← Back to catalogue
Published

Service Level Objective / Reliability Commitment

vr.wm-sft-016 · wm-sft-016-service-level-objective-reliability-commitment

Describe versioned reliability objectives, evidence-bound evaluations and error-budget policies for a service or journey.

World Models Information and virtual systems INF.SFT.SLO

Bundle → Layer → Finding → Questions Filled

3 bundles · 3 layers · 5 findings · 11 questions

Indicator definition What is measured and how.

Service level indicator

The measurement the objective is based on.

Indicator specification

The events counted as good and valid, and the data source.

  1. Which indicator does this objective use, and how are good and valid events defined?
  2. From which telemetry source and query is the indicator computed?

Measurement scope

The service, journey and user population covered.

  1. Which service and user journey does the indicator cover, and for which users?
  2. Which traffic is excluded from the measurement, and on what grounds?
Objective and budget The target and what happens when it is missed.

Target

Threshold, window and version of the objective.

Threshold and window

The target value and the period it is evaluated over.

  1. Which target value and evaluation window does this objective set?
  2. Which version of the objective is in force, and since when?

Error budget rules

The budget derived from the target and the actions it triggers.

  1. How much error budget does the target allow in the window, and how much is consumed?
  2. Which actions follow when the budget is exhausted or burning fast?
Evaluation and accountability How results are recorded and who answers for them.

Evaluation record

Results per window and the owning team.

Evaluation result

Whether the objective was met in a window.

  1. Was the objective met in the last completed window, and by what margin?
  2. Which team owns the objective, and which service does it protect?
  3. Which incidents consumed the budget in this window?

Classifiers Filled

Family
World Models
Category
Information and virtual systems
Entry kind
aggregate
Navigation path
NAV.INF.SFT.SLO
Domain
INF.SFT.SLO
Industry
Cross-industry
Tags
servicelevelobjectivereliabilitycommitmentinf.sft.slo

What it is Filled

A service level objective is a stated reliability target for a software service, defined as a service level indicator measured over a window with a threshold, together with the error budget rules that follow from it. The objective is the policy record; the telemetry that measures it and the contractual agreement with customers are separate things.

In scope

  • Commitment identity, immutable SLI/SLO policy revisions, target criteria and observability bindings
  • Resolved evaluation windows, observation references, evidence sufficiency, result corrections and budget accounting
  • Burn and response policy bindings, accountable review and controlled reliability reporting

Out of scope

  • Service identity, expected outcome and consumer-scope masters; software-product lifecycle
  • Observation identity, value, method, uncertainty, time and correction; collection and telemetry storage lifecycle
  • Contractual obligations, obligees, credits, remedies and enforceability
  • Calendar arithmetic and calendar lifecycle; notification delivery, incident operations, deployment control and generic audit-system execution
  • Executable instance schemas, production evaluator, automatic legal compliance and universal target recommendations

Why it exists Filled

Describe versioned reliability objectives, evidence-bound evaluations and error-budget policies for a service or journey.

Distinguishing features Filled

  • The objective is an internal reliability policy; a service level agreement is a contract with consequences for customers.
  • The indicator definition is part of the objective, so changing the query changes the objective.
  • Error budget rules link measured reliability to decisions about shipping and remediation.
  • Distinct from the telemetry signal it is computed from and from the incident records that consume the budget.

What robots and AI may and may not do Filled

Must not

  • Change a threshold, window or indicator query to make a missed objective look met.
  • Exclude traffic from the measurement without recording the exclusion and its reason.
  • Present an internal objective as a contractual commitment to customers.
  • Report results from an incomplete window as if the window had closed.
  • Suppress budget burn alerts to protect a release schedule.

Only with a human decision

  • Approving a new objective or a change to a threshold, window or indicator.
  • Deciding to halt or continue releases when the error budget is exhausted.
  • Communicating reliability results to customers or regulators.

May

  • Compute indicator values and error budget consumption from the declared telemetry source.
  • Report objective results per window with the data source and query version.
  • Raise alerts when budget burn exceeds the declared rules.
  • Propose objective changes with the evidence behind them.

Moral aspects Filled

  • Reliability targets decide which users' failures are treated as acceptable, and averages can hide groups that always suffer.
  • Error budget pressure falls on on-call staff, whose working conditions are part of the cost.
  • Services people depend on for health, money or safety deserve targets set with that dependence in mind.

Who is affected

  • Users of the service
  • On-call engineers and support staff
  • Customers relying on contractual commitments
  • Teams owning dependent services

Owners Filled

Steward

The team that operates the service and answers for its reliability targets.

Roles

Reliability steward
Maintain scope, review evidence and propose policy revisions
Policy approver
Approve targets, effectivity, response rules and scoped exceptions
Evidence custodian
Maintain authorized evidence bindings, quality assessment and retention references
Evaluation operator
Run an approved profile, preserve inputs and emit indeterminate outcomes when necessary
Assurance reviewer
Challenge coverage, corrections, mappings and unsupported assurance claims

Master systems

  • Observability and monitoring platform
  • Service catalogue
  • Incident management system

