← Back to catalogue
Published

Research, hypothesis and experiment

EM-PRD-04 · em-prd-04

Research question, hypothesis, experiment design, run and interpretation of the result. AI training and evaluation use specialized child models.

Enterprise profiles Enterprise subject EnterprisePRD

Bundle → Layer → Finding → Questions Filled

2 bundles · 3 layers · 4 findings · 11 questions

Question and hypothesis What the research asks and what it expects before any result is seen.

Study framing

The research question, its hypotheses and when each was stated.

Research question

The question a study answers, with the decision it is meant to inform.

  1. Which question does the study answer, and which decision depends on it?
  2. Which earlier studies or findings does this study build on?

Hypothesis timing

Each hypothesis with the date it was stated relative to the result.

  1. Was the hypothesis recorded before the result was evaluated?
  2. Is a hypothesis added after the result marked as post-hoc?
  3. Which outcome would count as evidence against the hypothesis?
Experiment and result How a design is run and what can be concluded from each run.

Design and runs

The design as a reusable plan and each run as a separate record.

Run reproducibility

A run with pinned input versions, environment and design version.

  1. Which design version and input versions did this run use?
  2. Can the run be repeated with the same inputs and give a comparable result?
  3. Do two runs with contradictory results remain separately recorded?

Interpretation

The finding drawn from runs, with its limits and its status.

Finding and limits

A conclusion with its applicability limits and evidence strength.

  1. Does a null result show no effect or only too little data?
  2. Under which conditions does the conclusion stop applying?
  3. Which later decision cites this finding, including a negative one?

Classifiers Filled

Family
Enterprise profiles
Category
Enterprise subject
Entry kind
subject
Domain
EnterprisePRD
Industry
Cross-industry
Tags
EM-PRD-04W2subjectResearchStudyHypothesisExperimentDesignExperimentRunResearchFinding

What it is Filled

Research question, hypothesis, experiment design, run and interpretation of the result. AI training and evaluation use specialized child models.

Why it exists Filled

Research question, hypothesis, experiment design, run and interpretation of the result. AI training and evaluation use specialized child models.

Distinguishing features Filled

  • A hypothesis is dated relative to the result, so a post-hoc hypothesis stays visibly post-hoc.
  • The experiment design is a plan and each run is a separate record with its own pinned inputs.
  • A negative or failed result is kept as a finding rather than discarded.
  • Distinct from a general knowledge article, which states accepted knowledge rather than a tested claim with limits.

What robots and AI may and may not do Filled

Must not

  • Delete a failed experiment or replace its hypothesis retroactively.
  • Present a post-hoc hypothesis as if it had been stated in advance.
  • Treat absence of data as evidence that there is no effect.
  • Generalise a conclusion beyond the limits recorded with it.
  • Merge runs with different input versions into one result.

Only with a human decision

  • Deciding to stop, extend or redesign a study.
  • Approving a product or policy decision based on a finding.
  • Approving experiments that involve people or their personal data.

May

  • Link a finding to the runs, inputs and design versions it rests on.
  • Flag a hypothesis whose recorded date falls after the evaluation of its result.
  • Report where results of two runs of the same design contradict each other.
  • Summarise the stated limits of applicability of a finding.

Moral aspects Filled

  • Selective reporting of results misleads the decisions that rely on them.
  • Experiments on users can affect them without their knowledge or consent.
  • Hidden negative results waste effort when others repeat the same work.

Who is affected

  • Participants and users exposed to experiments
  • Product and research teams
  • Decision makers who rely on findings

Owners Filled

Steward

The head of product who answers for the product portfolio and its evidence.

Master systems

  • Product catalogue
  • Product discovery repository
  • PLM system

Links to other meta-models Filled

neighbor

  • EM-KNW-02

references

  • WM-ACT-022 - conceptual-candidate
  • WM-KNW-009 - conceptual-candidate
  • WM-ACT-036 - conceptual-candidate

What else AI and robots need to interact with it Filled

Identity and identifiers required Filled

  • Identified by a study identifier, with each run identified by its own run identifier and the design version.

Direct properties not applicable Not applicable

Not applicable

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

Recognition optional Filled

  • A record states a question, a hypothesis or a run with pinned inputs and a result.
  • Often confused with a product backlog item or with a general knowledge article.

Capabilities and actions required Filled

  • Runs can be compared across design versions when inputs were pinned.
  • Negative findings can be traced to the decisions they informed.

Hazards and failure modes required Filled

  • Hindsight rewriting of hypotheses after the result is known.
  • Irreproducible results when input or environment versions were not recorded.
  • Overreach when a finding from one segment is applied to all users.

Standards and interfaces required Filled

  • W3C PROV-O for the provenance of runs, inputs and findings.
  • ISO 5725 series on trueness and precision of measurement methods.
  • DataCite Metadata Schema for citing datasets and results.

Context of use required Filled

  • Used in product discovery, applied research and engineering experiments.
  • AI training and evaluation runs are handled by specialised child models.

Sources Filled

  1. TM Forum SID: product/service/resource and commercial roles
  2. IDTA AAS: physical assets and composite descriptions
  3. Product management and engineering configuration; compare selected PLM and software practices
  4. OMG Systems Modeling Language (SysML) - Checked live (HTTP 200) on 2026-10-07.
  5. W3C PROV-O: Provenance Ontology - Checked live (HTTP 200) on 2026-10-07.
  6. W3C RDF Data Cube Vocabulary - Checked live (HTTP 200) on 2026-10-07.
  7. IDTA Asset Administration Shell Specifications - Checked live (HTTP 200) on 2026-10-07.

Open questions

  • Which conditions make a result reproducible?
  • How to distinguish absence of effect from absence of data?
  • How to link a negative result to a subsequent decision?
  • 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-prd-04/brief.json, ver-cy/world-models/card-supplements/em-prd-04.json

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