# Tracer revision implementation and validation 9 October 2026. TaskT continuation of the pack AI and roster plan. Implementation is accepted within the automated/Editor/network evidence below; baseline was364f41e (1057 EditMode/65 PlayMode passes). Human and representative-hardware gates remain separate. **Current contract correction — 9 October:** The user revoked terrain/obstacle pass-through. Tracers must stop at the nearest solid terrain/cover/prop or eligible living player, with no damage through an intervening obstacle. Restore world absorption and actual obstacle impact. Prior player-only/pass-through decisions and passing fixtures below are historical and superseded. Damage6, bounded wave, protocol11 and clock/particle repairs remain required. Corrected collision tests, live fixtures and fresh builds now pass; see the final integration checkpoint below. ## Reviewed contract and ownership Three independent technical/gameplay/UX perspectives agree on the bounded ordinary-tracer scope. Saved Ranged Drone base damage and its rebuild recipe become6 instead of10. Normal.7 and elite1.5 outgoing scaling remain separate, giving4.2/9 before defenses at stage one. Melee, bosses, bombardier shells, player weapons and future Lancer precision shots are excluded. Initial personal sight/facing/muzzle clearance and attack/grace rules stay intact; paid launched tracers survive source death and stop on the first solid obstacle or eligible player. The host captures seed, wave settings, damage, lifetime, speed/turn and tracking identity at launch. Proposed8-degree2Hz sine and2-degree7Hz smooth seeded jitter form one desired heading before the existing total steering limiter. Normal8m/s and60degrees/s remain fixed; requested angular amplitudes are clipped when necessary rather than adding a second turn budget or positional offsets. Ruling: launch-anchored120Hz simulation processes the captured final lifetime interval before expiry, with a finite maximum960 full steps for the existing8s lifetime ceiling. Partition determinism applies to equal sampled target timelines; moving targets sampled at different times cannot imply identical historical trajectories. No elapsed time is silently dropped or flight restarted. If incorrect, a hitch could skip a player or extend effective life; pure/native prefix and expiry tests are required. The existing shell/world SweepSegment remains unchanged. Corrected ordinary tracer queries include solid world geometry and scoped living players, then select the nearest valid contact with initial-overlap/child-hit handling. Saturated queries require a same-mask correctness fallback; exceptional allocations/cost must be recorded. A larger fixed buffer alone does not establish nearest-contact correctness. Ruling: clients receive the host path through existing correction payloads; no hidden target/guidance data or new wave RPC payload.30Hz corrections and a small fixed presentation history require measured delayed/lossy interpolation and bounded extrapolation. Missing ordinary-tracer visuals can adopt current correction state for late join, with terminal-generation fences preventing ended-flight resurrection. Shell/boss historical late-join behavior is not claimed as improved. If insufficient, remote waviness/readability remains a failed acceptance gate rather than assumed seed reconstruction. One implementer owns bounded tracer runtime/rules/settings, new rule/native tests, the saved-definition authoring helper/rebuild recipe and tracer-specific old validation expectations. Lead owns actual asset operations, builds/network/render evidence and documentation. No source writes during testing/building. Preserve manual terrain/scenes/GUIDs, pins, jump0 and preexisting ProBuilder settings. ## Evidence ledger Focused compiling RED tests are being authored. Runtime and asset authoring have not yet changed. Require actual base/scaling/excluded-family proof, deterministic bounded motion/final lifetime sweep, saturated/overlap/nearest-player collision, pool/reuse/stale messages/source-target lifecycle, before-launch refusal versus after-launch pass-through, host/remote rendered flow and128-slot pressure. Historical world absorption is not evidence for the revised policy. - `tracer-rules-red`:1 pass/16 failures. Existing correction RPC payload signature passes; failures prove saved base10, missing family/wave contracts and unsafe invalid steering inputs. - `tracer-native-red`:0 passes/12 failures, with