Build Notes: a replay receipt is not a fresh run
Separate recorded results from re-executed work before testing an agent. A small run manifest makes the distinction inspectable.
Source-based editorialReviewed Oct 11, 2026
Deterministic AI
Latest stories
Our latest coverage, with original sources and clear evidence limits. Use the filters to explore a beat.
What Auditors Will Ask: show the decision, the rule and the exception
A proposed evidence packet for AI workflow reviews: authority, inputs, checks, exceptions and outcomes, with each claim kept within scope.
Source-based editorialReviewed Oct 11, 2026
Counterpoint: demand repeatability, then ask the next question
Repeatable computation makes an AI claim easier to inspect. Correct decisions still need authoritative inputs, rules and verification.
Source-based editorialReviewed Oct 11, 2026
LLM-42: take the fast path, then verify the computation
LLM-42 proposes verified speculation for deterministic inference, retaining a fast path while checking which tokens can be committed.
Microsoft Research2026-09
Same request, different batch: test the serving configuration
A documentation watch on vLLM and SGLang’s opt-in controls for batch-invariant inference, with scope and implementation limits.
vLLMReviewed Oct 11, 2026
Before you press replay, list what will run again
LangGraph and Temporal show why replay needs a precise definition. Check the checkpoint, event history and external actions.
LangChainReviewed Oct 11, 2026