain name solana in its API responses. The Shared Chain Name List — adopted by the crypto group as its canonical Shared List — uses sol for the same chain. A schema in the crypto selection that emits solana (the provider’s convention) instead of sol (the Shared List) is in direct conflict with the group’s Domain-Knowledge content.

Grading consequence. The grader records a Forbidden-Conventions violation in the domainConformance dimension. The violation is a penalty, NOT a Categorical Veto — it does not raise aggregateGrade = REJECTED. It IS a strong score reduction on domainConformance via a high weight. The high weight reflects the repeated experience of having to explain the Shared-List rule across multiple schema reviews.

The choice of “high weight, no veto” deliberately keeps the door open: a schema that violates a Forbidden Convention is fixable by remapping the value at the schema’s output layer, and the grading should incentivise the fix rather than reject the schema outright.


Diversity Maxim — “More Namespaces, Better”

Section titled “Diversity Maxim — “More Namespaces, Better””

The maxim from Diversity Argument has a second concrete consequence beyond the hard threshold: more namespaces in a group widen the basis against which domainConformance can be evaluated. A 7-namespace crypto selection that draws from many distinct providers exposes provider quirks (e.g. one provider’s solana vs. another’s native chain identifier) to direct comparison; a 3-namespace selection cannot do this comparison at all.

The grading consequence is indirect — there is no dimension named “namespace diversity” — but the maxim is reflected in:

Groups that aspire to aggregateGrade = A SHOULD aim for ≥ 7 namespaces and a comprehensive Forbidden-Conventions section.


The Domain-Knowledge content is the selection’s About Resource. It is stored at:

selections/<selection>/resources/about/

as a markdown Resource (see 11-about-convention.md and 19-folder-layout.md). It is graded by about-selection. There is no separate Domain-Knowledge file path — the selection’s About Resource is the single internal document that carries the seven sections of Mandatory Sections.

A grading entry that uses Domain-Knowledge content records the resolved About Resource reference in its selectionContext (see 08-grading-model.md).


DimensionTierEffect
domainConformancegroup-boundScore reflects alignment of the members with the group’s Conventions and absence of Forbidden-Conventions violations. Without Domain-Knowledge content, domainConformance cannot be scored above stale / n/a. Carried by selection-aggregate.
personaUseCaseFitgroup-boundReads the Personas Reference and Use Cases sections of the Domain-Knowledge content as ground truth. Carried by selection-aggregate.

Binding rule. Without Domain-Knowledge content for a group, no group-bound grading entry can be produced, and aggregateGrade = A is consequently NOT attainable for the schemas in that selection. This rule is the structural enforcement of the diversity maxim: groups have to invest in their About Resource to unlock the top grade.