← Back to catalogue
Published

AI and research landscape

EM-LND-14 · em-lnd-14

Support research oversight, artifact reuse, withdrawal and release decisions by connecting usage, rights, evaluation applicability and resource consumption.

Enterprise profiles Enterprise landscape EnterpriseLND

Bundle → Layer → Finding → Questions Filled

2 bundles · 2 layers · 4 findings · 8 questions

Artifacts and usage Models, datasets and systems, and where each artifact is used.

Usage network

Which systems and runs use which artifacts.

Artifact use

Links from systems and runs to the artifacts they use.

  1. Which systems and runs use this model or dataset?
  2. Which systems still use a withdrawn artifact?

Input rights

Rights attached to datasets and other inputs.

  1. Under which rights or licences was each input used?
  2. Do the rights still allow the current use?
Research and evaluation Runs, evaluations and the resources they consumed.

Runs and evaluations

Training and evaluation runs as records separate from artifacts.

Evaluation applicability

Whether an evaluation still applies to what is deployed.

  1. Which evaluations lost applicability after an input or version changed?
  2. Does a benchmark result cover the context of this deployment?

Compute and research

Compute linked to research work without a mandatory product.

  1. Which research work consumed this compute?
  2. Is the research linked to a product, or does it stand alone?

Classifiers Filled

Family
Enterprise profiles
Category
Enterprise landscape
Entry kind
landscape
Domain
EnterpriseLND
Industry
Cross-industry
Tags
EM-LND-14W3landscapeAIResearchLandscapeAIUsageNetwork

What it is Filled

Links between research, models, datasets, runs, evaluations, systems and consumed resources.

Why it exists Filled

Support research oversight, artifact reuse, withdrawal and release decisions by connecting usage, rights, evaluation applicability and resource consumption.

Distinguishing features Filled

  • A run is a record of work done; an artifact is what the work produced.
  • A benchmark result applies to its own context and never acts as a general clearance.
  • Research may stand on its own without being tied to a product.
  • Distinct from the individual AI system, model and evaluation contours that it connects.

What robots and AI may and may not do Filled

Must not

  • Create a commercial product record from a research checkpoint by default.
  • Treat a benchmark score as clearance for every deployment.
  • Drop the rights attached to an input when it is reused.
  • Decide on its own to withdraw a deployed system after an input is withdrawn.

Only with a human decision

  • Deciding what happens to each deployment when an input is withdrawn.
  • Approving reuse of a dataset under a different purpose or licence.

May

  • List training runs, evaluations and systems affected by a withdrawn dataset.
  • Flag evaluations whose inputs or versions no longer match the deployment.
  • Report compute use per research line.

Moral aspects Filled

  • Deployments built on withdrawn or unlawfully used data can harm the people in that data.
  • Over-broad use of benchmark results can hide risks to affected users.

Who is affected

  • People whose data was used
  • Users of deployed AI systems
  • Researchers and model owners

Owners Filled

Steward

The enterprise AI and research governance steward is accountable for landscape integrity and coordination with authoritative research and operational record owners.

Master systems

  • Research management systems govern research activities and their approved objectives.
  • Dataset catalogues govern dataset identities, versions and provenance.
  • Rights management repositories govern recorded permissions, licences and use restrictions.
  • Model registries govern model versions and lifecycle status.
  • Experiment tracking systems govern run records and their input and output references.
  • Evaluation repositories govern evaluation results and their applicability conditions.
  • Deployment inventories govern recorded operational artifact use.
  • Resource accounting systems govern compute consumption and allocation records.

Links to other meta-models Filled

references

  • WM-AI-001 - conceptual-candidate
  • WM-AI-003 - conceptual-candidate
  • WM-AI-006 - conceptual-candidate
  • WM-AI-008 - conceptual-candidate

What else AI and robots need to interact with it Filled

Identity and identifiers required Filled

  • The usage network has an identifier and version, while artifacts, runs and systems retain their own identifiers.

Direct properties not applicable Not applicable

Not applicable

Enterprise record contour: physical properties belong to referenced world models.

Recognition required Filled

  • The landscape connects research, datasets, models, runs, evaluations, systems and compute.
  • It can be confused with a model registry or an experiment tracker.

Capabilities and actions required Filled

  • Withdrawal of an input can be traced to affected runs, evaluations and systems.
  • Recorded rights information can be followed from inputs to outputs.

Hazards and failure modes required Filled

  • Withdrawn or unlicensed inputs may remain in use.
  • Stale evaluations may be treated as current safety evidence.

Standards and interfaces optional Filled

  • PROV-O from the World Wide Web Consortium supports provenance of runs and artifacts.
  • DCAT from the World Wide Web Consortium supports dataset descriptions.
  • ISO/IEC 42001 from the International Organization for Standardization and the International Electrotechnical Commission provides requirements for AI management systems.

Context of use required Filled

  • The landscape supports AI governance, dataset withdrawal and model release decisions.
  • Applicability of the European Union AI Act or other regulation depends on jurisdiction, statutory scope, relevant exclusions and applicable dates.
  • A high-risk classification alone does not establish that a particular regulation applies.

Sources Filled

  1. Vercy composition and whole-object: a separate landscape object and delegation
  2. ArchiMate/ISO 42010: views, questions and representation rules
  3. Compare participant subject models and the practical assembly of a permitted context
  4. Vercy Whole-Object Description Profile - Checked live (HTTP 200) on 2026-10-07.
  5. NIST Artificial Intelligence Risk Management Framework - Checked live (HTTP 200) on 2026-10-07.

Open questions

  • Which systems use a withdrawn artifact?
  • Which evaluations have lost applicability?
  • How is compute linked to research without a mandatory product?
  • Установить границу и решение reuse/extend/new по действующим спецификациям.
  • Подтвердить semantic crosswalk, права и source mastership.
  • Выбрать immutable refs; провести проверки fixtures до заявления о публикационной готовности.

Machine files

Provenance

enterprise research programme · published-partial

Built from: enterprise/models/em-lnd-14/brief.json, ver-cy/world-models/card-supplements/em-lnd-14.json

Still in Russian: suggested owner, blocking decisions, vercy candidates. Translate in research/enterprise/i18n/units.en.json.