actual hosted fixture setup accepted. Missing separate tracer sweep/fixed producer/captured settings and visual adoption/terminal fences are the intended failures. Runtime implementation is now authorized; actual asset mutation awaits the supported helper. - Pinned compilation initially rejected a new obsolete GetInstanceID call; overlap ties now use stable player NetworkObjectId. No Editor/package upgrade was made. - Supported authoring `tracer-authoring-green.json` preserves Ranged GUIDa2ad0a9dff069fd4bafc082f0777516d and resolves base6/normal4.2/elite9. Saving also makes its existing default coverGraceSeconds5 explicit; no gameplay meaning changes. `tracer-rules-green` passes17. - First native candidate passes10/fails5. Native diagnostic `tracer-player-mask-diagnostic` proves actual enabled owner CharacterController is Players8, with one raw overlap/hit; CombatTargets9 sees zero. The original pre-review assumption that player hulls were on CombatTargets was incorrect. The sweep now queries explicit Players|CombatTargets for normal and saturated fallbacks, preserving world/trigger/scoped-player exclusions. Final-interval damage and nearest/overlap/saturation retests remain pending; clocks are not changed to mask this defect. Ruling: the unchanged correction payload has no projectile-family marker. Its shared non-shell visual factory/history can therefore also adopt/smooth a boss's early correction; this is incidental presentation behavior, not a boss weapon revision. Boss damage/physics/collision/tracking/correction cadence remain unchanged, as do shells. Do not infer a weapon family from velocity or claim exact wave reconstruction. This avoids a wire/schema change while preserving the explicit physics exclusions; actual remote presentation is the proof gate. The no-schema ruling above is superseded by integrated review evidence: excluded boss corrections end early, so bounded extrapolation freezes their surviving remote flight; delayed tracer motion also cannot show an immediate untimestamped terminal impact coherently. The bounded correction adds explicit tracer-family/captured-expiry metadata to spawn/correction and authoritative simulation end time to terminal messages. Protocol11 is required, with the existing canonical build guard and a native source-prefab-only update. Client tracer terminal presentation follows its delayed timeline; excluded nontracers retain legacy coast and shells retain their existing path. No target/guidance positions or wave seed/settings are replicated. The earlier ambiguity is not solved by inferring family from speed/cadence or changing boss physics/rates. - `tracer-native-torso-green`:15/15 native passes; actual raw player-hull prerequisites prevent the feet-height fixture misses and vacuous dead/foreign-scene assertions. `tracer-full-editmode`:1074/1074 before presentation metadata correction. - Metadata-correction RED: `tracer-presentation-metadata-rules-red`16pass/2missing-signature failures; `tracer-protocol-eleven-red`10pass/2protocol10 failures; `tracer-presentation-native-red`13pass/9contract failures, including the actual excluded visual coast28m gap. The new terminal/client branch tests are isolated render contracts, not actual remote-network proof. Source schema implementation and protocol11 guard/native prefab update now proceed; stale signature expectations are revised deliberately with the approved compatibility boundary. - Metadata-correction GREEN: `tracer-protocol-eleven-green`12/12, `tracer-metadata-rules-green`18/18 and `tracer-metadata-native-green`22/22. Fresh full suites: `tracer-protocol11-full-editmode`1077/1077 and `tracer-protocol11-full-playmode`87/87, zero failures/skips. These include actual native collision and isolated production presentation branches; actual remote rendering and dense performance are still pending. - Supported protocol update changes only saved application-root ProtocolVersion10→11; MainMenu inherits it without an override. `tracer-protocol-source-before.json`/`after.json` preserve recorded configuration and GUID38b57d87ab30cbf44bfb5bbd09cd7496. Existing protocol10 builds are obsolete and must be rebuilt before network acceptance. - Live-proof mini-plan approved after full suites finished: bounded opt-in development-only read-only service snapshots and local probe observer, owned offscreen actual-camera frames, genuine AI exposed-launch-then-cover/hidden-before-launch scenarios, natural Update128-slot timing and separately