Framework Engineering definition and boundary validation REP

1. Executive conclusion

Framework Engineering is not currently supported as a distinct engineering discipline. The repository has a coherent research program, executable pilot profile, theory, hypotheses, comparison machinery, provider execution paths, and validation tools. Those are genuine assets. They do not establish a distinct causal mechanism, independent outcome value, a mature professional practice, or an institutional field.

The strongest current classification is:

Framework Engineering is a repository-centered research program and a candidate integrated engineering profile for designing, evaluating, operating, and evolving analytical frameworks as versioned knowledge artifacts for human and machine use.

No single comparator fully subsumes that exact unit-of-analysis and integration boundary. Collectively, however, adjacent fields substantially cover every mechanism identified in FE. Systems engineering supplies lifecycle and integration; method engineering supplies construction and tailoring; design science supplies artifact build/evaluate logic; knowledge and ontology engineering supply semantics and provenance; requirements, decision analysis, quality engineering, cybernetics, organizational learning, and agent engineering supply the remaining controls.

The distinction that remains is a profile boundary, not a demonstrated discipline boundary. Its value must be tested.

This REP is a proposed successor synthesis, not an experimental override. EX-FE-0002 remains Stage A incomplete, Stage B blocked, and experimentally inconclusive. This execution supplies a broad source registry and subsumption analysis that can repair part of its input gap, but it does not satisfy independent curation/audit, blinding, recognition, sealed-key, or reviewer gates. EX-FE-0003 has no execution data.

2. Research question and candidate explanations

Question: Does FE have a useful non-redundant boundary relative to adjacent disciplines, or is it better classified as an integrated method profile or repository-specific research program?

Four candidate explanations were frozen before external-source interpretation:

1. distinct discipline;
2. integrated method/engineering profile;
3. repository-specific research program;
4. full subsumption.

The 12 comparison dimensions, ten adjacent comparators, evidence rules, and decision rules are preserved in the execution pre-registration. They were not revised after observing sources.

3. Method

3.1 Evidence collection

The review prioritized:

• current standards and official standards-body records;
• professional bodies and maintained bodies of knowledge;
• original peer-reviewed work defining adjacent methods;
• official handbooks for lifecycle, decision, measurement, and professional practice;
• direct repository inspection and executed validation baselines.

Several ISO records were available only as official abstracts and metadata. They were not treated as if their paywalled normative text had been read.

3.2 Falsification strategy

The review did not search only for terminological similarity. It steelmanned each adjacent field, mapped each FE mechanism to its strongest existing home, searched for individual and collective full subsumption, and applied institutional maturity tests. Evidence that FE has many files, schemas, or tools was excluded from distinctiveness.

3.3 Limits of method

Literature can establish precedent, overlap, and open hypotheses. It cannot establish FE's causal utility. No systematic bibliometric review, practitioner survey, matched experiment, or independent field replication was performed.

4. Direct findings

F1 — FE is an executable internal research program

The repository contains governing documents, a provisional theory, hypotheses, experiment records, comparison instruments, evidence workflows, an accepted-for-pilot ROS profile, record schemas, provider executions, generated registries, and tests. Current baselines execute successfully. [EV-FEB-R01, EV-FEB-R04]

F2 — Current repository authority does not support discipline status

The current theory and prior evaluation retain low confidence in distinctiveness and empirical value. No approved successor establishes otherwise. [EV-FEB-R02]

F3 — FE mechanisms are substantially inherited

The mechanism-level matrix found established coverage for artifact lifecycle, method construction and tailoring, architecture representation, semantics, provenance, requirements, decision quality, measurement, validation, configuration, feedback, organizational learning, and agent coordination. No clearly non-subsumed mechanism was found. [EV-FEB-002, EV-FEB-004, EV-FEB-006, EV-FEB-008–025]

F4 — Individual full subsumption is incomplete

Systems engineering can include conceptual systems and provides the broadest lifecycle umbrella, but it does not foreground cross-domain analytical frameworks. Method engineering directly covers method construction and tailoring, but not every framework is a method. Design science directly covers artifact build/evaluate logic, but is artifact-generic. [EV-FEB-002, EV-FEB-004, EV-FEB-008–010, EV-FEB-014–015]

F5 — Collective functional subsumption is substantial

When adjacent fields are combined, every current FE mechanism has an established conceptual or operational home. The remaining FE candidate is the deliberate integration of those mechanisms around one object and one portable evidence/operation model. This is an inference from the comparison matrix, not a source's direct claim.

