Performance Opportunity Sweep

Performance Opportunity Sweep

This page ranks source-backed performance opportunities found in the Chernarus source mission. It separates quick low-risk wins from higher-value fixes that are already bundled with security, correctness or gameplay behavior.

For the old BennyBoy WarfareBE comparison that sparked the current FPS discussion, see Old WarfareBE performance comparison. That page separates old-code lessons from A/B test variables so current Wasp changes can be measured rather than guessed.

Ranked Opportunities

Priority	Opportunity	Source evidence	Why it matters	Implementation shape
P0	PVF dispatcher allowlist completion / old-branch backport	Current stable origin/master@0139a346 uses missionNamespace getVariable plus CODE guards at Server_HandlePVF.sqf:14-15 and Client_HandlePVF.sqf:32-33; Init_PublicVariables.sqf:55-61 still has no registered-handler allowlist/logging. Miksuu b8389e74 and perf 0076040f still use Spawn (Call Compile _script).	The per-message recompile is source-present/fixed on current stable, but dispatcher hardening still needs allowlists/logging and old-branch backports. This remains performance-adjacent because the branch-backport removes runtime compile and the current-stable completion keeps lookup bounded to registered handlers.	Use the PVF dispatch playbook: registered namespace/allowlist lookup, keep Spawn, log rejects, and keep DR-55/direct-PV work separate.
P1	Hosted server FPS loop smoke/consolidation	Repo docs branch HEAD@4a1bc13edde6 is source-unchanged from d0161083 / a27086cd and guards both maintained-root publishers at serverFpsGUI.sqf:1 and monitorServerFPS.sqf:1; current stable/B74.1 origin/master@f8a76de34, current B74.2 21b62b04, B69, B74, feat/client-fps@709258e7, perf/asr-ai-server-fps@07609ef4 and historical a96fdda2 guard the single serverFpsGUI.sqf publisher at :4 and have no maintained-root monitorServerFPS.sqf; Miksuu b8389e748243 and perf 0076040f still keep the old branch-only sleep shape. No release/* or origin/dev/july-update-hosted-server-fps-loop-fix head was found on 2026-06-24.	The old busy-loop is no longer a current stable/B74-shaped defect, but Miksuu/perf old-shape refs still need deliberate propagation if they remain targets. Docs-source and stable/B74-shaped refs differ on two-publisher versus single-publisher compatibility, the planned July branch must be created/restored before code work, and Arma smoke remains pending.	Smoke dedicated RHUD FPS updates and hosted/listen no-spin behavior on the target branch first; decide separately whether to keep WFBE_VAR_SERVER_FPS compatibility publishing or consolidate to SERVER_FPS_GUI. See Hosted server FPS loop sleep.
P1	B74.6/B74.9/B75 current-master client RPT/FPS, settings/FPS picker, JIP retry and AICOM cap route	Current master origin/master@c4c9a6479 contains PR #85 from origin/claude/b745@dd1401811, PR #86 from origin/feat/b749-enrollment-bodyfix@2327e5c6a and direct branch origin/VBIED_factory_kill_rewards@756d6918. The broad B745 merge from previous stable f8a76de34 to 34f652ae4 is 226 files / +7611 / -9082; PR #85 adds the B748 settings-menu delta, 9 files / +265 / -7; PR #86 adds B749 enrollment/settings propagation, 14 files / +308 / -21; B75 adds the maintained-root VBIED reward delta, 2 files / +88 / -2; checked ranges are clean under git diff --check. Current master carries RHUD/common helper nil guards, 30 FPS picker and AICOM hard cap/veteran cap in source Chernarus plus maintained Vanilla: RHUD _baseStructures guard, Common_GetSideStructures.sqf:7-10, Common_GetFactories.sqf:7-10, Common_GetClosestEntity.sqf:13, Rsc/Dialogs.hpp:3528, FPSPicker_Open.sqf:48, Init_CommonConstants.sqf:159 and AI_Commander_Teams.sqf:113-114,184. PR #86 also carries B749 JIP body enrollment in both roots (RequestJoin.sqf:95-96, Server_OnPlayerConnected.sqf:32,78-80) and maintained-Vanilla settings (GUI_Menu.sqf:230, Rsc/Dialogs.hpp:1258,1266,3436, Settings_Open.sqf:19,75-79). B75 changes server GUER reward handling, not the checked UI/performance files.	These paths are now current-master static/source evidence, but B74.6 per-frame undefined-variable/RPT-spam claims, B749 JIP claims, B748/B749 settings live-use claims and B75 reward claims still need target mission package/hash and runtime logs. The merge improves maintained Vanilla parity for several UI/performance/AICOM routes, but it does not make the package release-ready or full-modded-fork complete.	Treat as target-package smoke, not a general performance cleanup. Boot source Chernarus and maintained Vanilla early-join/JIP before side structures replicate; exercise RHUD/action availability/factory scans, death/respawn camera paths, FPS picker 30/45/50/60 profile persistence, B748/B749 settings toggles including no A2 allMapMarkers RPT error, B749 body-store enrollment retry, plus low-pop/banked-funds and veteran-slot AICOM founding. Smoke B75 VBIED reward attribution separately from client performance. Compare client/server RPT volume, client FPS and AICOM team count against older B74.5/B74.6/B74.7 heads. Record full-modded fork scope separately.
