Add macOS parity reference evidence for the Partial ledger rows

investigation/macos-parity-evidence/README.md

@@ -0,0 +1,69 @@
+# macOS parity evidence
+
+Reference evidence captured from the **shipping macOS app** so the Windows port
+has an authoritative comparison target for the `Partial` rows in
+[`../ui-parity-matrix.md`](../ui-parity-matrix.md).
+
+It was produced by running
+[`../macos-parity-evidence-agent-prompt.md`](../macos-parity-evidence-agent-prompt.md)
+on a macOS host.
+
+| | |
+| --- | --- |
+| App build | GraphCode 0.1.76 (299) |
+| Built from | `717240cdb373a2c00644842c429f06e520d15f51` |
+| Host | macOS 26.7 (25G229), Apple Silicon |
+| Captured | 2026-09-28 PDT |
+| Displays | 3840x2160 physical / 1920x1080 logical (2x); 1920x1080 (1x) |
+
+## Contents
+
+- [`macos-parity-evidence-report.md`](macos-parity-evidence-report.md) — row-by-row
+  findings for the 35 `Partial` rows, verified requirements, Windows
+  implementation candidates (W1–W7), blockers, an explicit *Not verified* list,
+  and an artifact index.
+- [`evidence/`](evidence/) — 26 unaltered window screenshots, indexed by the
+  report. `evidence/README.md` maps them to ledger surfaces.
+
+## How to read it
+
+The report marks every claim with `S` (inspected source), `T` (test source), or
+`R` (direct observation of the running app), and rates confidence on the
+**bounded** finding rather than on blanket platform equivalence. Ten Swift
+suites were executed once — 113 passed, 0 failed, 0 skipped. That is focused
+automated evidence, not a full `make test`.
+
+Read the report's *Not verified* section before citing anything. Terminal VT
+output, IME, clipboard, Codespaces and a real update were deliberately **not**
+exercised, and are recorded as unobserved rather than absent.
+
+## What this evidence does and does not do
+
+It **does** establish macOS behavior that Windows work can be measured against,
+and it supplies fixed-size 1x/2x captures for visual comparison.
+
+It **does not** validate any Windows code path. A ledger row still requires its
+own Windows runtime evidence before it can leave `Partial`, and citing this
+report is not a substitute for that.
+
+Two observed macOS behaviors are **defects, not parity targets**, and must not
+be copied:
+
+- Nested Main `Rename…` is offered but does not open the rename alert
+  (root-only lookup in `ProjectFeature.swift`).
+- The edge cycle-guard form visibly clips its labels on this fixture.
+
+Accessibility sample buttons were also found unlabeled on macOS; fix Windows UIA
+on its own merits rather than matching that gap.
+
+## Provenance and hygiene
+
+Captures are unaltered and unannotated. Only disposable `Alpha`/`Beta` Git
+repositories, an isolated bundle ID and a temporary unregistered daemon were
+used; no real project, account, provider, Codespace or release was opened. The
+worktree sweeper's full-window capture was deliberately withheld because the
+product's recovery note renders a local path.
+
+Do not retouch, recompress or crop these files. Their evidentiary value depends
+on being byte-for-byte what the app rendered. Add new captures alongside them
+rather than replacing them.

investigation/macos-parity-evidence/evidence/README.md

@@ -0,0 +1,36 @@
+# macOS parity capture originals
+
+All PNGs in this folder are **unaltered window-specific screenshots** of an
+isolated GraphCode 0.1.76 (299) build from commit
+`717240cdb373a2c00644842c429f06e520d15f51` on macOS 26.7, captured
+2026-09-28 PDT. Only disposable Alpha/Beta local Git repositories and a
+temporary, unregistered daemon were used. No real project, provider, account
+or update was opened. The report at
+`../macos-parity-evidence-report.md#artifact-index` supplies **one row per
+original** with ledger mapping, exact PNG dimensions, logical capture window
+size, display scaling and the bounded claim each image supports.
+
+| Ledger surface | Original filenames |
+|---|---|
+| Main split view, welcome, global/Quick Chats/project empty states | `isolated-window-a-original.png`, `empty-welcome-original.png`, `empty-global-graph-original.png`, `empty-quick-chats-original.png`, `alpha-empty-project-original.png` |
+| Workspace lifecycle | `manage-workspaces-original.png` |
+| Global graph, lane/card layout, notebook grid and loop rows | `two-project-overview-original.png` (pre-Fit/offscreen), `two-project-overview-fitted-original.png`, `two-project-overview-1400x820-original.png`, `two-populated-project-lanes-original.png`, `alpha-composite-empty-original.png` |
+| Node creation and validation | `alpha-new-node-form-original.png`, `alpha-composite-form-original.png`, `alpha-goal-validation-original.png`, `alpha-nested-two-nodes-original.png` |
+| Edge creation and presentation | `alpha-new-edge-form-original.png`, `alpha-edge-cycle-controls-original.png` (observed form clipping), `alpha-created-handoff-original.png`, `alpha-handoff-fitted-original.png` |
+| Node rename | `nested-rename-no-dialog-original.png` (post-click; see report for AX/source context), `alpha-root-rename-prompt-original.png` |
+| Worktree notice boundary | `eight-worktree-notice-original.png`, `seven-worktrees-no-notice-original.png` |
+| Native appearance | `native-settings-default-original.png` (light Settings versus dark app) |
+| Per-monitor DPI and sample visual fidelity | `two-populated-project-lanes-1x-original.png`, `two-populated-project-lanes-2x-return-original.png` |
+
+The 900x524 and 1400x820 figures refer to **logical app window points**.
+PNG dimensions include variable shadow/sheet margins. The ASUS display uses
+2x (3840x2160 physical, 1920x1080 logical); the Dell uses 1x
+(1920x1080 physical/logical). Changes in elapsed labels, selection, and
+native material prevent unqualified whole-image pixel comparisons.
+
+Menus were inspected via macOS Accessibility rather than photographed outside
+the app window. The worktree sweeper was also inspected via Accessibility:
+its full-window capture was withheld because the product's recovery note
+displays a local path. No runtime evidence was fabricated for terminal-backed
+tabs, Codespaces, IME/clipboard, or a real update. The 113 focused Swift tests
+are described in the report; screenshots are **not** screenshots of those tests.