F6 — Institutional maturity is absent

Established disciplines exhibit independently maintained bodies of knowledge, education and accreditation structures, competence assessment, professional communities, standards, ethics duties, and independent practice. FE has internal analogues for some research infrastructure but no external community, curriculum, credential, standard, profession, or independent outcome base. [EV-FEB-001, EV-FEB-005, EV-FEB-028–030, EV-FEB-R05]

F7 — Incremental utility is the decisive unknown

No current FE experiment matches the profile against strong adjacent methods while controlling provider, task, documentation dose, and evaluation conditions. Existing work cannot distinguish FE-specific effects from generic structure, prompting, recognition, or review effort. [EV-FEB-R03]

F8 — Prior experiment gates remain controlling

EX-FE-0002 already specifies a stronger blinded mechanism-boundary test. Its v1.1 Stage A gate is incomplete and Stage B is blocked; the machine-only v1.2 pilot is also incomplete. EX-FE-0003 is defined but its final analysis records no execution data. This review does not convert those missing runs into evidence. [EV-FEB-R06]

5. Definition and boundary

5.1 Current-state definition

FE is a repository-centered research program that develops and tests representations, comparison methods, evidence controls, and operating practices for analytical frameworks.

5.2 Recommended normative definition

FE is an integrated engineering profile for designing, evaluating, operating, and evolving analytical frameworks as versioned knowledge artifacts, with explicit purpose, semantics, evidence, uncertainty, provenance, context, and human/machine execution boundaries.

5.3 Object

An analytical framework is a reusable, bounded structure that organizes concepts, relations, questions, procedures, evidence, or decision logic to help a person or machine explain, diagnose, design, compare, decide, coordinate, or learn.

5.4 Boundary

FE is warranted when the framework is the primary object, the work crosses multiple lifecycle activities, and semantics/evidence/provenance/transfer materially affect outcomes. Use an established discipline directly when it covers the work without meaningful loss. Keep work in research posture while net benefit remains untested.

6. FEH-001 confidence update

This is not a claim that a distinct FE discipline is impossible. It is a decision that the current burden of proof is unmet and the narrower classification should govern present work.

7. Decision record

DF-FE-BOUNDARY-2026-07-28

Status: proposed for governance review.

Decision:

1. Describe current FE as a research program and candidate integrated engineering profile.
2. Do not describe it as an established or validated engineering discipline.
3. Preserve ROS–FE Profile v1.0 as the accepted pilot baseline.
4. Route profile enforcement defects to a v1.1 proposal rather than silently changing v1.0.
5. Make the matched incremental-utility experiment the next empirical gate.
6. Delay general framework generation, credentialing, organization-wide mandates, and autonomous canonical promotion.

Alternatives rejected:

• Declare a discipline now: rejected because mechanism, efficacy, institutional, and independent-practice evidence are absent.
• Terminate FE immediately: rejected because individual full subsumption is incomplete and an integrated profile may still yield useful transfer, auditability, or recovery outcomes.
• Continue building infrastructure before utility testing: rejected because infrastructure growth cannot resolve FEH-001 and increases sunk cost.

Reversal condition: replace this decision when controlled, independently replicated evidence establishes or falsifies incremental value, or when a stronger systematic review demonstrates individual full subsumption.

Compatibility with existing experiment authority: before the utility study is treated as the next executable run, use this source packet to repair and complete the independent Stage A/Stage B decision path in EX-FE-0002, or record an authorized decision explaining why the mechanism test is superseded. Until then its inconclusive classification remains controlling.