P1	Old-FPS opportunity pack	Returned scout threads old-vs-current-fps-opportunity-scout and old-be-current-wasp-fps-opportunity-scout-2; reports C:\Users\Steff\Documents\Codex\2026-06-05\wasp-old-vs-current-fps-investigation\outputs\Old-BE-vs-Current-Wasp-FPS-Opportunity-Audit.md and C:\Users\Steff\Documents\Codex\2026-06-05\wasp-old-mission-fps-opportunity-window\outputs\Old-BE-vs-Current-Wasp-FPS-Opportunity-Scout.md; owner pages Old WarfareBE performance comparison, Player AI caps, Testing workflow	The final evidence-backed order is now: AI-source scoreboard, role-aware player-owned AI cap pilot, HC proof pack, low-SV town group branch, static-defense lifecycle/HC accounting, cleaner/object-count evidence, then PV/server-FPS/network cleanups. Old BE is a conservative Takistan Lite OA baseline, not proof that old FSM town loops or bandwidth settings are cheaper.	Treat as a test/config/docs bundle first. Public pilot can cap normal roles near 10, but final design should be a server-published role-aware dynamic budget with buy menu and RHUD parity. Only patch source after the 10/15/20/25/30 test matrix chooses a lane.
P1	Supply mission command-center scan narrowing	Docs/source 15563691 supplyMissionStarted.sqf:25,28,44; current stable origin/master@0139a346 :7,55,61,83; current Miksuu b8389e74 and perf 0076040f :25,28,44	Current stable already uses the typed Base_WarfareBUAVterminal scan with heli/truck radius in both maintained roots, while current Miksuu and perf still scan all object classes inside 80 m and post-filter in SQF. Historical release commits a96fdda2 / 7ff18c49 also carry the heli-aware typed shape, but current origin exposes no live release/* head. Arma smoke remains pending.	Do not reopen the scan on current stable. For old-shape targets, port/adopt the typed scan deliberately; for current stable, smoke truck/heli delivery at command centers and no-completion near unrelated objects. See Supply mission scan narrowing.
P2	Duplicate Skill_Init and non-idempotent Soldier AI cap	origin/master Client/Init/Init_Client.sqf:873 (single call), WFBE_SK_FNC_Apply at :896; Client/Module/Skill/Skill_Init.sqf:39-49	origin/master and origin/release/2026-06-feature-bundle still run Skill_Init.sqf twice before applying skills. This docs branch removes the second init and keeps WFBE_SK_FNC_Apply; Arma smoke remains pending before calling the fix shipped.	Merge/adopt the branch fix, then smoke Soldier/non-Soldier caps and respawn skill reapply. See Client skill init idempotency.
P2	Factory queue broadcast churn and soft-lock	Client/Functions/Client_BuildUnit.sqf:167-172, Client/Functions/Client_BuildUnit.sqf:365-369, Client/Functions/Client_BuildUnit.sqf:467-469	Current source still has the random queue token and empty-vehicle counter leak. Public queu broadcasts also remain on queue mutations.	Smoke repeated crewless buys, normal crewed/infantry buys, concurrent buyers and factory-dead cleanup. Broadcast reduction remains a follow-up. See Factory queue cleanup.
P2	WASP marker dialog wait propagation/smoke	Docs/source HEAD@c9df2a8b7264 is unchanged from 46840f048bd4 for checked WASP paths and still has the unslept display-54 wait at WASP/global_marking_monitor.sqf:57,62,64,68-69,80; current stable/B74.1 origin/master@f8a76de34, B74.2 origin/claude/b74.2-aicom@21b62b04, B69 origin/claude/b69@8d465fce, B74 origin/claude/b74-aicom-spend@b23f557f, origin/feat/client-fps@709258e7 and live map branches already add sleep 0.1 at :64 in both maintained roots; Miksuu b8389e748243, perf 0076040f8a5e and historical a96fdda28087 remain old-shape.	The hot-loop patch is no longer a current-stable/B74.1/B74.2/B69/B74/client-fps/live-map code task, but docs/source, current Miksuu, perf and historical old-shape targets still need the one-line throttle or a deliberate waitUntil refactor if they remain targets. All targets still need marker-dialog smoke before verified-runtime wording.	For old-shape targets, add sleep 0.1 before findDisplay 54 or refactor to throttled waitUntil, then smoke map double-click marker naming, Enter prefixing, Escape cleanup and timeout/no-dialog input re-enable. See WASP marker wait cleanup.
P3	Uninstrumented client modal menu loops	Client/GUI/GUI_BuyGearMenu.sqf:503, Client/GUI/GUI_TransferMenu.sqf:94, Client/GUI/GUI_RespawnMenu.sqf:113, Client/GUI/GUI_UpgradeMenu.sqf:282; maintained Vanilla has the same paths/lines. Broader cadence map in Client UI systems.	These four dialog-only controllers poll at sleep .01 and have no local PerformanceAudit_Record writer in the checked files. They are the highest under-covered client UI cadence, but they only run while the modal dialog is open.	Do not blindly coarsen sleeps. Add client-side measurement first, then test dirty flags/event latches or lower refresh only with gear, transfer, respawn selector and upgrade-list smoke.
P3	Cleaner/restorer wide scans	Server/FSM/cleaners/crater_cleaner.sqf:14-49, Server/FSM/cleaners/droppeditems_cleaner.sqf:62, Server/FSM/cleaners/ruins_cleaner.sqf:9-28, Server/FSM/restorers/buildings_restorer.sqf:10-26; cadence/cost guide in Marker cleanup/restoration atlas	These use very wide class-filtered nearestObjects scans, but run on long timers, sleep between per-item work, and already emit PerformanceAudit records. droppeditems_cleaner.sqf now runs a single ~20 km weaponholder scan at :62 (the B40 header at :5-14 documents the two redundant Mine/MineE scans being removed; _perfMines stays 0), so it is largely audit-first rather than a scan-narrowing target. Garbage/empty-vehicle collectors are the opposite shape: 0.5-second registry drains where idempotent queue/flag behavior matters more than scan narrowing.	Do not patch first. Use RPT audit rows to prove actual cost, then consider smaller terrain-aware centers/radii, combined scan strategy or tracked-object lists. Fix mine pair removal and garbage flag mismatch as correctness items before speculative cadence changes.

