Orchard Mk II foundations v0.1

Copyright © 2026 Kimberley Laverne Asher. CC BY-NC-ND 4.0. Licence and scope.

Framework reconstruction, reproducible evidence and a proposed common development structure
26 September 2026 · Stages 1 and 2 · Unicode mathematics

What is delivered

This release reconstructs the full eighteen-responsibility architectural map, records its algorithms and heuristics, explains its mathematical foundations and preserves the distinction between demonstrated finite mechanisms and wider proposals. It supplies approachable companions, a terminology and notation glossary, exact source snapshots, raw empirical data and a portable Python reproduction route.

Stage 2 proposes a common structure for development across different kinds of change. The user's new insight is also captured: contextual descriptions could construct mathematical investigations through typed intermediate representations, retain unresolved choices among remedies, and eventually earn extensions to the language or composition machinery itself. This remains a destination and design proposal, not a newly demonstrated compiler.

A–D have been reproduced; no Stage 3 experiment has been started. The next experimental choice is reserved for discussion. Historical A–D archives and source snapshots retain their original bytes.

Read in the way that suits you

Open the offline reading edition for a navigable view of all new documents. It works without a network connection. Markdown files below remain the primary editable source format; the rendered edition is generated from them.

How much is established

The architectural reconstruction covers all eighteen responsibilities in the foundational brief. The source library contains 98 exact files, including three newly supplied proposal dialogues, the twenty-chapter v0.10 framework and supplements, companions, primary anchors and forward/backwards records. Seven empirical archives preserve A–D and four additional historical evidence folders. The historical folders are included for inspection; they were not newly replayed here.

The new portable reproduction runs 25 focused tests across the three A–D packages, their full declared populations and separate auditors. A/B has 5,128 executions; C has 4,608 plus 388 composition microtraces; D has 224 sessions containing 49,536 transfer tasks. All logical comparisons pass, with the expected 13, 12 and 12 negative controls rejected respectively. These are different denominators, not a pooled count of independent experiments.

M means mathematical consequence under premises; E means executed finite evidence; P means proposal; H means architectural hypothesis; U means unresolved. Historical results read from their reports remain source-level evidence unless explicitly identified as reproduced. A source snapshot, schema validation or clear diagram cannot raise a capability's standing by itself.

What remains open

This is not a general alignment solution, a deployed model harness, a validated whole cognitive runtime or an unrestricted self-improving language. Context grounding, coverage, general composition, qualifier evolution, physical containment and the newly proposed cognitive compiler retain explicit open obligations. The source register's unresolved identity and canonical-release questions also remain open.

The package preserves the whole ambition while showing where the strong finite mechanisms actually connect. See citation and provenance and licensing scope before sharing.