labelled saturated fallback costs. No debug server RPC, hidden target replication, clock injection or per-frame file I/O. The lead owns build/network orchestration; the sole implementer owns helper/source hooks. - Integrated hook reviews find no new material production defect. Evidence boundaries: inserted cover is host-local, so genuine-AI host crossing and correlated client playback are distinct proofs; scene renders from the actual player pose use a clone, omit HUD and are not exact final gameplay frames. A separate direct-service normal/elite collision control through the saved arena wall is planned to show the same world obstacle on both peers. Latejoin's15s host observer may finish before its45s join window; prejoin generation/launch records and fresh actual client adoption/expiry evidence are required. Performance samples deduplicate the service simulation-frame counter and describe that section's cost, not whole-frame/native-memory performance. - Live hooks compile on the pinned Editor. Fresh post-hook full suites `tracer-live-hooks-full-editmode`1077/1077 and `tracer-live-hooks-full-playmode`87/87 pass. Subsequent helper-only controls add direct shared-wall `cover-normal`/`cover-elite`, reject non-finite local observer durations and count unique service simulation frames. Final compilation passes; no production authority/physics/schema changed in that helper follow-up. - Fresh protocol11 Windows builds succeed: prototype0errors/569warnings (52.019s), debug0/640 (14.710s), release0/706 (39.380s). Release packed-asset guard passes. Full retained CLI output is `TestResults/tracer-{normal,debug,release}-build-cli.json`. Graphical host/client collection is running; these builds do not by themselves establish gameplay or rendered acceptance. - First graphical cohort `tracer-peers-3fa49a22c28f40be96860ad83cae4890` joined two actual delayed/lossy owners, but the helper never executed: CLI could not resolve its type. Diagnosis proves file/meta existed on the real Editor filesystem but no MonoScript/type was imported. The new helper meta GUID had33hex digits; all other new GUIDs had32. The lead repaired only that never-imported new GUID to32hex and invoked supported ImportAsset. Earlier compiler checks therefore covered imported runtime/observer code, not this new Editor helper; normal/debug/release player builds remain valid because the helper is Editor-only. The harness now resolves the helper/imported GUID during preflight before opening any session. Failure and cleanup-operation-check error are retained; actual host leave and original clean Verdant setup restoration succeeded. Actual flight acceptance remains unrun. - Corrected import verified `asset=True`, class `ProjectI.Editor.EnemyTracerLiveValidation`, idle/no compile failures. Graphical cohort `tracer-peers-cff9366199a04d37aa7bdd04a382d692` completes all8 host phases and collection/cleanup, buildSHA256437691663C5FB21231524F4886E60A10B1841F13E42FBBC93EE39FBD827930CD. Genuine normal/elite cover-after-launch each fire once/contact once/apply4.2/9; shared saved-wall direct controls each contact once; hidden-before-launch, actual source despawn/target down, capacity/expiry/reuse/cancel and saturated queries pass. Collection success is explicitly not visual acceptance. - **Actual first-normal visual RED:** generation1 client has25 observations but only one path sample (timestamp16.061), render time freezes16.094 and positionz-26.94347 until fixture cancel, with no pending end/sparks. Host first camera capture causes1.2s real hitch: launch16.061, next sample17.267 after real contact/end16.9409. Client first capture also stalls about1s; its sampled server clock advances15.956→16.956. Shared-cover normal/elite otherwise show actual passage and queued terminal/particles, while ordinary elite's delayed terminal and wide-view subtle curvature remain review limits. Three perspectives agree first-normal blocks presentation acceptance; do not hide it by warming captures alone. - Pinned NGO source processes transport/RPC queues in EarlyUpdate (`NetworkManager.cs351–364`) and advances NetworkTimeSystem in PreUpdate (`411–414`). This supports the stale sampled-clock/future-time rejection hypothesis but does not prove receipt versus non-delivery. Approved next step is bounded development-only receipt diagnostics recording kind/generation/time/Now/realtime/frame-delta/reason/fences, without changing any accept/reject behavior. Only a reproduced rejection authorizes