What Was Read

Source files:

• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/GUI/serverFpsGUI.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/Module/serverFPS/monitorServerFPS.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/Init/Init_Server.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/Module/supplyMission/supplyMissionStarted.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/WASP/global_marking_monitor.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Client/Functions/Client_BuildUnit.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/Functions/Server_HandlePVF.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Client/Functions/Client_HandlePVF.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Common/Init/Init_PublicVariables.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Client/FSM/updatetownmarkers.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Client/FSM/updateteamsmarkers.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Client/Client_UpdateRHUD.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Client/Module/Skill/Skill_Init.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/FSM/server_town.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/FSM/server_town_ai.sqf
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/FSM/cleaners/*
• Missions/[55-2hc]warfarev2_073v48co.chernarus/Server/FSM/restorers/buildings_restorer.sqf

Wiki/docs:

• AI, headless and performance
• Feature status register
• Hardening roadmap
• Arma 2 OA external reference guide
• agent-hardening-backlog.jsonl

Notes By Area

PVF Dispatch

Current stable origin/master@0139a346 compiles each listed client/server handler into CLTFNC* / SRVFNC* globals and installs PVEHs at Init_PublicVariables.sqf:55-61. Server_HandlePVF.sqf:14-15 and Client_HandlePVF.sqf:32-33 now resolve _script with missionNamespace getVariable and spawn only CODE, so the DR-38 runtime recompile is source-present/fixed there.

