R0023/2026-03-25/Q004/H3¶
Statement¶
Frameworks are emerging but focus narrowly on versioning and testing, not full lifecycle. Published guidance exists for parts of the lifecycle (versioning, evaluation) but no comprehensive framework addresses the full prompt lifecycle including deprecation, maintenance, and cross-model migration.
Status¶
Current: Supported
The evidence clearly shows partial coverage: versioning and testing are well-addressed by both vendor frameworks (AWS) and tooling (Langfuse, PromptLayer). However, no published framework covers:
- Deprecation: How to formally retire prompts that no longer work
- Cross-model migration: How to port prompts between model families
- Maintenance: Ongoing prompt health monitoring and update processes
- Governance: Decision frameworks for when to update vs. rebuild prompts
Supporting Evidence¶
| Evidence | Summary |
|---|---|
| SRC01-E01 | AWS covers versioning and testing but not deprecation or cross-model migration |
| SRC02-E01 | PEPR covers regression prediction but not full lifecycle |
| SRC03-E01 | Industry tools focus on versioning and evaluation |
Contradicting Evidence¶
No evidence directly contradicts H3. All sources confirm partial coverage with gaps.
Reasoning¶
H3 is the most accurate description of the field. The tooling ecosystem is maturing for versioning and evaluation, but the conceptual framework for full lifecycle management has not been published. The field draws heavily from software engineering lifecycle concepts but has not developed prompt-specific methodology for the unique challenges of managing probabilistic text inputs.
Relationship to Other Hypotheses¶
H3 reconciles H1 (frameworks exist — partially true) and H2 (no formal frameworks — partially true) by identifying what exists and what is missing.