4MKB: a picked row for a Lua function bind calls the function through the registry ref Hyprland reports, only when one reported bind, and no other, matches the row and Hyprland did not reload its Lua state after the lookup. It is new on the mirror (commit 19d55809, proof work, not upstream). Its 7 function-level matches are the functions the PR adds plus dispatch_binding. It satisfies SYS-REQ-260922-6642. The requirement is in review status, not approved.

Line 735: __lua) [[ -n $arg ]] && dispatch_lua_function_binding "$arg". Before this change a __lua row with no cached expression reached the "") return 1 arm with an empty dispatcher. 0W96 says Lua binds "remain dispatchable", and before this change function binds were not. The PR closes that gap for binds whose ref Hyprland reports.

9DMS is code_changed because bin/omarchy-menu-keybindings and the test file changed. In dynamic_bindings, the PR keeps the code:$keycode derivation and stores the raw key separately (raw_key, line 301) for the identity. My reading: the displayed key for code:N binds goes through the same lines as at base. Its trace was re-reviewed on the mirror (proof work).

On the evaluated tree I extracted the PR's functions the way its test does and drove them with its stub_hyprctl_dispatch (macOS, luajit as lua). Results: one matching bind with ref 264 gives exit 0 and calls 264. Two binds with the same identity, a non-numeric ref, a reload between lookup and call, and the same chord and description in another submap each give exit 1, call nothing, and leave no mark. The picked-row awk: two identical exec rows dispatch the first, as before. An exec row and a __lua row that read the same give no record. A label containing \t matches through ENVIRON but not through the base's awk -v. In our Linux runs, keybindings-menu-test.sh passed 33/33 on the PR head and on its merge onto quattro 035ce29f, and 16/16 on quattro. These results are outside the package.

I reproduced the engine's witness-name scan on both versions of the test file. A bash function name() { matches none of its patterns. At base nothing matches, so the witness falls back to <file>:<line>. At the head, the first match is the generic "typed declaration" pattern (<type> <name>(), which reads if type( at line 333 as a declaration named type. The keyword filter checks only the captured name, not if. Both compat findings follow from this. The fix belongs in the engine (skip heredoc bodies in shell files, or use the shell function pattern), not in this PR. This is my reading of the engine, not a package fact.

test_diff reports that no existing test file was modified or deleted. Upstream e03ad1ea adds 323 lines to keybindings-menu-test.sh and removes none. On the mirror our proof work changed existing test code: the omarchy-menu-select stub moved into use_pick_file_picker(), which run_keybindings re-installs on every run (pick_from_menu overwrites it). It also extended the Verifies header to 4MKB. No existing assertion changed.

At base the picked row was found with awk -v selection="$selection" '$1 == selection'. awk -v processes escape sequences, so a label containing \ matched no row, for every bind kind. At the head the selection is matched through ENVIRON (line 758), which is literal. The PR states the intent: "The picked row is now matched through the environment instead of awk -v, which reads a backslash in a description as an escape and then matches no row." The test at line 598 pins it. review.behaviorDiff.inventory is empty, so this is my inventory, not the engine's.

At base, two rows with the same text dispatched the first ({ print; exit }). At the head (line 762) a single match is dispatched. Two or more matches are dispatched (first row) only when none of them is __lua; otherwise nothing is printed and the script exits 1 with no message. The PR states this: "When two rows read the same and one of them is a function bind, the pick is refused". Tests at lines 625 and 637 pin it. Our stubbed run of the shipped config found no two rows that read the same.

From the audit run (mcdc_coverage passed): keybindings-menu-test.sh witnesses 4MKB rows 1, 3, 4 and 5 (meets_policy true). Our proof work added the # MCDC markers (lines 393, 418, 456, 578) and the 3 new scenarios: undecodable identity, refused mark, and menu closed without a pick. The two remaining rows are mcdc:ignore:defensive (test lines 5-6, REVIEW-261005-THHB): a called function with the mark gone, and no call with one match and the mark present. Each needs the guard or the hand-off removed. These are requirement-level witnesses, not changed-line coverage.

