Goal specification: minimum schema and completeness validation

Research Question

What properties must a Goal specification carry for an automated system to determine whether it is complete enough to act on -- specifically, what is the minimum schema, and what happens when fields are absent or contradictory?

Findings

(Populated from §6 Synthesis above.)

Executive Summary

A Goal specification requires five minimum fields for an automated system to determine whether it is actionable: intent statement (what the goal aims to achieve), initial or context conditions (what is currently true), success criterion (how completion is verified), unique identity, and scope boundary. These five fields appear in equivalent form across all four major Goal schema frameworks surveyed: GORE/KAOS, TOGAF/ArchiMate 3.2, IEEE 29148, and PDDL/STRIPS. When fields are absent, schemas split into two error modes: automated planning schemas (PDDL) use hard errors that halt execution, while human-mediated schemas (KAOS, IEEE 29148, ArchiMate) use degraded mode that flags incompleteness but continues processing. No schema surveyed provides automated contradiction resolution; all require human arbitration once a contradiction is detected.

Key Findings

1. The cross-schema minimum Goal schema contains five fields: intent statement, initial or context conditions, success criterion, unique identity, and scope boundary -- all of which appear in equivalent form across GORE/KAOS, TOGAF/ArchiMate, IEEE 29148, and PDDL.

2. A sixth field -- agent or responsibility assignment -- is mandatory in GORE/KAOS for leaf goals and recommended in IEEE 29148, but is not a required goal-level attribute in PDDL or ArchiMate, making it consensus-strong but not universal across all four schemas.

3. Automated planning schemas (PDDL) enforce a hard-error model: a missing :goal block or an undeclared predicate in the planning domain causes an immediate parse error and the planner refuses to execute any plan.

4. Human-mediated schemas (KAOS, IEEE 29148, ArchiMate) use a degraded-mode model: an incomplete specification is flagged but the system continues processing other goals, allowing partial models to exist and authoring workflows to proceed.

5. A missing success criterion is the most operationally severe absent field: without it no schema can verify plan completion, and automated planners halt while human-mediated schemas leave the goal permanently in an unverifiable state.

6. KAOS explicitly models contradictions between goals using obstacle analysis and conflict links, making contradiction detection a designed feature of the KAOS specification tooling rather than an ad-hoc check.

7. PDDL detects contradictory goals through planning search: a :goal conjunction of mutually exclusive predicates is syntactically valid but returns UNSOLVABLE at plan-search time, meaning the contradiction is not caught until execution is attempted.

8. No schema surveyed provides automated contradiction resolution; the consistent escalation path across all four schemas is to detect the contradiction, identify the conflicting fields, and require human arbitration before the specification can be acted on.

9. The IEEE 29148 CUBCOVF quality rubric (Complete, Unambiguous, Bounded, Consistent, Observable, Verifiable, Feasible) maps directly to the five minimum schema fields: Completeness requires all fields present, Verifiable requires a success criterion, Bounded requires scope definition, and Consistent requires absence of contradictory fields.

10. ArchiMate 3.2 motivation model Goals tolerate absent fields because they are communication artefacts rather than execution specifications; enforcing the five-field minimum therefore requires supplementary governance rules not built into the ArchiMate notation.

11. In PDDL, the closed-world assumption means an absent :init block effectively specifies that all predicates are false, which is a syntactically valid but behaviourally incorrect initial state -- a class of silent error that absent-field validators in human-mediated schemas do not need to guard against.

12. A 2023 goal-oriented requirements ontology paper proposes formalising completeness and consistency checks as first-order logic ontology reasoning rules, confirming that the missing-field and contradictory-field problems remain active research targets for automated enforcement.

Assumptions

• Assumption: The four schemas surveyed (GORE/KAOS, TOGAF/ArchiMate, IEEE 29148, PDDL) are sufficiently representative of the space of Goal schema proposals to identify a cross-schema minimum. Justification: These schemas span requirements engineering, enterprise architecture, international standards, and AI planning -- the four principal sub-fields in scope per the research question; other Goal schema proposals (OCL constraints, SysML v2) exist but are domain-specific extensions of the surveyed base schemas. Sources: Van Lamsweerde (2001; www.semanticscholar.org IEEE 29148-2018 (IEEE 29148-2018 Systems and Software Engineering -- Requirements Engineering planning.wiki (planning.wiki ArchiMate 3.2 (pubs.opengroup.org

• Assumption: Functional equivalents across schemas count as the same field for minimum-schema purposes (e.g., :goal in PDDL is functionally equivalent to operationalization / success criterion in KAOS). Justification: The schemas use different terminology for structurally analogous concepts; treating functional equivalents as equivalent fields is necessary to enable cross-schema comparison and is the standard approach in requirements engineering survey literature. Source: Van Lamsweerde (2001; www.semanticscholar.org planning.wiki (planning.wiki

Analysis

The five-field minimum is a cross-schema intersection result. A system designer who wants a Goal specification that is actionable across all four schema families must include all five fields; a system targeting a single schema family may operate with fewer (ArchiMate enforces only name; PDDL enforces :init, :goal, and :domain).

