Research and Agriculture Base Layers in the Ontology Spine

ADR-0003: Research and Agriculture Base Layers in the Ontology Spine

Status

accepted

Context

Background and Problem Statement

The corpus already layers shared supertypes into base ontologies that domain packs extend: mif-base (the cognitive triad) → mif-generic (the MIF built-in entity kinds) → engineering-base (shared engineering supertypes: component, control, policy, artifact, provenance, …) → leaf engineering packs (software-engineering, data-engineering). The leaf packs declare subtype_of against the base supertypes, so a base-typed relationship endpoint accepts any specialization (ADR-0011 substitutability in the harness).

Two families of packs do not yet have their own base layer:

• Research packs. scientific carries generic research-methodology types (study, research-investigation, method, measurement, hypothesis, dataset, research-publication, research-funding, research-instrument) that are not biology-specific — they are what any research-observation pack needs. A future agriculture-research, market-research-as-research, or clinical-research pack would re-derive them.
• Agriculture packs. regenerative-agriculture carries generic farm types (farm, field, soil-profile, crop, animal/herd, equipment, input, harvest-record, weather-event) mixed with regen-specific ones (crop-rotation, grazing-plan, carbon-baseline, carbon-credit).

Current Limitations

Without these layers the spine is inconsistent and lossy:

• regenerative-agriculture-research extends engineering-base — an engineering lineage — though it models agriculture research. It inherits component/control rather than research or agriculture supertypes. That is the wrong parent.
• scientific and regenerative-agriculture both extend mif-base directly, so there is no shared research or agriculture supertype to type a cross-pack relationship endpoint against, and every new research/agriculture pack re-mints the generic core instead of specializing it (violates the reuse-before-mint rule ADR-0001/the build-spec set).

Decision Drivers

Primary Decision Drivers

• When a new research or agriculture pack is authored, the pack shall specialize shared supertypes via subtype_of rather than re-mint generic research or farm types.
• If a relationship endpoint is typed at a research or agriculture supertype, then any pack’s specialization of that supertype shall satisfy the endpoint (substitutability).
• A research pack shall extend a research lineage and an agriculture pack an agriculture lineage; no pack shall inherit a domain it does not belong to (no agriculture-research under engineering-base).

Secondary Decision Drivers

• The two base layers shall mirror the shape of engineering-base (extends [mif-base, shared-traits], supertypes under the cognitive-triad namespaces), so the spine stays uniform.
• Existing typed findings shall keep resolving across the re-parenting (migration, not a break).

Considered Options

Option 1: Keep the flat structure (no new base layers)

Leave scientific/regenerative-agriculture extending mif-base and regenerative-agriculture-research under engineering-base.

• Pro: Zero migration; nothing moves.
• Con: The wrong-lineage parent persists; every new research/agriculture pack re-mints the generic core; no shared supertype for cross-pack endpoints.

Risk Assessment

• Technical: Medium — re-minting drift accumulates; substitutability is unavailable.
• Schedule: Free now, costlier per future pack.
• Ecosystem: The spine stays inconsistent with the engineering lineage.

Option 2: Introduce research and agriculture base layers; re-parent (chosen)

Mint two base ontologies, each extends: [mif-base, shared-traits], modeled on engineering-base:

• research — generic research-methodology supertypes extracted from scientific’s non-bio core: research-study, research-investigation, research-method, measurement, hypothesis, dataset, research-output (publication), research-funding, research-instrument, research-finding.
• agriculture — generic farm supertypes extracted from regenerative-agriculture: farm, field, soil, crop, livestock, agricultural-input, agricultural-equipment, harvest-record, weather-event.

Re-parent: scientific → extends: [research]; regenerative-agriculture → extends: [agriculture]; regenerative-agriculture-research → extends: [research, agriculture] (it is agriculture research). Leaf packs declare subtype_of against the new supertypes; the generic types they previously minted become specializations.

• Pro: Correct lineage; reuse-before-mint for every future research/agriculture pack; shared supertypes enable cross-pack substitutable endpoints; the spine is uniform with the engineering lineage.
• Con: A migration: re-parenting changes extends closures and the vendoring index extends fields, and every affected finding’s type must still resolve. It also supersedes the just-completed ADR-0002 promotion of scientific (its extends changes), so the registry promotion must be re-run on the layered packs before deploy.

