Determinism & Performance Guardrails — temp=0, schema validation, caching, budgets

[DevVig]

Summary
Codify guardrails across the pipeline to ensure deterministic, fast, and repeatable runs: temperature=0, strict JSON validation, caching, and time/latency budgets.

Why

• Directly supports NFRs: determinism, performance, reliability

Scope

• Cross-cutting controls and configuration

Tasks

• Enforce temp=0 and top_p=1 defaults in a2a/llm.py with per-call override
• Require JSON schema validation before writing outputs (where applicable)
• Implement caching for expensive steps via a2a/core/cache.py hooks
• Add time budgets/limits in .a2dev/perf.yaml and respect them in orchestrator steps
• Add a run-id and record start/end times in docs/timeline/
• Update status line to include elapsed time where useful
• Document these guarantees in README_A2Dev.md

Acceptance Criteria

• Repeated runs produce identical artifacts (unless inputs change)
• Cached re-runs show improved latency within budget
• Status lines and timeline reflect timing

References

• suggestion.md (Guardrails implied in all recommendations)
• a2a/llm.py, a2a/core/cache.py, a2a/orchestrator.py