Which Harness objects survive live provider replacement without duplicating Host or Runtime state?
Prove or delete a Host-local Harness boundary across provider-faithful lifecycles, replacement, faults, and completion adjudication.
01 / Current position
A hypothesis is not the current judgment.
The dossier preserves the live research position. Dated articles preserve the complete evidence and argument that changed it.
Host-local Assignment, HarnessRunReceipt, CompletionProposal, capability manifest, generation fencing, and completion adjudication will survive Codex App Server and Hermes ACP, while provider Thread and Session identities remain local evidence.
Answered by H1–H5. Both replacement orders completed through one Task Attempt and fresh Assignment generation; stale completion, missing Artifact, and response loss were handled correctly. Provider final text was not portable evidence, and Codex/Hermes lifecycles remained materially different. The durable Host boundary survived; a shared internal Provider lifecycle did not.
02 / Decision boundary
What keeps this Question alive?
A shared Harness abstraction can erase useful provider capabilities or create a second Task and Session system. Direct drivers can also duplicate assignment, evidence, fencing, and completion logic unless the stable overlap is measured.
Keep the shared mature-Provider lifecycle closed. Reopen only if a second independent consumer proves stable shared semantics and measurable net deletion without Provider capability loss. Use the separate Ordivon Harness v0 question for bare model APIs.
Direct provider drivers plus ordinary Host Task state remain simpler and equally correct, or a shared interface destroys material provider capability or adds more lifecycle code than it removes.
03 / Supporting publications
Complete arguments connected to this Question.
3 dated publications currently document this research line.
From Tokens to Work: The Complete Agent Execution Stack
A first-principles walkthrough from token generation to tool execution, verification, TaskOutcome, and graph advancement, compared with OpenAI, Anthropic, Microsoft, Google, and Ordivon's tested boundaries.
31 July 2026 ↗Research reportWhat Survived When Codex and Hermes Replaced Each Other Mid-Task
How one task remained coherent through Codex→Hermes and Hermes→Codex replacement, three injected faults, and materially different provider lifecycles.
31 July 2026 ↗Architecture decisionWhy Ordivon Needs a Harness—but Not a Universal Harness
Why H5 correctly rejected a common Codex/Hermes lifecycle while a separate Ordivon Harness remains strategically necessary for bare model APIs, local inference, and controlled agent-loop research.
31 July 2026 ↗04 / Source discipline
The dossier is an index, not the evidence authority.
Owns the current judgment, next test, and deletion condition.
Own complete dated arguments, limitations, comparisons, and source links.
Own exact code, tests, releases, receipts, and machine evidence.