Risk Assessment

• Technical: Medium — supertype extraction + subtype_of wiring must validate; resolve-ontology must still type the corpus.
• Schedule: Medium — adds a base-authoring + re-parent + re-promote pass before the ADR-0002 deploy.
• Ecosystem: Low once landed; the spine becomes consistent and extensible.

Option 3: Rename scientific to research (no agriculture layer)

Treat scientific itself as the research base; add no agriculture layer.

• Pro: Less new content.
• Con: scientific carries bio-specific types (sample-organism, protocol-application) that do not belong in a generic base; conflates the base with one domain; leaves the agriculture lineage unfixed.

Risk Assessment

• Technical: Medium — a polluted base re-creates the reuse problem one level up.
• Schedule: Cheaper, but half the fix.
• Ecosystem: Asymmetric (research fixed, agriculture not).

Decision

Adopt Option 2. Mint research and agriculture base ontologies (each extends: [mif-base, shared-traits], supertypes under the triad namespaces, shaped like engineering-base). Re-parent scientific onto research, regenerative-agriculture onto agriculture, and regenerative-agriculture-research onto [research, agriculture]. Leaf packs declare subtype_of against the new supertypes; the generic types they minted become specializations.

Sequencing (load-bearing): because this changes scientific‘s (and the agriculture packs’) extends, it supersedes the ADR-0002 promotion of those packs. Land this layering — bases + re-parent + re-merge + re-run the registry promotion + regenerate the object+sha index — before the mif-spec.dev deploy, so the deploy publishes the final layered closure once, not twice.

Consequences

Positive

• Correct domain lineage; no pack inherits a domain it does not belong to.
• Reuse-before-mint for every future research/agriculture pack.
• Shared supertypes give substitutable cross-pack relationship endpoints.
• The spine is uniform with the engineering lineage.

Negative

• A real migration: extends closures, vendoring-index extends fields, and finding type-resolution all move; the ADR-0002 promotion must be re-run on the re-parented packs before deploy (extra pass, and the Phase-A scientific/ agriculture commits are superseded).
• Two new published base ontologies to maintain and version.

Neutral

• The new bases follow the existing engineering-base shape, so no new conventions are introduced.
• Versioning: the re-parented leaf packs take a version bump (their extends changed); the new bases ship at 0.1.0.

Decision Outcome

The decision meets its drivers: new packs specialize shared supertypes (driver one); base-typed endpoints accept specializations (driver two); research packs sit under research and agriculture packs under agriculture, ending the engineering-lineage miscategorization (driver three). The bases mirror engineering-base (secondary one) and the re-parenting is a migration that keeps findings resolving (secondary two). The residual cost — a re-promote pass that supersedes part of the ADR-0002 work — is mitigated by sequencing the layering before the single mif-spec.dev deploy.

Related Decisions

• ADR-0001: Underscore-Prefixed Base-Type Namespaces
• ADR-0002: Object-Keyed, Hash-Pinned Ontology Vendoring Index — its promotion of scientific/agriculture packs is superseded by this re-parenting and must be re-run before deploy.

Links

• ontologies/engineering-base.ontology.yaml: the base-layer shape this follows.
• ontologies/scientific.ontology.yaml, regenerative-agriculture.ontology.yaml, regenerative-agriculture-research.ontology.yaml: the packs re-parented.

More Information

Auditable in the extends: blocks of the affected packs and the new base ontologies’ entity_types/subtype_of wiring: a research/agriculture pack whose extends does not name its base layer, or a generic type re-minted in a leaf instead of specialized from the base, signals this decision has been violated.

Audit

• 2026-06-30: Pending. Accepted; bases authored (collision-free abstract supertypes, discovery disabled), scientific/regenerative-agriculture/regenerative-agriculture-research re-parented with subtype_of (all reachable via extends closure), corpus resolves 36/36. Flip to Compliant once deployed + the harness vendors against the live registry.