The remaining performance-adjacent work is to add explicit registered-handler allowlists/logging on current stable and backport the full lookup+allowlist to older targets that still use Spawn (Call Compile _script). Do not treat it as a micro-optimization only; it belongs in the P0 hardening branch because DR-55 and direct-PV authority work still depend on a clear dispatcher boundary.

Validation:

• Source-only: every registered _clientCommandPV and _serverCommandPV resolves to CODE after init and is present in the matching allowlist.
• Dedicated smoke: one server-bound PVF and one client-bound PVF still work.
• Negative smoke: bogus command string is rejected and logged without executing.

Hosted Server FPS Loops

On Miksuu master@b8389e74 and origin/perf/quick-wins@0076040f, serverFpsGUI.sqf and monitorServerFPS.sqf both run while {true} with their only sleep 8 inside if (isDedicated). On non-dedicated/hosted server, that branch is skipped and the loop has no sleep.

Repo docs branch HEAD@4a1bc13edde6 is source-unchanged from d0161083 / a27086cd for the checked FPS paths and already has the low-risk DR-19 shape in both maintained roots: both publishers exit immediately on !isDedicated, then keep the dedicated 8-second cadence. Current stable/B74.1 origin/master@f8a76de34, current B74.2 21b62b04, B69, B74, client-fps, ASR FPS and historical a96fdda2 use a different fixed shape: one guarded serverFpsGUI.sqf publisher and no maintained-root monitorServerFPS.sqf; B74.2 only line-drifts source-Chernarus init/removal anchors for this lane. The remaining current work is smoke plus the separate decision of whether to consolidate two closely related FPS variables; do not tell stable/B74-shaped refs to port the old two-publisher guard blindly. No release/* or origin/dev/july-update-hosted-server-fps-loop-fix head was found on 2026-06-24, so create or restore that branch before code work.

The old-FPS scout rechecked the consumers: RHUD reads SERVER_FPS_GUI (Client/Client_UpdateRHUD.sqf:113), while no obvious current-source consumer was found for WFBE_VAR_SERVER_FPS beyond its publisher. Do not remove it on that search alone; use a branch smoke or full consumer grep across maintained/generated mission copies before consolidation.

Validation:

• Hosted/listen source smoke: both publisher scripts exit when !isDedicated.
• Dedicated smoke: both HUD/status consumers still receive server FPS at the expected cadence, or documented consumers are migrated to one shared variable.

Supply Mission Scan

Docs/source supplyMissionStarted.sqf:25,28 now uses a truck-only typed Base_WarfareBUAVterminal scan in both maintained roots. Current stable origin/master@0139a346 uses the heli-aware typed scan at :55,61, with 400 m heli / 80 m truck radius selection. Current Miksuu b8389e74 and origin/perf/quick-wins@0076040f still scan all object classes within 80 meters and post-filter Base_WarfareBUAVterminal at :25,28.

Docs/source 15563691 carries the truck-only typed 80 m scan; current stable carries the heli-aware typed scan; historical release-line anchors a96fdda2 and 7ff18c49 carry the PR #1-compatible truck/heli shape, but current origin exposes no live release/* head in the 2026-06-22 remote scan. Current Miksuu/perf still need the old-shape port if they remain targets. The code already wanted that class family, and the 8-meter nearby-player scan intentionally remains broad. Arma smoke remains pending.

Do not "optimize" the command-center scan by switching it to nearEntities: the target is a Base_WarfareBUAVterminal structure, so the safe shape is class-filtered nearestObjects/nearObjects plus isKindOf (supplyMissionStarted.sqf:25,28). In this mission, nearEntities belongs to entity/logics proximity scans such as camps, towns, vehicles and units.

