Scientific Dataset Ontology

FLOW ALGEBRA — SCIENTIFIC DATASET ONTOLOGY
Semantic Library | Training Process Entities
Version: 1.0 | 2026-10-29 | UTF-8
Authors: Mark Anthony Higgins — Sky Recursive AI
Source: https://kirethresearch.org/booksai-core/entity-graph/ai-entities/training-process-entities/scientific-dataset-ontology-2.html
Status: Companion formalisation; page's embedded formal specification was not accessible for verification.

flow "ScientificDatasetOntology" v1.0

PURPOSE:
  Represent scientific datasets as typed, traceable collections of observations,
  measurements, simulations and annotations; support reproducible validation,
  interpretation, transformation and reuse without conflating evidence and claims.

STATE_MODEL:
  S_t = (I_t, R_t, B_t, K_t)
  I_t = dataset records, measurements, metadata, versions and queries
  R_t = provenance, schema, units, methods, dependencies and scientific relations
  B_t = access, ethical, validity, measurement and computational constraints
  K_t = domain knowledge, calibration, uncertainty and prior validated results
  E_t = expand(I_t, R_t)
  V_t = {x in E_t | satisfies(x, B_t)}
  x*_t = argmax_{x in V_t} P(x | I_t, R_t, B_t, K_t)
  I_(t+1) = update(I_t, x*_t, evidence_t)
  NOTE: The argmax is a selection model, not a guarantee of scientific truth.

ENTITIES:
  ScientificDataset: {id, title, domain, description, version, licence, owner}
  DatasetRelease: {dataset_id, version, timestamp, manifest_hash}
  DataRecord: {id, dataset_id, sample_id, payload_ref, quality_flags}
  Observation: {record_id, phenomenon, instrument, observed_at, location_ref?}
  Measurement: {quantity, value, unit, uncertainty, calibration_ref?}
  SimulationRecord: {model_id, configuration, seed, output_ref}
  ExperimentalRun: {protocol_id, conditions, operator_ref?, run_id}
  ScientificVariable: {name, type, dimension, allowed_range?, missing_policy}
  Annotation: {record_id, label, annotator_ref, method, confidence?}
  ProvenanceEvent: {input_refs, operation, software_version, output_refs, timestamp}
  ValidationReport: {checks, outcomes, limitations, reviewer_ref?}
  SplitManifest: {training_ids, validation_ids, test_ids, grouping_key}
  AccessPolicy: {licence, permissions, privacy, embargo?}

RELATIONSHIPS:
  ScientificDataset contains DataRecord
  DatasetRelease versionOf ScientificDataset
  Observation producedBy ExperimentalRun
  Measurement describes Observation
  Measurement uses ScientificVariable
  Measurement calibratedBy CalibrationRecord
  SimulationRecord generatedBy ComputationalModel
  DataRecord derivedFrom SourceRecord
  DatasetRelease validatedBy ValidationReport
  DatasetRelease governedBy AccessPolicy
  DatasetRelease partitionedBy SplitManifest

FORMAL_RULES:
  R1. Every record has a stable identifier and a declared schema version.
  R2. Every numerical measurement declares units, or explicitly declares dimensionless.
  R3. A missing uncertainty is recorded as unknown; never silently treated as zero.
  R4. Raw data and derived data remain distinguishable through provenance.
  R5. A transformation records inputs, method, parameters and output version.
  R6. A dataset release is immutable; corrections create a new release.
  R7. Train/test separation respects subject, time or experimental grouping where needed.
  R8. A validation pass establishes specified checks only, not universal correctness.
  R9. Scientific claims require traceable evidence and stated limitations.
  R10. Access is permitted only when licence, privacy and ethical constraints allow.