8. Operating-system result

The accepted v1.0 profile passes its current tests and registry workflow. It also has material enforcement gaps:

• execution-local records are not validated;
• execution packet completeness is not checked;
• duplicate IDs and many reference types are not rejected;
• several supported record types and taxonomies lack generated registries;
• stale registries are not detected;
• tests lack negative and compatibility cases;
• cross-repository ROS dependencies are not machine verified.

The v1.1 priority is evidence and execution integrity, not additional surface area. Full findings are in ros-fe-profile-v1-audit.md.

9. Research-to-engineering architecture

Research work should flow through:

governance → mission/pre-registration → source/observation → interpretation → artifact engineering → experiment/evaluation → decision/release → monitoring/retirement

Draft, submitted, reviewed, accepted, superseded, and rejected states must remain distinct. Generated indexes are not authority. Research is exported to engineering only through a versioned packet of accepted claims, requirements, schemas, conformance tests, risks, evidence, unresolved hypotheses, migration, rollback, and telemetry.

The immediate maturity gate is comparative utility (G3). Professional credentialing or discipline branding is several gates downstream.

10. Next discriminating experiment

The next study compares four matched arms:

1. minimal structured practice;
2. a strong adjacent profile combining SE lifecycle, situational ME tailoring, DSR/FEDS evaluation, and provenance;
3. the full FE profile;
4. an FE-shaped ablation without its candidate-specific genome and explicit evidence/contradiction/confidence separation.

Across multiple domains and task families, blind evaluators score downstream correct action and evidence integrity. Transfer, adaptation, error detection, provenance, confidence calibration, time, tokens, review effort, cognitive load, maintenance, safety, and recovery are secondary outcomes.

The primary contrast is full FE versus the matched adjacent profile. The mechanism contrast is full FE versus the ablation. A null or equivalent result against the adjacent profile triggers renaming/simplification; a replicated benefit justifies continued profile research, not immediate discipline status.

Execution dependency: complete or explicitly supersede the still-blocked EX-FE-0002 mechanism-boundary protocol first. Its blinded mapping can identify which FE components merit the ablation and prevent an outcome study from testing an undefined bundle.

11. Claim traceability

12. Risks and controls

13. What not to build yet

• general automatic framework generation;
• autonomous acceptance or theory mutation;
• a universal framework ontology;
• mandatory organization-wide full-mode governance;
• credentialing, accreditation, or discipline marketing;
• additional dashboards without measured decision or recovery value;
• production claims based solely on repository process quality.

14. Limitations

• The review is broad but not systematic in the formal bibliometric sense.
• Some standards were accessible only through official metadata.
• Comparator fields contain schools and variants not exhausted here.
• Exact-term negative search does not prove no external field exists.
• No external practitioner adjudicated the steelman descriptions.
• No new efficacy study was run.
• Existing EX-FE-0002 and EX-FE-0003 experimental gates remain incomplete; this same-provider synthesis is not their substitute.
• The matched experiment still needs independent method review, power analysis, case construction, and preregistration.

15. Completion and handoff

FE-MISSION-001 is complete at the boundary-research level:

• candidate explanations and dimensions were preregistered;
• primary and official evidence spans all comparators and dimensions;
• strongest adjacent-field descriptions and full-subsumption search are recorded;
• descriptive, normative, and operational definitions are separated;
• FEH-001 is updated;
• v1.0 is audited without silent modification;
• the next discriminating experiment is specified;
• machine-readable evidence and a cold-start handoff exist.

The mission does not establish efficacy, external validity, or discipline status. The next authorized work is experiment design review and preregistration, followed by the matched artificial evaluation. A successor must preserve null and negative outcomes and update this package through a new record, never by erasing the current result.

16. Verification

At closure:

• all new JSON parsed successfully, and all 11 execution-local records passed a separate required-field, profile-identity, and unique-ID check;
• profile registry generation and validation passed;
• profile tests passed 2/2;
• experiment tests passed 7/7;
• all ten registered experiments verified;
• research validation passed;
• the research publisher built 1,267 pages and indexed 1,200 in the final current-worktree build;
• ROS rebuilt six affected registries, validated successfully, and reported registries current.

These are conformance and publication results, not efficacy evidence.