# Elevated platform firing fix — 15 October 2026 Intent: a visible, exposed player remains a valid ranged threat even when their platform cannot be reached. Preserve cover, reaction, paid recovery and immutable shots. Live diagnosis on `2a40e43` plus pre-existing dirty director assets/docs (untouched): Unity 6000.6.4f1, PlayTestScene manual one-shooter sandbox; compiled/imported, idle test runner, no Console errors. Player `(382.411,4.080,396.522)`, enemy around `(400.828,0,393.434)`. Personal sight/reaction verified; only 22 released shots while planning repeatedly moves about three metres forward/back. Eye-selected +1.35 m point and +0.9 m muzzle spheres hit `Disconnected island (1)` first; +1.6 m reaches the player first with the same .08 m clearance. Seeing the gun does not imply the chosen lower body point clears the intervening platform lip. Preserved runtime reproduction boxes, under the temporary sandbox terrain: `Disconnected island (1)` position `(382.5,3.5,396.5)`, scale `(10,1,10)`; `Disconnected island (2)` position `(384,1,396.5)`, scale `(10,1,10)`. Both have identity rotation and unit centred BoxCollider. No saved terrain/nav changes. Hypothesis: first eye-visible point need not be muzzle-clear; rejecting only that point also makes the current firing hold appear useless. A shared bounded eye/muzzle point should permit repeated stationary releases. Contradiction: unchanged elevated geometry still oscillates or cannot release despite a common clear point. Scope/owner: lead writes targeting helper, release verification, current-hold lane and native regressions. Eye-only sampled memories, hypothetical travel lanes, all authority, request/lifetime/status/cooldown/commitment gates and saved assets remain unchanged. No new pursuit policy, permanent level changes, rebakes or dependency changes. Medium risk: bounded targeting/holding behavior within existing validated execution; no authority/lifecycle ownership changes. Independent technical and gameplay pre-reviews support the approach; both require a common clear point, immutable commitment and real current-lane hold. Design/UX adds little independent scope to this narrow repair. Ponytail inspection belongs in technical post-review. Checks: first RED captured elevated-platform native graph regression; focused GREEN including no-common-sample refusal, immutable blocked commitment and preserved memory. Then affected full shooter/perception/navigation/retrospective native fixtures and rules; final integrated scope and matching peer checks selected from actual diff. Existing gates remain open until actually executed; no historical proof labelled fresh. Duration unknown (Editor native runs/builds dominate). Escalate if authority/lifecycle or unseen target semantics change; stop after bounded proof and human checkpoint. Human timing: user chose **ASAP**. Present first safe compiled, focused-verified manual checkpoint with reconstructed temporary decks and original player placement; pause feedback-dependent closure and further Editor runs while the user tests. ## Implementation and focused evidence Implemented shared bounded firing-point selection in EnemyTargetingRules and a read-only personal firing verifier in EnemyPerception. Bolt preparation/commitment uses a common eye-ray/muzzle-sphere point. The current visible hold uses that real firing check; hypothetical travel destinations keep their existing checks. No authority/payment/committed-lane/lifetime changes, assets or navigation rebakes. Independent technical and gameplay post-reviews found no source blockers, including Ponytail scope/completeness inspection. A later technical delta review supported the time-correct fixture assertions and cautioned that the cover miss below remains open. All reports below are immutable `TestResults/foundation-runs/