Validation:

• Current stable: typed truck/heli command-center scan is source-present in both maintained roots.
• Docs/source: typed truck-only command-center scan is source-present in both maintained roots.
• Current Miksuu/perf: still broad-enumerate the 80-meter command-center scan, then post-filter Base_WarfareBUAVterminal.
• Release 7ff18c49: both maintained release roots have the typed ["Base_WarfareBUAVterminal"] command-center scan and still have the broad nearby-player scan.
• Dedicated smoke: supply completion still detects command centers for both sides and PR #1 supply helicopter/truck variants.
• JIP note: completion detection is server-side; cooldown JIP behavior should remain pull-based and unchanged.

Duplicate Skill Initialization

origin/master runs Skill_Init.sqf once at Init_Client.sqf:873, then calls WFBE_SK_FNC_Apply at :896. The duplicate init is no longer present in current master. This docs branch has the intended shape: Skill_Init.sqf once at :547, then WFBE_SK_FNC_Apply at :571.

Skill_Init.sqf:49 mutates WFBE_C_PLAYERS_AI_MAX for Soldier class by setting it to ceil (1.5 * current value). Because there is no guard or remembered base value, the double init compounds the cap. With default WFBE_C_PLAYERS_AI_MAX = 16, first init gives 24 and second init gives 36.

This is patched in this docs branch's source Chernarus by removing the second client init call while preserving the later WFBE_SK_FNC_Apply call. origin/master and the release branch still need adoption; Arma smoke remains pending for Soldier/non-Soldier caps and respawn reapply.

Validation:

• Branch-local source: done. This docs branch runs Skill_Init.sqf once and still calls WFBE_SK_FNC_Apply; origin/master and origin/release/2026-06-feature-bundle still have the duplicate init.
• Local smoke: Soldier sees the intended AI cap; non-Soldier classes keep their normal cap.
• Respawn smoke: Client_PreRespawnHandler.sqf can still call WFBE_SK_FNC_Apply.

Factory Queue

Client_BuildUnit.sqf:167-172 still uses the low-entropy random FIFO token. Queue changes are still broadcast with _building setVariable ["queu", _queu, true] at :172, :191 and :207. Empty vehicle purchases still need local queue-cap cleanup before exiting at :365.

The hardening backlog item factory-queue-cleanups is patch-ready: soft-lock cleanup and token identity are the first pieces, while public queue broadcast reduction remains a separate UI-aware follow-up. Because the queue is visible to clients, do not silently remove network publication without checking UI consumers.

Validation:

• Repeated crewless empty-vehicle purchases do not soft-lock the local queue.
• Two simultaneous buyers cannot collide on the front token.
• Queue UI remains correct with reduced broadcasts.

WASP Marker Dialog

Docs/source HEAD@c9df2a8b7264 is unchanged from 46840f048bd4 for checked WASP paths, and docs/source, current Miksuu b8389e748243, origin/perf/quick-wins@0076040f8a5e and historical a96fdda28087 still need a throttled display-54 wait while preserving input re-enable after display-open or timeout (global_marking_monitor.sqf:57,62,64,68-69 with display-12 already throttled at :80). Current stable/B74.1 origin/master@f8a76de34, B74.2 origin/claude/b74.2-aicom@21b62b04, B69 origin/claude/b69@8d465fce, B74 origin/claude/b74-aicom-spend@b23f557f, origin/feat/client-fps@709258e7 and live map branches already add sleep 0.1 at global_marking_monitor.sqf:64 in both maintained roots; the launch lines are stable/B74.1/B69/B74/live-map Init_Client.sqf:397, B74.2 source Chernarus :403 and maintained Vanilla :397, client-fps Chernarus :308 and client-fps Vanilla :287. B74.2's checked Init_Client.sqf delta is unrelated vehicle-tint/join-ACK/intro-music drift, not a WASP monitor behavior change.

This remains a small local UI cleanup, not a server-wide performance fix. See WASP marker wait cleanup for the target-specific patch shape and smoke plan.

Validation:

• Double-clicking the map still opens marker creation and prefixes marker text with player name on Enter.
• Escape still clears handlers.
• Input is always re-enabled, including timeout/no-dialog cases.

Not First-Patch Items

Town And Town-AI Loops

server_town.sqf and server_town_ai.sqf use per-town nearEntities scans, but both already have cooperative per-town sleep 0.05, a sleep 5 between full cycles, and PerformanceAudit records with scan/detected/cycle metrics.