The hard-error vs. degraded-mode split is a design choice that reflects the recovery capability of the consuming system. PDDL planners have no mechanism to query a user for missing predicates mid-run; halting is the only defensible response to a missing field. KAOS tools and requirements management systems operate in an interactive authoring environment where an author can be prompted; degraded mode preserves workflow progress while surfacing the gap.

An automated delivery system or agentic AI workflow that consumes Goal specifications faces this same design choice. Hard-error semantics are safer for autonomous execution because silent continuation with an incomplete specification is harder to detect and diagnose than an explicit failure. Degraded-mode semantics are more suitable for iterative authoring workflows where partial specification is a normal intermediate state. The right choice depends on whether the Goal consumer has a channel to request missing information from its caller; if no such channel exists, hard-error semantics are the correct default.

Contradictory fields present a structurally distinct class from absent fields. Absence is a structural gap detectable with a schema validator (field present or not). Contradiction is a semantic conflict detectable only by reasoning over field values (are the intent statement and the success criterion mutually achievable given the initial conditions?). A completeness validator that checks only field presence will pass a Goal specification with contradictory fields because all fields are present even if their values conflict. A validation pipeline for autonomous action therefore requires two stages: a structural completeness check followed by a semantic consistency check.

The companion item on Goal scope change propagation (Mitchell 2026; davidamitchell.github.io establishes that constraint re-enumeration is not automatic in any current GORE or MBRE framework. The companion item on Goal constraint feedback convergence vs. cycling (Mitchell 2026; davidamitchell.github.io shows that goal-constraint feedback loops involving contradictory goal sets can cycle without converging, reinforcing the finding that contradiction detection must lead to human arbitration rather than automated re-resolution. The companion item on formal methods feasibility for interdependent inputs (Mitchell 2026; davidamitchell.github.io confirms that automated feasibility checking requires formal specification rather than schema validation alone, supporting the two-stage validation pipeline finding here. All three companion findings converge on the conclusion that automated systems cannot resolve semantic conflicts between Goals without a formal model of the stakeholder intent that produced the conflicting fields, and human arbitration is the only safe escalation path.

Risks, Gaps, and Uncertainties

• The TOGAF 9 Motivation Architecture specification page (The Open Group TOGAF Standard -- Motivation Architecture required authentication and was not directly accessed; TOGAF motivation model content was confirmed via the ArchiMate 3.2 Specification, which supersedes and extends the TOGAF motivation model for open publication.

• IEEE 29148-2018 full text requires an IEEE Xplore subscription; the field inventory was confirmed via the NASA Systems Engineering Handbook Appendix C, which operationalises IEEE 29148 criteria in an openly available form. The standard may contain nuances not captured in secondary sources.

• The SCIRP abstract for Taye and Ghoul (2023) was accessible; the full paper text was not. Findings attributed to this source are drawn from the abstract and section headings only.

• The scope covers four schemas; Object Constraint Language (OCL) constraints in Unified Modeling Language (UML), SysML v2 requirement assertions, and domain-specific Goal schemas (healthcare, defence) are not covered and may yield additional required fields or different error-mode patterns.

• No empirical benchmarks on the operational cost of absent-field errors in production delivery systems were found; the severity ranking of "missing success criterion" as the most severe absent field is inferential based on logical consequence rather than measured outcome data.

Open Questions

• For agentic AI systems that accept natural-language goal specifications, which of the five minimum fields are most frequently absent in practice, and what failure modes result? This could become a targeted empirical study.

• Is there a formal mapping from the five-field minimum schema to the PDDL formalism that would allow GORE/KAOS-style goals to be automatically translated into PDDL problem files, and what information loss occurs during that translation?

• What validation rule set would be sufficient for a lightweight Goal completeness checker embedded in a CI/CD (Continuous Integration and Continuous Delivery) pipeline, and could the CUBCOVF rubric be operationalised as automated unit tests on Goal specifications?

sources

• [ ] Van Lamsweerde (2009) Requirements Engineering: From System Goals to UML Models to Software Specifications -- GORE theoretical foundations and Goal schema from Axel van Lamsweerde
• [ ] The Open Group TOGAF Standard -- Motivation Architecture -- TOGAF motivation metamodel defining Driver, Goal, Outcome, Requirement
• [ ] IEEE 29148-2018 Systems and Software Engineering -- Requirements Engineering -- IEEE standard requirements for well-formed requirement and goal statements
• [ ] Russell & Norvig Artificial Intelligence: A Modern Approach -- Goal Representation -- AI planning goal specification (STRIPS/PDDL formalism)