a timestamp/receive-phase repair. - Actual128-flight natural-update simulation sample:226 unique service frames over0.8s, mean0.22345ms/max2.6606ms,46110 segments/11271 corrections across lifecycle, zero saturated sweeps. Scope is this machine and simulation section, excluding presentation/capture; it is not whole-frame/hardware acceptance. Eight exceptional saturated queries take1.629ms. Reported allocation counter0 is **unavailable**, not zero allocation: a known managed16KiB calibration also returns0 (`tracer-allocation-calibration.json`). SphereCastAll's allocating fallback remains explicit; updated reports will use availability=false/bytes=-1. - Diagnostic-only guards compile and preserve focused18/22 passes. Debug receipt build succeeds0errors/572warnings (48.630s). Instrumented first-shot cohort `tracer-peers-7eb0796188144e709d47935ca5f8d76e` does **not** reproduce the earlier capture hitch: End gen1 is accepted/QueuedTerminal at packet17.74666, sampledNow17.66611, frame delta0.00694; client has40 samples, eight-sample history and six particles, with no receipt gaps. This is a positive replay, not proof of the original cause. The next explicitly opted-in local client-stall probe waits for a real tracer then blocks that owned main thread once for nominal1.25s (measured≥1.2s), within strict1.5s/10s-arm bounds. It changes client frame timing deliberately, without altering clocks, actors, packets or server simulation; the observer remains passive. Runtime timestamp acceptance is still unchanged until the stall receipts establish causation. - Stall probe compiles; fresh debug0errors/73warnings (24.907s). `tracer-peers-441314b623e34517b24ce4af3995686e` at75/15/1 executes1.2502858s real stall, gen1 frame695, clock17.530849 frozen during sleep, resumes696 at18.87585. End arrives701 and is accepted CompletedImpact (packet18.44047/Now18.94146), six particles. This does not reproduce a rejection and is retained as a separate positive stall condition. - **Causation proved under diagnostic0/0/0:** `tracer-peers-fe1e4fd5ee5d44aabd2abbd93862f2c3` executes1.2507988s client stall, gen1 frame981/Now16.1777826 frozen. In EarlyUpdate982,13 legitimate corrections (16.68725–17.08725) and End17.100825 are rejected `FutureTimestamp` against stale16.1777826+.5. Receipt gaps/truncation0, no client sparks. Native pinned ordering plus actual received-message reasons establish this defect; the original uninstrumented failed shot remains separately retained, not retrospectively asserted to have an identical receipt trace. - The0/0/0 host gameplay case completes, but helper cleanup times out restoring an actual player pose under the unchanged4s settlement gate. Outer client/host leave and clean original scene/build-table restoration succeed. Do not label the cohort fully passing or increase its deadline. Next helper instrumentation records exact owner/requested/accepted/body/lifetime/warp-phase/tick state at cleanup request and failure; production Warp changes require independent evidence. - Ruling after three independent perspectives: defer all three client presentation message types to a fixed per-slot Spawn/newest-Correct/End envelope flushed in service Update after NGO's PreUpdate time refresh. Keep host simulation/damage/presentation immediate, original payload/captured timestamps/expiry and strict future guard at the fresh clock. Bound storage/one-frame processing; discard still-invalid messages; fence queued terminal/cancel/reuse/expired traffic, with no target data/new RPC/clock slack/lifetime restart or unbounded extrapolation. Implementer mini-plan and new behavioral RED precede this repair. Actual0/75 stall replays, warm closer motion/terminal views and latejoin/full lifecycle remain acceptance gates. - Queue design refinement bounds storage at768 value records (128slots×3message kinds×2candidates). A protected currently validated generation cannot lose its terminal to an unvalidated higher-generation future packet. Retained records flush by receipt order, with newest same-generation correction; intermediate unrendered generations may coalesce during a severe hitch. This is bounded playout, not complete packet/history replay. - `tracer-clock-queue-red` compiles and runs **22 existing native passes /10 new failures**. The stale-clock case fails because the core remains active instead of reaching its terminal; valid current terminal followed by invalid higher future data also fails; higher valid unknown terminal leaves