10 projects, three kinds of work, no single mandatory stack.

The 0.3 contraction removed unconsumed Goal coordination, foreign execution, legacy execution workloads, generic capability policy, and Runtime-integration residue. Current 0.4.x then makes external semantic continuity first-class: one WorkingCheckpoint can survive client replacement with revision-coherent resume and exact response-loss replay, while every Runtime/Git/domain hint remains explicitly subject to owner-native revalidation.

Current Status and Architecture freeze six distinct state domains—canonical history, durable cognition, interaction cognition, attempt cognition, execution control, and effects—while HarnessAgentRun provides the supported Python composition surface. Recent verification work binds typed evidence to creation lineage and fails closed on structured-result conformance without promoting semantic evaluation or Provider continuation into Harness.

The C1–C10 operational-realization programme saturated its current evidence horizon without reopening Runtime Foundations: currentness, consequential retry, evidence scope, identity/lineage, authority/binding, lifecycle retention, and reclamation remained separable. Current production also exposes workspace.content for digest-bound pixel observation, runtime.describe for capability/ceiling discovery, and workspace.execBound on Linux plus configured limited windows_native targets; the latest fixes preserve fresh remote source currentness, reclaim lifecycle contracts, and terminal observation time instead of guessing after the fact.

On 2026-08-24 natural Product Value dogfood traced a Core cargo successor defect, rejected unrelated Heat/scan fixes, and promoted one objective-scoped correction after exact regression, 18-profile negative controls, evidence revalidation, and 350/350 repository checks. The current matrix keeps Rescue at 3/6 focus-complete victories and Hive at 6/6 while Core rises from 1/6 to 5/6; v2 remains the only registered product.

The 2026-08-18 WDF4 Causal Reality, WDF5 Property Evaluation, and TSAF0 Temporal Structure reconciliation admitted no new Causal, Measurement, or Temporal subsystem: current World already separates owner-native occurrence/admission from causal truth, bundle/observation integrity from measurement validity, and provider execution time from World availability and Host admission. Those results were consumed into the existing 0.4 production boundary rather than expanding it.

The current capital-state path now aggregates portfolio observations fail-closed across component scopes, preserves balance FinancialResources even when risky positions are zero, separates held state from valuation time, records replayable carrier/exposure mappings, and bridges the exact managed-capital scope into owner contribution. Nonzero managed FinancialResources explicitly forbid the historical QB5 zero-return shortcut when aligned return evidence is missing. Context v21 additionally projects current execution/reconciliation work while keeping completed historical detail canonical and replayable; the state-changing Decision baseline fence remains enforced at Proposal creation.

Phase 1 completed the Studio→Media owner inversion and pressure-tested OMPC on four materially different real consumers: Host projection, Studio transformation lineage, Web interaction trajectory, and Finance metric/legibility. The result retained a six-role / eleven-invariant shared reference contract while explicitly rejecting a universal Media framework, engine, runtime, SDK, serializer, validator service, or owner absorption.

The current programme demonstrates both expansion and refusal to expand. HUMAN-AI-001 now consumes mature external evidence first and targets only Ordivon-specific transport residuals; HF0–HF23 preserve reusable distinctions across materially different Human questions; and HD9 completed the organismic health/disease/pathophysiology route while explicitly rejecting HF24 and leaving HD10/NextDeepRoute UNKNOWN rather than manufacturing the next stage.

On 2026-08-28 Security completed a destructive executable-topology contraction and retained the P1/R2 current-capability result: owner-native mechanical applicability can compile the exact model-facing operation set, and the returned request is checked against that same surface. In deterministic real-provider fault injection, a broad static surface let UNKNOWN/UNAVAILABLE intents reach providers (one UNKNOWN action even produced a verified physical consequence), while the current-compiled surface rejected the same requests before provider invocation. This strengthened the admission boundary without earning a shared planner, registry, compiler, gateway, Trust system, or freshness service.

The 1760–2026 historical dogfood closed without turning historical resemblance into proof of Ordivon theory, and the current Computer contraction reduced active shared machinery to a small responsibility map. On 2026-08-22 the canonical Research Method was strengthened to preserve imported mature knowledge through novelty subtraction, expose that method at first look, adjudicate closeout destinations, and materialize the Resource Ecology result as reusable Knowledge without creating a Resource owner or calculus.