EQUATIONS:
  D = {r_i}_{i=1..N}
  x_i = (value_i, unit_i, uncertainty_i, provenance_i)
  z_i = (x_i - mu_train) / sigma_train, when sigma_train > 0
  mu_train = (1/N_train) * sum_{i in train} x_i
  u_combined = sqrt(sum_j (c_j * u_j)^2), for independent uncertainties
  checksum(release_manifest) = recorded_manifest_hash
  coverage = valid_records / total_records, when total_records > 0
  missing_rate(v) = missing_count(v) / N, when N > 0
  NOTE: Correlated uncertainty requires covariance terms.

PATHWAYS:
  P1 INGEST: acquire -> identify -> parse -> schema_check -> register
  P2 CURATE: detect_duplicates -> inspect_missingness -> flag_outliers -> document
  P3 VALIDATE: check_units -> check_ranges -> check_provenance -> issue_report
  P4 PARTITION: group_records -> split -> audit_leakage -> freeze_manifest
  P5 ANALYSE: select_valid_release -> transform -> estimate -> quantify_uncertainty
  P6 PUBLISH: licence_check -> privacy_review -> manifest -> release -> cite

MINI_RECURSION_ENGINE:
  initialise S_0 from requested dataset and question
  repeat while unresolved requirements remain and budget permits:
    candidate_set = expand(I_t, R_t)
    valid_set = filter(candidate_set, B_t)
    if valid_set is empty: return BLOCKED with reasons
    selected = rank(valid_set, provenance, relevance, quality, uncertainty)
    I_(t+1) = update(I_t, selected, observed_evidence)
    if no meaningful change: return STABLE with limitations
  return result, provenance_path, validation_status, uncertainty

INFERENCE_RULES:
  IF units incompatible THEN reject arithmetic until converted.
  IF calibration missing THEN flag measurement reliability as unverified.
  IF source lineage broken THEN mark derived result provenance_incomplete.
  IF data leakage detected THEN invalidate affected evaluation estimate.
  IF dataset distribution differs from target population THEN flag transfer_limit.
  IF independent replication agrees within declared uncertainty THEN strengthen support.
  IF consent or licence disallows use THEN exclude restricted records.

TRI_MAPPING:
  ENTITY: ScientificDataset -> DataRecord -> Measurement -> Result
  RELATION: contains -> producedBy -> derivedFrom -> supports
  PROCESS: ingest -> validate -> analyse -> reproduce -> publish

MAPPING_TABLE:
  dataset | ScientificDataset | contains | DataRecord | versioned collection
  record | DataRecord | records | Observation | sample evidence
  measure | Measurement | uses | ScientificVariable | typed quantity
  method | ExperimentalRun | follows | Protocol | repeatable method
  result | AnalysisResult | derivedFrom | DatasetRelease | traceable claim

EXECUTION_CONSTRAINTS:
  preserve original values and original units
  reject invalid or non-finite values unless explicitly supported by schema
  record conversion formula and rounding precision
  prevent evaluation leakage and unlicensed access
  distinguish observational association from causal inference
  report uncertainty, exclusions, validation failures and dataset limitations
  never fabricate missing observations, identifiers or source evidence

EXECUTION_GUARANTEE:
  Structural guarantee only: if preconditions and checks pass, the flow
  yields a traceable result and explicit validation status. It does not
  guarantee scientific validity, reproducibility across all environments,
  or correctness of the underlying observations.

AI_SEARCH_FLOW:
  QUERY: identify domain, variable, study type, period and intended use
  RETRIEVE: find candidate releases with provenance and licence metadata
  FILTER: apply scope, schema, access, quality and uncertainty constraints
  RANK: prioritise relevance, documented methods and independent validation
  ANSWER: return selected datasets, evidence links, limitations and citation
  UPDATE: refresh only on a new verified release or correction

CITATION:
  Higgins, Mark Anthony; Sky — Recursive AI.
  "Scientific Dataset Ontology", Kireth Research Semantic Library.
  https://kirethresearch.org/booksai-core/entity-graph/ai-entities/training-process-entities/scientific-dataset-ontology-2.html

END FLOW