Links to other meta-models Filled

references

  • WM-MAT-008 - Evaluation cites observations and absence evidence for a window. Observation identity, value, method, uncertainty, time and lifecycle remain observation-owned. Abstract references need a pinned instance binding before operation.
  • WM-ACT-004 - SLO-local reliability subject binding may reference a service definition. Service identity, expected outcome and consumer scope remain service-owned; a journey can be selected without creating a service master.
  • WM-ECO-006 - Contract may cite policy and evaluation evidence. Obligations, obligees, credits, remedies and enforceability remain contract-owned; no containment or automatic breach conclusion.
  • WM-XCT-009 - Calendar-aligned profiles pin calendar identity, timezone and period conventions. Calendar arithmetic, version and lifecycle remain calendar-owned; elapsed windows need no calendar instance.
  • wm-sft-017-telemetry-operational-signal - The indicator is computed from telemetry signals.

aligned

  • OpenSLO specification - Conceptual interchange alignment only. Current README supports common SLO exchange intent, not a validated field mapping or evaluator implementation.
  • OpenTelemetry Metrics Data Model - Observation adapter preserves resource and metric identity, temporality, units and missing-value semantics; the SLO aggregate does not implement collection.
  • PROV-O - Conceptual derivation and revision links for result provenance. No ontology conformance or evidentiary truth certification.

neighbor

  • WM-MAT-008 - Evaluation cites observations and absence evidence for a window. Observation identity, value, method, uncertainty, time and lifecycle remain observation-owned. Abstract references need a pinned instance binding before operation.
  • WM-ACT-004 - SLO-local reliability subject binding may reference a service definition. Service identity, expected outcome and consumer scope remain service-owned; a journey can be selected without creating a service master.
  • WM-ECO-006 - Contract may cite policy and evaluation evidence. Obligations, obligees, credits, remedies and enforceability remain contract-owned; no containment or automatic breach conclusion.
  • WM-XCT-009 - Calendar-aligned profiles pin calendar identity, timezone and period conventions. Calendar arithmetic, version and lifecycle remain calendar-owned; elapsed windows need no calendar instance.
  • Registry and supplement - The registry standalone-mm value is a record-plane classification; aggregate is the subject kind. No legacy_alias exists. The unreviewed supplement is reconciled as a lead, with properties and cross-service comparability qualified.
  • world-k4-service - Contractual service level commitments to requesters live in the service model.

part-of

  • wm-sft-001-software-product - An objective protects a service of a software product.

What else AI and robots need to interact with it Filled

Identity and identifiers required Filled

  • An objective is identified by a name scoped to its service in the service catalogue, with a version.
  • The indicator definition cites the telemetry query and its version as part of that identity.

Direct properties not applicable Not applicable

Not applicable

An objective is a policy record, not a physical object; its threshold, window and budget figures are part of its structure.

Recognition optional Filled

  • Marked by an indicator definition, a numeric target, an evaluation window and an owning team.
  • Often confused with a service level agreement, with an alert threshold or with a raw metric.

Capabilities and actions required Filled

  • Budget consumption and burn rate can be computed continuously within a window.
  • Results can be compared across windows and services.
  • Objective versions can be compared to show what changed and when.

Hazards and failure modes required Filled

  • Targets quietly weakened to avoid reporting a miss.
  • Indicators that measure the system rather than the user experience.
  • Alert fatigue from burn rules set without regard to on-call load.
  • Objectives treated as contractual promises they were never meant to be.

Standards and interfaces required Filled

  • OpenTelemetry for metrics, traces and logs feeding the indicator.
  • Prometheus and OpenMetrics exposition for metric collection.
  • ISO/IEC 20000-1 where objectives feed a service management system.

Context of use required Filled

  • Used by platform and product engineering teams to steer reliability work and release decisions.
  • Objectives for regulated or critical services may have to be shown to supervisors.

Sources Filled

  1. Service Level Objectives - Google
  2. Implementing SLOs - Google
  3. Alerting on SLOs - Google
  4. OpenSLO specification project overview - OpenSLO community
  5. Metrics Data Model - Cloud Native Computing Foundation
  6. Histograms and summaries - Cloud Native Computing Foundation
  7. RFC 3339: Date and Time on the Internet: Timestamps - Internet Engineering Task Force
  8. PROV-O: The PROV Ontology - World Wide Web Consortium
  9. OpenTelemetry Specification, Cloud Native Computing Foundation
  10. ISO/IEC 20000-1 Service management system requirements, ISO and IEC
  11. OpenMetrics specification, Cloud Native Computing Foundation

Open questions

  • Run coordinator source checks, pin mutable documentation and restore independent external review before canonical or publishable-draft promotion.
  • Develop instance schemas and fixtures for zero traffic, absent streams, 100 percent targets, partial windows, late evidence, query changes, resets, calendar transitions, overlapping windows and correction lineage.
  • Validate weighted, percentile, batch, durability and composite profiles plus loss-aware OpenSLO and monitoring mappings with adopting service owners.
  • Independent second-provider review is waived; the later no-tools phase is local self-audit only.
  • Direct source HTTP status, redirects and response hashes are unmeasured; mutable documentation releases and commit pins require coordinator verification.
  • Nested instance schemas, executable evaluator semantics and tested numeric, calendar and export profiles remain incomplete.
  • Composite dependency budgets, weighted or sampled SLI estimators, batch freshness and durability profiles require domain-specific extensions and validation.
  • No legal, regulated-service, licensing or security certification; adopting owners must set retention, confidentiality and critical-service assurance requirements.

Machine files

Provenance

world-models research · reviewable-draft

Built from: models/wm-sft-016-service-level-objective-reliability-commitment/spec.yaml, ver-cy/world-models/card-supplements/wm-sft-016-service-level-objective-reliability-commitment.json