testResults is ingested and current (68 passed, 0 failed); 1 tagged test, joined. Our proof work (97e2bbf3) added obligation tags to keybindings-menu-test.sh: 9DMS error_handling nominal (line 181) and negative (line 202), and 4MKB error_handling nominal (line 394) and negative (lines 534, 566). The junit report records the whole file as one test, so each of those cells rests on that file's single pass (tests: 1), not on the individual scenarios. 9DMS external_call_timeout_bounded is not_linked: no test is tagged with that class, so it is not shown as tested. 0W96 error_handling is suppressed by a recorded suppression in the base data. The earlier test_results_not_joined disclosure is gone at this package.

called() asserts which registry ref the stub recorded as called. It is added by the upstream PR. It has no behaviour of its own beyond the scenarios that call it, and the file's # Verifies: header already ties those scenarios to 0W96, 9DMS and 4MKB. No requirement is missing. A function-level trace would add nothing.

called_nothing() asserts the stub recorded no call. It is added by the upstream PR. It has no behaviour of its own beyond the scenarios that call it, and the file's # Verifies: header already ties those scenarios to 0W96, 9DMS and 4MKB. No requirement is missing. A function-level trace would add nothing.

left_no_mark() asserts no mark is left in the stub's Lua state. It is added by the upstream PR. It has no behaviour of its own beyond the scenarios that call it, and the file's # Verifies: header already ties those scenarios to 0W96, 9DMS and 4MKB. No requirement is missing. A function-level trace would add nothing.

lua_function_bind() prints one hyprctl binds record in the 0.56.2 function-bind shape. It is added by the upstream PR. It has no behaviour of its own beyond the scenarios that call it, and the file's # Verifies: header already ties those scenarios to 0W96, 9DMS and 4MKB. No requirement is missing. A function-level trace would add nothing.

pick_from_menu() installs a picker that returns one row and runs the real script. It is added by the upstream PR. It has no behaviour of its own beyond the scenarios that call it, and the file's # Verifies: header already ties those scenarios to 0W96, 9DMS and 4MKB. No requirement is missing. A function-level trace would add nothing.

stub_hyprctl_dispatch() installs a hyprctl stub that runs the dispatched Lua and records calls and marks. It is added by the upstream PR. It has no behaviour of its own beyond the scenarios that call it, and the file's # Verifies: header already ties those scenarios to 0W96, 9DMS and 4MKB. No requirement is missing. A function-level trace would add nothing.

use_pick_file_picker() re-installs the file-driven picker stub before each run. It is added by our proof work (it wraps the existing picker stub). It has no behaviour of its own beyond the scenarios that call it, and the file's # Verifies: header already ties those scenarios to 0W96, 9DMS and 4MKB. No requirement is missing. A function-level trace would add nothing.

pocs/crs-260930-jvx8-c02.sh is the reproducer of KI-MENU-KEYBINDINGS-FUNC-BIND-NO-DISPATCH, not part of the upstream PR. Commit dc773390 makes its stub print arg: 264 for the Select all bind, the shape Hyprland 0.56.2 prints. With the old empty arg: a fix that dispatches through the ref still read as broken. Its hyprctl() stub is untraced and the file is out of scope (changed_files_out_of_scope). It realizes the KI's check on 0W96, so no product requirement is missing.

80236345 sets the KI's status to fixed by omacom/omarchy#11851. 49ebebcf re-captures its evidence as not reproduced. In our Linux runs the fixed reproducer exits 0 (defect present) on quattro-proof 1e433732 and exits 1 (fixed) on the PR head and its merge.

The PR text lists Zoom in, Reset zoom and the three universal clipboard actions. The base also ships Select all (default/hypr/bindings/clipboard.lua:45, send_shortcut_once("CTRL", "A")), a Lua function bind next to them. The fix is generic, so it covers that bind too; the reproducer uses it.

review.effects is absent for this bash change, so no new_call_path, effect_gain or new_mutable_state claim can be checked, and the scope analysis could not detect uncodified connections. From reading: the new callees are dispatch_lua_expression (existing; runs hyprctl dispatch) and hyprctl binds, base64 and awk. The only product caller of dispatch_binding is the script's own main block (line 767). These facts come from reading, not from the engine.