Experiment Run
Describe one governed experiment execution with a pinned manifest, distinct attempt identity, recorded departures and traceable output references.
Bundle → Layer → Finding → Questions Filled
3 bundles · 4 layers · 5 findings · 11 questions
Run definition What the run was supposed to do before it started.
Frozen manifest
The design, parameters and materials fixed for this run.
Manifest identity
The manifest version the run executes and the experiment it belongs to.
- Which experiment and which manifest version does this run execute?
- Was the manifest frozen before the run started, and by whom?
Planned conditions
Parameters, materials and procedures the run is meant to use.
- Which parameters, materials and procedure versions does the manifest fix for this run?
- Which approval or ethics clearance covers this run, if one is required?
Execution and lineage What actually happened during the attempt.
Attempt
The concrete execution with its own identifier and relation to earlier attempts.
Attempt lineage
How this attempt relates to earlier attempts of the same manifest.
- Is this a first attempt, a repeat or a replacement of an earlier attempt?
- Who executed the run, on which equipment or compute, and when did it start and end?
Deviation log
Append-only record of departures from the manifest.
Recorded deviations
Each departure from the manifest with time, cause and impact.
- Which deviations from the manifest were recorded during this run?
- Was any deviation edited or removed after it was recorded?
- Does any deviation invalidate the run for its intended analysis?
Outcome How the run ended and what it produced.
Result status
The final status of the run and its outputs.
Run outcome
Completion status and links to produced data.
- Did the run complete, fail or get aborted, and for what reason?
- Which datasets, samples or artefacts did this run produce?
Classifiers Filled
- Family
- World Models
- Category
- Activities and processes
- Entry kind
- event
- Navigation path
- NAV.ACT.EXP
- Domain
- ACT.EXP
- Industry
- Cross-industry
- Tags
- experimentrunact.exptrial
- Also called
- Experiment, Trial
What it is Filled
An experiment run is one governed execution of an experiment: a single attempt carried out against a manifest of design, materials, parameters and procedures that was frozen before it started. Its record holds the attempt lineage and an append-only list of deviations; the experiment design itself and the resulting datasets are separate things.
In scope
- Run identity, one study binding, immutable baseline manifest, assigned conditions and local readiness assertions
- Run-specific time, executor/context references, attempt lineage and append-only departure or correction assertions
- Output bindings, qualified closure and handoff, disclosure and record continuity
Out of scope
- Study design and protocol authoring, randomization algorithms, hypothesis revision or assessment, cohort and participant masters
- Observation measurement, dataset editing, sample handling, analysis execution, statistical inference and evidence adjudication
- Clinical care, safety control, consent collection, ethics approval, operating equipment or implementing hazardous procedures
- External system writes, public result registration and legal or standards-conformance certification
Why it exists Filled
Describe one governed experiment execution with a pinned manifest, distinct attempt identity, recorded departures and traceable output references.
Distinguishing features Filled
- One run is one attempt; the experiment design and its manifest are separate records the run points to.
- The manifest is frozen before execution, so any change during the run is a recorded deviation, not an edit.
- Repeats keep their own identity and link to earlier attempts instead of overwriting them.
- Distinct from the dataset it produces and from the analysis that interprets it.
What robots and AI may and may not do Filled
Must not
- Edit a frozen manifest after the run has started.
- Remove, rewrite or back-date a recorded deviation.
- Report a failed or aborted attempt as completed, or hide it from the lineage.
- Start a run that requires ethics or safety approval before that approval is recorded.
- Select attempts for reporting only because their results were favourable.
Only with a human decision
- Approving a manifest and freezing it for execution.
- Deciding whether a deviation invalidates a run.
- Abandoning a run that involves participants or animals.
May
- Register a run against a frozen manifest and record its start, end and executor.
- Append deviations with time, cause and observed impact.
- Link the datasets and samples a run produced to its record.
- Report runs whose outputs are not traceable to a manifest version.
Moral aspects Filled
- Selective reporting of successful attempts distorts science and can harm those who rely on it.
- Experiments involving people or animals depend on valid approval and consent for each run.
- Honest deviation records protect the credibility of everyone who uses the results.
Who is affected
- Research participants and animal subjects
- Researchers and laboratory staff
- Users of the results, including reviewers and regulators
Owners Filled
Steward
The principal investigator or study lead who answers for the experiment and its runs.
Roles
- Run steward
- Maintain identity, study binding and reconciliation; do not claim ownership of external records.
- Executor or recorder
- Supply attributed actual progress, outputs and departures without rewriting prior evidence.
- Authorized scientific reviewer
- Supply scoped external decisions about continuation and intended-use suitability.
- Safety and ethics authority
- Provide applicable decisions or exceptions through the responsible external governance process.
- Record custodian
- Enforce recipient views, holds, lawful retention and disposition with access evidence.
Master systems
- Electronic laboratory notebook
- Laboratory information management system
- Experiment tracking system
Links to other meta-models Filled
references
- WM-ACT-036 - Each governed run belongs to one study and pins one immutable study-scoped design release. Study protocol, design, cohorts and scientific finding lifecycle remain study-owned; inbound CONTAINS does not transfer their ownership.
- WM-KNW-009 - Pin hypothesis revisions and preregistered or post-hoc timing assertions only. Hypothesis identity, revision and assessment remain external.
