Metric Definition

Stable metric identity

Proposed identity separates an enduring metric from its immutable revisions and localized labels. Identical names do not establish identical semantics.

1. Which master identifier and namespace distinguish this metric from similarly named metrics? identity
2. How are immutable revision identifiers associated with the enduring metric identifier? lifecycle
3. Which preferred labels, aliases and concept references describe the metric without merging distinct definitions? classification

Business meaning and intended use

Record the measured concept and decision context. DQV provides a metric/measurement distinction for quality metrics only; extension to business metrics is a local design choice.

1. What business concept does this metric quantify and what similar concepts does it exclude? definition
2. For which use cases is this definition suitable and where is use explicitly unsupported? requirement
3. Where are any targets, desired direction or decision thresholds governed separately from the metric definition? decision

Accountable roles and permitted use

Assign role references without naming a company as owner. Authorship, approval authority and execution permission remain distinct. Access rules are local proposals interpreted by an external policy service.

1. Which accountable role maintains the meaning and which technical role maintains implementation bindings? ownership
2. What evidence authorizes approval or retirement of a metric revision? authority
3. Which audiences may see formulas, source bindings and examples, and what disclosure exceptions apply? access

Formula and operand roles

The proposed formula contract includes operator order and distinct numerator/denominator roles where applicable. A source expression is inert metadata; untrusted text is never executed during registration.

1. What expression and language/version define the metric, including numerator and denominator when relevant? definition
2. Which input measures or other metric revisions supply each operand, and how are dependency cycles rejected? composition
3. In what order do row-level derivation, filtering, aggregation and final arithmetic occur? process

Parameter contract

Separate a metric family parameter from a changed definition. Record defaults and allowed domains explicitly, with rejection for missing required values or unauthorized overrides.

1. Which parameters are permitted and what type, unit, range or code set constrains each one? constraint
2. What happens when a parameter is absent, invalid or attempts to override a fixed semantic choice? exception
3. Which parameter changes preserve this revision and which require a new metric or revision? identity

Units, precision and exceptional values

Record quantity meaning separately from presentation scale. Ratios, percentages and counts need explicit interpretation; monetary conventions are profile dependencies, not SI rules.

1. What quantity, unit, scale and output datatype define the result before display formatting? measurement
2. Which precision, rounding stage and approximation or uncertainty declarations apply? quality
3. How are empty input, missing operands, zero denominators, overflow and nonfinite values represented? exception

Population and counting unit

A business population is a definition-level eligibility rule. No actual subject records are stored here. Deduplication choices must not be mistaken for source-data repair.

1. What entity or event is counted and what eligibility rule defines the population? definition
2. Which key distinguishes contributions and how are repeats, reversals or duplicate events treated? identity
3. What evidence distinguishes an absent contribution from an unobserved or excluded one? quality

Grain and dimension bindings

Input grain and result grain are separate. Local dimension bindings do not own hierarchy members. Join multiplicity must be documented before a metric binding is accepted.

1. What input grain and output grouping key determine one result cell? composition
2. Which dimension levels and hierarchy revisions may be used for slicing or rollup? relationship
3. How does the binding prevent join fan-out, missing-key loss or accidental many-to-many multiplication? validation

Filters and evaluation stage

Proposed filter metadata distinguishes population eligibility, pre-aggregation filters, post-aggregation tests and user slices. Access enforcement may restrict visible data but cannot silently redefine a metric.

1. Which predicates are fixed and which may consumers supply as slices? constraint
2. At which stage is each predicate applied and how do null or unknown predicate values behave? process
3. How are access-based data restrictions or suppression disclosed so differently authorized results are not presented as equivalent? privacy

Time basis and windows

A metric declares which time field governs inclusion and the calendar/window semantics required of its evaluator. RFC 3339 encodes instants, not business calendar policy.

1. Which event, effective, snapshot or ingestion time determines membership in the measurement period? temporal
2. What calendar, time zone, interval boundaries and window anchor define a period? temporal
3. How are incomplete periods, ambiguous local times and events exactly on a boundary handled? exception

Late data and restatement semantics

Declare how consumers should distinguish original-as-reported and restated results. Runtime ingestion, recalculation and publication of values remain external responsibilities.

1. What cutoff, lateness allowance and snapshot reference define an as-reported result? temporal
2. Which late-data or correction cases permit restatement, and how is the original result distinguished? lifecycle
3. Which definition revision and data snapshot must accompany a reproduced historical result? provenance

Aggregation and comparability

Data Cube deliberately does not define aggregation operators. This proposal requires a per-axis rule and evidence for combining values, including ratios, distinct counts and snapshot measures.

1. Across which axes is the metric additive, conditionally additive or non-additive? measurement
2. When must a total be recomputed from components instead of summing or averaging displayed values? process
3. What prevents comparison across incompatible units, periods, population scopes or definition revisions? validation

Source field and lineage references

Define logical inputs and evidence references without copying datasets or owning lineage collection. A derivation link records a claim; it does not prove data correctness.

1. Which dataset and schema revisions supply the required logical input fields? relationship
2. What evidence links each operand to its source or derived dependency? provenance
3. What happens when a dependency is unavailable, changes type or loses its required version pin? exception

Implementation equivalence claims

Keep executable representations separate from the semantic definition. Syntax acceptance or a shared metric name is insufficient evidence of equivalent results.

1. Which evaluator, language version and implementation digest are bound to this revision? interoperability
2. What fixture evidence supports equivalent behavior for each implementation binding? evidence
3. Which unsupported operators or semantic differences prohibit a binding from being treated as equivalent? constraint

Semantic mappings and loss disclosure

Map only compatible concepts. A quality-metric profile may align to DQV; general business metrics are not automatically instances of that vocabulary. No serialization is the definition itself.

1. Which external concept or specification term is mapped, and what is the exact scope of the mapping? interoperability
2. Which formula, grain, time or policy semantics are lost in each export projection? quality
3. What round-trip evidence shows that import and export preserve the declared revision semantics? validation

Semantic revision and approval

Proposed states are draft, in-review, approved, deprecated and retired. Approval is a scoped local decision; it is neither canonical research status nor proof of runtime correctness.

1. What is the current revision state and which evidence permits the next transition? state
2. Which changes to formula, grain, filters, units or time basis require a breaking revision or a new metric identity? lifecycle
3. When does a revision apply and how are overlapping applicability intervals or historical corrections resolved? temporal

Retirement and retained interpretability

Retiring a definition does not erase existing observations. Retention follows the adopting policy, with controlled evidence disposal and minimal tombstones where permitted; no indefinite personal-data retention is implied.

1. Which policy controls retention or deletion of definitions, approvals and sensitive examples? retention
2. Which consumers and observations depend on the retired revision and how are replacement references disclosed? relationship
3. What minimal reference remains after authorized disposal, without retaining restricted payloads? privacy

Semantic acceptance fixtures

Propose fixture expectations for metric bindings, not a new evaluator. Metadata schema success, fixture success and domain fitness are separate claims. No execution reports are supplied by this research.

1. Which positive and negative fixtures test the declared metric semantics? validation
2. What expected results and tolerances distinguish weighted ratios, duplicate contributions and time-boundary cases? measurement
3. Which evidence supports acceptance and which untested cases remain explicitly unknown? evidence