Independent Process native review — run 3

Reviewed on 2026-09-23, read-only. Directly inspected run-3 images 12–23, 27–44 and 49–52, and run-1 frame 26. No source edits, builds, UI actions, commits or publication. The native window captures are 1442×932 including Windows chrome, with Process zoom 100%; this review does not generalize to other display sizes.

Capture paths below are members of native/evidence.zip. Only link formatting was adapted for this portable copy; original review SHA-256: bbe98aa59fa9000cb612843dd2a6f5e33692a19512de32cd0baff5ff059b3a69.

Result

No new selection, primary, qualified-return, group-order or saved-state blocker found in the reviewed evidence. The run-1 Inspector mismatch is visibly corrected. There remains a concrete routing readability problem after reorder, described separately below; do not call routing fully accepted.

Visible evidence

12 first qualified (runs/run-3/raw/12-first-qualified-0.jpg) selects Shared detail inside First use. 13 nonprimary menu (runs/run-3/raw/13-nonprimary-menu-0.jpg) addresses Shared detail inside Second use while First use remains selected. 14 definition (runs/run-3/raw/14-nonprimary-definition-0.jpg) shows the local Shared detail editor with alias reused_leaf.
16 exact Back (runs/run-3/raw/16-exact-qualified-back-0.jpg) and 17 (runs/run-3/raw/17-caller-tree-expanded-0.jpg) return to the Second use's inner Shared detail tile, with the exact Second use path in Details, Shared detail as Properties subject and a read-only Inspector. The outer card is not simultaneously highlighted. In contrast, run-1 frame 26 (runs/run-1/raw/26-back-to-caller-0.jpg) shows both inner/outer outlines and the editable Second use Inspector despite qualified Details.
18 explicit outer (runs/run-3/raw/18-explicit-outer-properties-0.jpg) deliberately selects Second use: outline, Properties label and editable Inspector name/alias reference_two agree. 19–21 (runs/run-3/raw/21-current-back-readonly-0.jpg) show current-qualified Open and Back again returning to read-only Shared detail. 22 same-leaf Explorer navigation (runs/run-3/raw/22-explorer-same-leaf-definition-0.jpg) changes to Reusable request detail and restores the editable local reused_leaf Inspector. Card bounds/100% scale are stable across equivalent caller states; no visual camera jump is apparent.
29 mixed selection (runs/run-3/raw/29-mixed-selected-0.jpg) displays two selected, Card A as member and A→U link as primary, with the link Inspector. 30 refusal (runs/run-3/raw/30-mixed-refused-0.jpg) retains those identities/positions and presents the readable status: “Reorder requires local contribution cards only; links cannot move with them.” Saved frame 31 preserves that display.
32 C (runs/run-3/raw/32-card-c-0.jpg), 33 Add A (runs/run-3/raw/33-card-a-menu-0.jpg), 34 primary A (runs/run-3/raw/34-group-primary-a-0.jpg), 35 Inspect C (runs/run-3/raw/35-group-menu-c-0.jpg) and 36 primary C (runs/run-3/raw/36-group-primary-c-0.jpg) provide consistent membership/count and primary feedback. Inspect changes the primary to C without dropping A. The menu fits and its unavailable Hide/group-delete explanations are visible.
37 selected anchor (runs/run-3/raw/37-selected-anchor-0.jpg) / 38 saved (runs/run-3/raw/38-anchor-saved-0.jpg) retain A/B/C order and primary C. 39 reordered (runs/run-3/raw/39-group-reordered-0.jpg) / 40 saved (runs/run-3/raw/40-reordered-saved-0.jpg) show B/A/C, U/V still in their original right-column slots, count two and primary C. This is consistent with source-order A,C despite selection C then A. 41 Undo (runs/run-3/raw/41-one-undo-0.jpg) returns A/B/C; 43 Redo (runs/run-3/raw/43-one-redo-0.jpg) restores B/A/C; primary C/count stay visible. Actions themselves were observed by the parent, not by this reviewer.
49 reopen (runs/run-3/raw/49-reopened-0.jpg), 50 Behavior (runs/run-3/raw/50-reopened-behavior-0.jpg), 51 order (runs/run-3/raw/51-reopened-order-0.jpg) and 52 saved (runs/run-3/raw/52-reopened-saved-0.jpg) show the reopened project and B/A/C order with zero selected. The process Inspector agrees with the newly selected process. The Explorer tooltip temporarily covers part of Card A's title in 51/52; this is transient pointer feedback, not missing content.

Concrete remaining visual issue

In reordered frames 39/40/43/44/51/52, the two orthogonal A→U and B→V routes overlap along their shared middle x-coordinate and horizontal segments. Portions of the central vertical path appear interrupted, and the resulting picture can be read as B→U and A→V without carefully tracing the bends. Card titles and primary outlines remain readable, but endpoint connectivity is not reliably legible from this drawing. This is consistent with the parent's existing routing issue #32; I have not independently established its source-level attribution. The complete-Project checker establishes unchanged link records separately, so this should not be reported as group reassignment or source corruption, nor hidden by a blanket “visual pass.”

Receipt cross-check, independent of screenshots

Read history-bytes.json (runs/run-3/raw/history-bytes.json), all eight snapshots/*/result.json and checker stdout/stderr, plus the actual ten saved source files for each snapshot. Independently recomputed SHA-256/size pins and confirmed:

All eight pinned checker receipts/logs/source maps match actual bytes; each checker reports PASS for complete Project equality against literal expected order and exits zero with unchanged baseline/checker/snapshot flags.
Baseline = undo = qualified-navigation = mixed-refusal = anchor-noop across all ten source files.
Reordered = redo = reopened across all ten source files.
Only oa/behavior/process_oa_process.arcadia differs between baseline and reordered.
finished.json (runs/run-3/raw/finished.json) records normal exit 0 at 2026-09-23T13:43:54.209222+00:00; both actual application logs are empty and match their recorded pins. GUI/checker/fixture unchanged flags are true.

The launch receipt records matching GUI/checker input-map pin f3fd180af294e0797f9dfdbff647274de30a907b685a4fe1a2b58027bfe7da95; the parent identifies this run with b743a10. This review did not independently repeat commit-to-build source identity proof or rerun the checker executable.

Limits

Screenshots establish visible states, not exact model equality, action counts or Undo stack depth. Those conclusions rely respectively on the checker/source readbacks and the parent's observed single Undo/Redo actions. Frames 37/39 are post-action states; they do not establish native mid-drag insertion-line/count feedback (separate WGPU evidence exists). No new evidence here for SA native interaction, keyboard Shift behavior, dirty/stale guards, Escape/out-and-back, other DPI, accessibility contrast, performance or routing correctness beyond the concrete issue above.

6.9 KiB Raw Blame History Unescape Escape