Projects

Four projects preserve different parts of one durable work trajectory.

Start from the failure each project prevents, the capability it currently provides, and the evidence that supports its maturity. Ownership details remain on the project page.

01Agent system
Tested architectural core

Ordivon Host

Problem

A capable model can finish one run while the user's larger task, accepted decisions, and unfinished work disappear with the session.

What it does

Preserves goals, tasks, accepted decisions, assignments, and completion evidence above replaceable model runs and local processes.

Latest proof

One repository task completed in both Codex→Hermes and Hermes→Codex orders while three injected faults were rejected or recovered.

What Survived When Codex and Hermes Replaced Each Other Mid-Task
Open question · testing

Can Host complete a general repository Goal without absorbing Runtime mechanics?

See the next test ↗
02Agent system
Production-tested local runtime

Ordivon Runtime

Problem

Shell commands and transient processes do not preserve enough identity or evidence for a later agent to know what actually happened.

What it does

Turns admitted local operations into durable jobs, observable process state, bounded output, artifacts, cancellation, and restart recovery.

Latest proof

Thirteen public tools supported Host correlation, replay, fresh-client recovery, and terminal evidence with no new Runtime semantic objects.

How Ordivon Runtime Grew Beyond Its Ten-Tool Core
Open question · testing

Which real structured operation can complete the minimal Effect contract across a second backend?

See the next test ↗
03World interaction
Experimental boundary

Ordivon World

Problem

An external action can lose its response while paths, identities, and provider conditions change, making a blind retry unsafe.

What it does

Correlates target, path, identity, authority, provider execution, receipts, observations, uncertainty, reconciliation, and rebinding around one external interaction.

Latest proof

The former Link and Edge projects were unified and retired; the boundary now faces one decisive lost-response comparison against direct Host integration.

Why Link and Edge Became One Ordivon World Boundary
Open question · testing

Can one Host Task combine path evidence and provider execution, reconcile a lost response, and continue without duplicate work?

See the next test ↗
RFoundations
Active research foundation

Ordivon Computing

Problem

Agent systems can accumulate shared abstractions faster than real workloads justify them.

What it does

Compares proposed contracts against mature baselines and keeps only the responsibilities that survive replacement, interruption, and a deletion test.

Latest proof

Strong workflow, retrieval, idempotency, and provider baselines reduced the proposed core instead of validating a larger platform.

The Smaller Core That Survived Strong Baselines
Open question · testing

Which contracts survive a second independent workload without being tailored to one repository?

See the next test ↗

How they connect

The projects cooperate without becoming one mandatory pipeline.

A local Runtime job may never reach an external service. World can observe a path before Host admits a task. Computing can delete a proposed contract without owning production state.

Host owns the meaning and accepted progress of durable work. Runtime owns local execution facts. World owns conditioned external interaction. Computing tests which shared contracts remain necessary across those boundaries.

Source, tests, releases, receipts, and each repository's current documents remain the technical authority. This site provides orientation, evidence paths, and dated judgment.