Elora Engine 0.2.8 and the Governed Inference 1.09 chronology combine deterministic non-neural language processing, bounded neural assistance, runtime resource governance, governed interaction evidence, and proposal-to-commit authorization. Inference remains a proposal; execution authority remains with Governance.
Elora's inference path is hybrid but deliberately separated. Governed non-neural methods can interpret and construct bounded responses directly; neural assistance is optional, separately admitted, and never receives commit authority.
NNLSL uses governed lexical, semantic, relationship, and construction evidence to support deterministic language work. Educational material remains bounded and reviewable, and incomplete knowledge produces an explicit learning gap rather than fabricated certainty.
When non-neural processing cannot satisfy a request, Elora can consider a separately governed neural handoff. Model selection, provider routing, resource admission, and final authorization remain explicit boundaries rather than implicit fallback behaviour.
Deterministic turns are admitted against bounded memory and CPU controls, with lifecycle evidence for admission, use, breach posture, release, and cleanup. Resource failure can stop work before further processing while neural execution remains independently governed.
Public-safe chronology through 26 August 2026 for NNLSL, governed neural handoff, Observer reporting, research claim discipline, runtime resource boundaries, and inference evidence.