the old visual. Remaining cases specify bounded staging, cancellation/session clearing, expired/future rejection,128-slot coalescing and excluded-family behavior. Runtime implementation and cleanup-only pose snapshots are now authorized; no production Warp change or deadline relaxation is authorized by this evidence. - Base queue implementation compiles and passes `tracer-clock-queue-rules-green`18/18 and `tracer-clock-queue-native-green`32/32; debug0errors/73warnings (25.660s). A fresh-context final technical review withholds approval for one P2: same-generation first-End/newest-Correct coalescing can discard a legitimate message before fresh-clock/full metadata validation. Active gen10 End+20 followed by legitimate End+.8 loses the latter; mismatched family/expiry or an invalid newer future correction can similarly replace valid traffic. Lead source inspection confirms the gap. - Approved refinement keeps the768 bound while retaining incompatible same-generation alternatives and a valid current fallback. Exact Spawn/End duplicates may coalesce; ordinary matching corrections retain newest safe data, with an unvalidated future candidate held separately. Receipt-clock classification is priority only, never acceptance; strict fresh-clock validation remains final. Six additional behavioral tests cover invalid End before valid (including future time within captured life), wrong family, wrong expiry, future correction and unknown legacy future Spawn before valid same-generation Spawn. Only test authoring is released until its RED run; no changes to host simulation, wire/schema or settlement deadline. - Same-generation refinement observes `tracer-same-generation-red`32passes/6failures, then `tracer-same-generation-green`38/38. Fresh technical recheck closes P2; fixed bound, strict guards and host path remain intact. Refined debug succeeds0errors/73warnings (32.370s). - Actual postqueue0/0/0 `tracer-peers-c1367651ffde4713a31eb8fd158ada13` accepts legitimate deferred correction and End at fresh29.95696 versus old29.20813 (End30.19684), with no receipt gaps. Its fixed.75s fixture cleanup precedes buffered terminal display. The independent movement restore still fails owner1: seq2/lifetime41 body reaches(4,.08494,-4), host accepted(4,.1,-4), but PublishedTick/EligibleTick0 and phaseFailed after4s; owner0 restores. Do not infer source from body position alone. - Actual required75/15/1 `tracer-peers-8bcfe3ee86854bf2b539ddfcfd77d34f` **passes** gameplay/cleanup and presents six particles. Old receipt clock17.96386 would reject packet18.6563; evaluation18.95694 accepts it after refresh and completes impact. This establishes the repair under the configured delayed/lossy stall condition. - Optional fixture peer ACK now freezes existing gameplay assertions before waiting≤3s for actual client particles at exact generation/slot/endpoint, through unique local request/ACK files. It changes no production timing or4s movement deadline. `-NoPeerTerminalAck` preserves the original cleanup ordering for diagnosis. Observer latches only actual `SparkParticles>0`; End receipt or PendingEnd cannot acknowledge. Debug build0/73 (26.661s). - Peer-ACK0/0/0 `tracer-peers-1176ddafac1a49ea831e7e297ebe9c4b` **fails the visual gate**, while later cleanup passes: End18.89013 is accepted/queued at18.95695, then core clears and ImpactUntil20.44758 becomes positive, but all eight samples/latched status show zero particles. Clock jumps18.95695→20.33252 across0.036s local monotonic time. This is no longer assumed to be fixture cancellation; approved diagnostic-only immediate-after-Emit/first-LateUpdate/next-Update particle count/time/delta records will distinguish automatic aging from missed emission/cancellation before a behavior fix. - Warp source investigation disproves the LocalTime-vs-ServerTime and half-float hypotheses: pinned NGO2.13.3 stamps explicit and regular native publication with ServerTime, and the saved player prefab has half precision disabled. Client trace proves correction2 applied, but raw publication/owner hook/host rejection outcomes are missing. Three perspectives approve opt-in512 value events per owner/max4 owners (128 would lose a4s30Hz send+decision episode), no stateful getters/extra queries/publication/RPC/allowance changes. `warp-trace-contract-red`0passes/8missing-storage failures precedes instrumentation; no production Warp repair is authorized by this evidence. Root collects both peers before leave using `-TraceWarps -NoPeerTerminalAck` with original1.25s/0/0/0/4s conditions. Human dodge/readability, representative hardware and separate-network acceptance remain distinct from automated, Editor and configured75/15/1 multiplayer evidence. - Diagnostic integration: all three independent technical/gameplay/UX final reviews approve observational-only hooks, with actual correlated evidence still required. `warp-trace-contract-green` passes8/8; fresh `tracer-diagnostics-full-editmode` passes1085/1085. Warp first-event coverage requires both peers enabled before correction, matching epochs and no decisive-event overflow/discard; missing events cannot prove a branch was absent without those checks. - Initial `tracer-diagnostics-full-playmode` aborted in the pinned framework (`PlayModeRunTask.cs:52`, destroyed controller) without a result; retained Editor log is not a test pass. Retry starts from an empty unsaved test scene with standard domain/scene reload; original EnterPlayModeOptions enabled/3 and manual Verdant scene are retained for restoration. No source was changed during the run. - Retry framework XML confirms `tracer-diagnostics-full-playmode-retry` **103/103 passed**,0failures/skips,215.644s (07:32:04–07:35:40UTC). Standard domain reload discarded the project callback; the fresh framework-owned XML was copied into TestResults rather than rerunning the passing suite. Original EnterPlayModeOptions enabled/3 and saved Verdant scene were restored before the combined diagnostic debug build. - Combined particle/warp diagnostic debug build succeeds0errors/601warnings in41.679s; this is the first player containing both latest observational integrations. - Actual combined-diagnostic0/0/0 ACK replay `tracer-peers-7891b4a5ffa146f8ade43bfad61f96cd` fails visual ACK, cleanup passes. Measured stall1.250096s. Gen1 emission frame1069 realtime14.522964 has6particles/time0; firstLate sameframe realtime14.524926 has0particles/time.2/stopped, with DeltaTime.333333/UnscaledDelta1.30368, Active true/Cancelled false. NextUpdate1070 remains0. These snapshots establish automatic same-frame aging of newborn particles before rendering, independently of cancellation. Three preimplementation perspectives now review an ordinary-tracer-only local-monotonic/manual-particle-step repair; no production fix released yet. - Original-order `-NoPeerTerminalAck -TraceWarps`0/0/0 replays `d3b13069a3fe40808356f3fc88bf4fee` and `fef375b135634d83b98c635a3e29a056` both pass the unchanged4s restoration and outer cleanup. First control has matching epoch1,0overflow/resets/discards; owner1 sequence2 native teleport599 follows issue516, owner submits RPC, host accepts both same599 native pose/publication and commits. These are successful controls, not proof of the intermittent historical failure cause or authorization for a movement repair. - Reconciled three-perspective decision: ordinary-tracer impact only captures local monotonic birth/last-step/deadline, pauses its six emitted native particles and incrementally Simulates actual post-birth elapsed without restart/fixed-step accumulation. Preserve .1–.18 particle lifetimes/.2 total window; later stalls may expire it, never extend it. Network ImpactUntil remains a labelled nominal diagnostic marker; explicit local fields govern these sparks. Host damage/flight/wire/strict clock guards and boss/shell behavior are unchanged. Local FX continue through timeScale0 for .2 actual seconds. Cancel/reuse/despawn/session clear manual state. Native actual-playerloop aging/clock/lifecycle RED tests are authorized before runtime implementation. - `tracer-particle-age-red` compiles and observes37passes/7failures. Meaningful failures reproduce six newborn particles becoming0 by LateUpdate and a fresh effect hidden by network-clock jump. Remaining failures specify local birth/deadline/manual state and the intentionally migrated tracer-expiry assertion. Runtime ordinary-tracer-only repair is now released to the sole implementer; no test/build/live work overlaps source writing. - Particle repair `tracer-particle-age-green` passes44/44 native cases,0failures/skips. This proves local-age/actual-birth-frame behavior under the still-old collision implementation; it does not accept the newly corrected terrain collision requirement. All source remains frozen pending the collision design gate. - Corrected collision decision after independent technical/gameplay/UX checks: ordinary tracer `SweepTracerContacts` queries all solid layers