Deterministic > Probabilistic

[visionik]

Whenever possible, we should use deterministic components for repeatable actions.

This strategy started with Directives use of Taskfile tasks for repeatable work; but we want to extended it to things that go beyond taskfile tasks.

[MScottAdams]

Roadmap Refresh Analysis (2026-04-03)

Category: Enhancement / design philosophy

Summary: Design principle: prefer deterministic components for repeatable actions over probabilistic (LLM) ones. Already started with Taskfile tasks; wants to extend beyond Taskfile to other repeatable workflows.

Relationship to existing issues:

• Deft as teacher: contract hierarchy, explain WHY, and adaptive teaching mode #84 (Phase 2/5) — Deft as teacher / contract hierarchy; complementary design principle about how agents should execute
• consider typscript instead of python #160 (Phase 5) — TypeScript CLI consideration; deterministic CLI with structured commands is one expression of this principle
• [Feature Request] Compliance-Aligned Constitution Templates + Readiness Scanners (SOC 2, ISO 27001, HIPAA, HiTrust) #95 (Phase 5) — Compliance scanners; compliance workflows are prime candidates for deterministic execution
• Artifact-branch binding and complete audit trail for SDD #86 (Phase 5) — Artifact-branch binding; deterministic audit trails align with this principle

Scope: Medium — cross-cutting design principle needing documentation then ongoing application

Recommendation: Phase 2 for principle documentation (meta/philosophy.md, contracts/hierarchy.md, or new meta/determinism.md). Ongoing application across CLI, skills, and workflows is naturally Phase 5.