Challenge 101: Inference Without an Automated Provider¶
Recorded: 2026-11-14.
Executive Summary¶
A Node should support an explicitly attended text-inquiry profile when its only available model is a chat operated separately by the user. Preparing an export packet and importing a response must not require an already working model.
Context and Problem Statement¶
Copy-paste can connect an external consultation to the existing inference contract, but does not establish provider identity, locality or execution. Waiting for a person must survive restart without keeping a runtime invocation open, replaying disclosure, or extending grants. User participation does not override confidentiality or authorize publication.
Proposed Direction¶
P101 defines a native host manual-exchange transport under Inquirium, reusing P089 result/commit patterns, S029 deferred operations, P066 consent and existing interaction surfaces. Agent, Assistant Channel and Corpus remain consumers, not separate bridge owners.
Trade-offs¶
This removes the requirement for an automated model provider for supported attended inquiries, at the cost of human attention, latency and unverifiable external execution. It does not supply embeddings or autonomous background work.
Failure Modes and Mitigations¶
Gate export before packet visibility; bind imports to principal and operation; persist one immutable packet under finite retention; reject ambiguous framing; never automate the external browser or upgrade an assertion into execution proof.
Open Questions¶
Higher-class export, multimedia and unattended batch providers remain outside the first profile. They require separate decisions and evidence.
Next Actions¶
Freeze P101's contracts and qualify a restartable exchange through the real UI endpoint with all automated model candidates absent.