with identical overlap/cast/fallback masks; classify enemy and ineligible player shapes before world classification. Return nearest world/player; exact equal-distance contact favors world. Initial world overlap ends immediately; world contact applies no player damage, player contact ends before one damage callback. Preserve shells/player/boss paths, launch permission, captured life/source survival, wave and protocol11. Newly created visual primitive colliders must be disabled before deferred destruction to prevent same-frame self-collision under the restored all-layer query. Add separate labelled open-normal/open-elite direct collision controls alongside genuine AI launch-then-cover and saved-wall absorption. Root harness syntax passes with these phases. Only revised native RED tests are released first. - Corrected collision RED `tracer-world-collision-red` compiles and observes44passes/10failures: real TerrainCollider ignored, normal/elite behind-wall damage4.2/9, initial overlaps/nearest ties/saturated world ignored and same-frame visual collider enabled. Root implements only production collision/service; assigned implementer updates only live/debug/Verdant assertions. All writers freeze before further Editor checks. - Corrected collision GREEN `tracer-world-collision-green` passes54/54 native cases with0failures/skips. Fresh `tracer-world-final-editmode` passes1085/1085. Independent fresh technical plus gameplay and UX final source reviews approve the integrated bounded collision/particle changes, with actual rendered/build/network gates still open. Saturation setup uses the existing open muzzle lane at forward8m,80shapes at.08m spacing; strict all-layer first-player preflight remains mandatory. The saved wall is centred atz-20; an earlier assumption ofz-22 was incorrect and its source comment was corrected without behavior changes. - Fresh full `tracer-world-final-playmode` framework XML passes119/119,0failures/skips,226.432s (08:25:11–08:28:57UTC); copied framework-owned result after the local callback was absent. Fresh full totals1085 EditMode/119 PlayMode. Original Editor options and saved manual Verdant scene are restored before final normal/debug/release builds. - Fresh normal/debug/release builds succeed with0errors and603/640/841warnings,32.926/16.014/43.777s. Release packed-asset exclusion guard passes. Only a subsequent fixture comment correction changes source text; runtime behavior and test/build contracts are identical. - Matching75/15/1 graphical cohort `tracer-peers-767f8484887d4640b3e2eff29a830f93` passes all10phases and cleanup. Open normal/elite apply4.2/9 once; genuine AI launch→inserted host-local cover and saved shared wall each produce WorldContacts1/PlayerContacts0/Damage0. All six hit phases receive actual six-particle endpoint ACKs with0targeted receipt gaps/truncation. Shared-wall client endpointsz-20.5; inserted host-local endpointsz-29.75. Sampled visible cores never cross these planes; wide actual client PNGs show small wall-face sparks. Local player-pose clones omitHUD. New all-solid128 host section:162unique frames,46,110segments,10,759corrections, mean.291574ms/max4.4155ms, no sweep/overlapfallbacks. Clientpressure receipt ringMissing448 means exhaustive128RPCcoverage is unavailable; host capacity/expiry/reuse/cancel and sampled16-slot remote observation remain separate. Saturated80ignoredshape control8fallbackqueries=.8118ms; allocationcalibration0 means measurement unavailable/-1, neverzeroallocationproof. - Corrected0/0/0 real-hitch replay `tracer-peers-68f62156819d4858b9fb2969abebcf96` passes gameplay, actual endpoint particle ACK and unchanged4s restore/outer cleanup. Measured stall1.250476s; emission frame1319 has6paused native particles/time0 atDeltaTime.333333/UnscaledDelta1.303695. FirstLate sameframe still6/time0, nextUpdate still6; fixed localbirth15.797048/deadline15.997048. This directly reverses the earlier measured same-frame6→0 defect. Actual terminal PNGs retained; no injected clock/damage or additional particles. -75/15/1 real-hitch replay `da99ebd13e1e40e6a1067a6f14df8224` passes actual particle ACK and cleanup. WarmClose Repeat2 open-normal `cfdc7a45ca6b4f2c8f97970f840195f5` and open-elite `d83f3c1a8bca43d2bd65c8166f2105f7` both pass; independent rendered review sees retained trails/modest curvature and no frozen/disconnected core. Actual latejoin `40f733eecfbb47f085c482dedf510cd3`:16 client-adopted gens49–64 match prejoin host flights at12.330638