# Issue 21 source implementation handoff

Commit **43c2d51**, branch `codex/nested-call-956atf-20260912`, worktree `.local/nested-call-implementation-20260912`, base `f7d84fa`. Root owns issue 21 coordination, integration, compilation, native validation and release. This contribution changes only `behavior.rs`, `behavior/scenario_activation.rs` and `behavior/scenario_activation/tests.rs`.

Implemented typed call/receiver/scope activation segments, semantic-order overlap columns, exact-correlated Reply source anchors, self-call receive versus send y, and shared final bar/route geometry for hit testing. Square activation fills paint over lifeline guides, under cards/messages, shallow before deep. Large line hit tolerances follow actual activation-body hits so zoomed-out shoulder points do not select a nearby unrelated message. Selected and unselected bars share geometry. No schema, source semantics, endpoint IDs, reply correlation or persisted layout change.

Twelve production Scene regressions construct actual OA/SA Projects with validation and canonical source round trips. Coverage includes both directions; sequential/nested/crossed and permuted UUIDs; self/reentrant calls; three exposed nested columns; exact canvas press/release selection under pan and scales 1/.25/.05; incomplete correlation; ALT/PAR scope separation; inherited outer execution; and fragment-boundary false-nesting falsifiers. Existing direct-canvas ALT/open tests now inspect typed activation records.

Independent source critic `/root/nested_call_design/activation_source_review` found (1) scope-clipped crossed calls appearing nested and (2) ALT sibling return order incorrectly rejecting true nesting. Corrections retain correlation-level reply endpoints and compare only compatible reply paths; critic rechecked both. This is source review, not independent runtime acceptance. The critic disclosed one initial read-only git diff despite its requested no-Git boundary; no Git mutation occurred.

Executed: rustfmt with edition 2024 and skip_children=true on the three owned files; unstaged and staged `git diff --check`; owned-files commit. No compiler, Cargo, desktop/native input, Gitea or root tracked edits. Initial checkout failed on historical long filenames; per-command `git -c core.longpaths=true worktree add` succeeded without changing global settings or overwriting existing work.

Root had previously compiled/run `.local/nested-call-fixtures-20260912.rs` against exact full425 core dependencies: `.local/native-qa/nested-call-f7-core/build.json`, `fixtures.json`/log; output `.local/native-qa/nested-call-fixture-20260912`; 132 elements / 77 completeness warnings. This validates that separate fixture generator, not the new desktop implementation.

Next: root compiles commit 43c2d51 and runs `behavior::scenario_activation::tests`, existing `behavior::direct_canvas_tests` activation/ALT/selection gates and affected workspace/input tests. Verify actual unselected and selected paint, both-direction message edge anchors, loops, shoulder selection, source correlation edits, Undo/Redo and save/reopen with a bound exchange. The fixture intentionally omits exchange references to isolate activation geometry; repeat the user model with actual exchanges/values for native acceptance.

Width 12 and offset 6 are world-pixel style candidates. At scale .05 they become .6/.3 screen pixels: pointer identity tests at that scale do **not** prove raster legibility or practical selection. Root explicitly retains native evaluation at realistic and minimum scale. Incomplete isolated calls keep the existing 26-unit open stub; branch continuations remain explicit open continuations. No general execution/thread metamodel or new LIFO rule is introduced. This work does not satisfy issues 20/22's free geometry or the broader Behavior UX objective.