- WM-MAT-008 - Reference actual observations, values, methods and uncertainty without editing them or declaring measurements locally authoritative.
- WM-DAT-001 - Pin immutable input and output dataset versions, partitions and available digests. Dataset content, identity, correction and lifecycle remain external; a run may produce no dataset.
- WM-ACT-053 - Reference separate analysis or processing executions and their results. Their retries and output production remain external; a processing retry is not automatically a new experimental run.
- WM-KNW-008 - Reference evidence and citation bindings used by later findings. Evidence stance, assessment and citation lifecycle remain knowledge-owned.
neighbor
- WM-ACT-036 Research Study - Each governed run belongs to one study and pins one immutable study-scoped design release. Study protocol, design, cohorts and scientific finding lifecycle remain study-owned; inbound CONTAINS does not transfer their ownership.
- WM-KNW-009 Hypothesis - Pin hypothesis revisions and preregistered or post-hoc timing assertions only. Hypothesis identity, revision and assessment remain external.
- WM-MAT-008 Observation / Measured Value - Reference actual observations, values, methods and uncertainty without editing them or declaring measurements locally authoritative.
- WM-DAT-001 Dataset - Pin immutable input and output dataset versions, partitions and available digests. Dataset content, identity, correction and lifecycle remain external; a run may produce no dataset.
- WM-ACT-053 Analysis / Processing Execution - Reference separate analysis or processing executions and their results. Their retries and output production remain external; a processing retry is not automatically a new experimental run.
- WM-KNW-008 Evidence / Citation - Reference evidence and citation bindings used by later findings. Evidence stance, assessment and citation lifecycle remain knowledge-owned.
- Trial label and legacy material - The queue retains Experiment / Trial, but the current registry names Experiment Run. A whole clinical trial stays study-owned. No legacy_alias or existing_spec_ref is registered; no legacy world specification is inherited.
- Plan amendments and observed departures - The original run baseline never changes. Authorized mid-run amendments are linked as prospective study-owned decisions with effective scope; unintended deviations are separately typed. Neither silently replaces the baseline.
parent
- WM-ACT-036
What else AI and robots need to interact with it Filled
Identity and identifiers required Filled
- A run is identified by a run identifier issued by the laboratory or tracking system, scoped to its experiment.
- The run record cites the manifest version and, where present, a registered study or protocol identifier.
Direct properties not applicable Not applicable
Not applicable
A run is an activity; the physical conditions and measurements it captures belong to the observation and dataset records it produces.
Recognition optional Filled
- A run record names one manifest version, an executor, a start time and a status.
- Often confused with the experiment design, with a protocol or with the dataset of results.
Capabilities and actions required Filled
- Runs can be repeated against the same manifest while keeping each attempt's identity.
- Deviations can be appended and reviewed without altering the manifest.
- Outputs can be traced back to the run, manifest version and executor.
Hazards and failure modes required Filled
- Silent changes to a manifest make results impossible to reproduce or audit.
- Hidden failed attempts bias the evidence base.
- Runs started without required approval expose participants to unapproved risk.
Standards and interfaces required Filled
- W3C PROV-O for the provenance of runs and their outputs.
- Electronic laboratory notebook and tracking system records.
- Clinical trial registries where the run belongs to a registered trial.
Context of use required Filled
- Used in laboratory science, clinical research, engineering tests and machine learning experiments.
- Good laboratory practice and good clinical practice rules apply in regulated settings.
Sources Filled
- PROV-O: The PROV Ontology - World Wide Web Consortium
- Completely randomized designs - National Institute of Standards and Technology
- Good Laboratory Practice: OECD Principles and Guidance for Compliance Monitoring - Organisation for Economic Co-operation and Development
- ICH E6(R3) Guideline for Good Clinical Practice - International Council for Harmonisation
- Process Run Crate - Workflow Run RO-Crate working group
- Basic HHS Policy for Protection of Human Subjects: 2018 Requirements - United States Department of Health and Human Services
- Constraints of the PROV Data Model - World Wide Web Consortium
- How do you select an experimental design? - National Institute of Standards and Technology
- OECD Principles of Good Laboratory Practice, OECD
- ICH E6 Guideline for Good Clinical Practice, International Council for Harmonisation
Open questions
- Run the prepared HTTP checker outside the sandbox and review current editions, errata, licensing and applicable consolidated rules for every adopting profile.
- Specify and test instance schemas and adapters for cancelled reservations, unauthorized starts, late amendments, clock contradictions, partial outputs, failed predecessors and lawful redaction.
- Pin neighboring model releases and obtain specialist review for actual clinical, animal, laboratory, field and computational use cases.
- Restore independent external review before promotion beyond a noncanonical reviewable draft.
- No independent external provider review; separate local self-audit required before synthesis.
- Direct HTTP status and response hashes are unmeasured; selected browser readings do not verify every source claim or current applicability.
- No nested instance schemas, state transition implementation, tested exchange mappings or pinned releases of neighboring world models.
- Clinical, animal, field, physical laboratory, online and computational specialist profiles are not exhaustively evidenced; consent exceptions and safety decisions require qualified local authority.
- Reproducibility, statistical power, scientific truth and compliance cannot be established from a run record alone.
Machine files
Provenance
world-models research · reviewable-draft
Built from: models/wm-act-022-experiment-run/spec.yaml, ver-cy/world-models/card-supplements/wm-act-022-experiment-run.json