Patch only after RPT evidence shows these rows dominate. Changing detection cadence or class filters risks gameplay behavior: capture, town activation, despawn and patrol behavior depend on those scans.

AntiStack And Collector Loops

AntiStack loop candidates are measurement-first, not first-patch simplifications. The server startup cluster launches countPlayerScores, updateScoreInternal, skillDiffCompensation and launch-join ACK workers, while garbage and empty-vehicle collectors drain registries at short cadence. These loops have correctness roles around join balance, skill totals, score persistence, cleanup idempotency and player-occupied object safety.

Do not reduce cadence or remove waits before collecting RPT/DB latency evidence and checking disabled-AntiStack, teamswap, launch join, late join and match-end behavior. AntiStack and PerformanceAudit should be explicit A/B variables in old-vs-current performance weekends because both are current-only support surfaces and both default on in current parameter/config paths. For collectors, validate array shape, idempotency flags and crew/player occupancy before optimizing queue drains.

Client Marker Loops

updatetownmarkers.sqf now caches marker names/text, delays closed-map refreshes, and records skipped text writes. updateteamsmarkers.sqf updates only when map/GPS/Warfare UI is visible, throttles AI leader updates to 1 second, and records marker operations.

Common marker helpers are in the same caution bucket: Common_MarkerUpdate.sqf:73-109,182-229 and Common_AARadarMarkerUpdate.sqf:54-75,105-106,184 already cache marker text/position/color or radar marker state. These are not first-patch targets. Use PerformanceAudit output before touching cadence.

RHUD

Client_UpdateRHUD.sqf sleeps 1 second, caches display/control state, and avoids redundant text/color writes. It should stay measurement-led.

Cleaner And Restorer Wide Scans

The whole-map cleaner/restorer scripts have wide scans, but they are class-filtered, low cadence and instrumented. Their sleep 0.5 per item intentionally spreads deletion/restoration work. Treat them as audit-first candidates.

The current source-level cadence readout is now in Marker cleanup/restoration atlas. Use these labels in the server RPT before tuning: cleaner_droppeditems, cleaner_craters, cleaner_ruins, restorer_buildings, cleaner_mines, server_garbage_collector and emptyvehiclescollector. A wide scan is not automatically a hot path when the timer is 30-90 minutes; a 0.5-second loop is not automatically unsafe if the registry stays tiny and idempotent.

Suggested Patch Order

1. PVF dispatcher lookup, because it is both P0 security and performance.
2. Hosted server FPS loop sleep/consolidation, if hosted/listen mode still matters.
3. Branch-scoped by target: docs/source and current stable carry typed supply mission scan narrowing, current Miksuu/perf remain broad/post-filter, no live release head is exposed by origin, and Arma smoke remains.
4. Factory queue cleanup is patch-ready for counter leak and token identity; current source still needs the code patch, and broadcast reduction remains a later UI-aware follow-up.
5. WASP marker dialog wait cleanup is target-specific: docs/source, current Miksuu, perf and historical old-shape refs still need the throttle/refactor; current stable/B74.1/B74.2/B69/B74/client-fps/live-map refs need marker-dialog smoke before verified wording.
6. Audit-led cleaner/restorer or marker-loop tuning only after RPT evidence.

Handoff

For Codex:

• Keep this page linked from dashboard/status/context/backlog.
• Avoid scattering performance notes across subsystem pages unless they are implementation-ready.

For Claude:

• Good contradiction checks: prove whether the double Skill_Init call is intentional or historical accident; inspect whether any UI path consumes every broadcast queu update; verify exact class inheritance for Base_WarfareBUAVterminal in OA before a scan-narrowing patch.

For a future code owner:

• Smallest code patch: hosted FPS loop sleep/exit.
• Highest value patch: PVF dispatcher lookup from the existing playbook.
• Completed branch-local cleanup packages: Client skill init idempotency and Hosted server FPS loop sleep; Supply mission scan narrowing is source-present on docs/source and current stable, old-shape on current Miksuu/perf, and still smoke pending.

Continue Reading

Previous: AI, headless and performance | Next: Hardening roadmap

Main map: Home | Fast path: Feature status | Agent file: agent-context.json