MOON / FIRST DAWN V3 / DOCUMENTATION

September 10 — supply the machines that make progress

Today's journal covers every main release: waiting-input logistics, Industry by Seed Base, stable carrier controls, asset Priority, Hide HUD, Explorer driving and handling, launch supply progress and Kestrel's closing cargo bay. The public board links this complete journal in post796610; the five next-horizon proposals are in thread796613. Those proposals are not unlocked features.

Updated visual overview · Driving controls · Transport guide · Five future cooperation ideas

Explorer is the rover’s name

Corey named the drivable rover Explorer. The Crew card, research, inspector, driving controls, API messages, manual and overview now use that name, including the lettering on its chassis. Existing vehicle IDs, saves, research progress, commands and driving links remain compatible. This is a naming update; vehicle physics and material costs are unchanged.

Logistics correction — live at 11:04 UTC

An enabled parts workshop could wait for metal while continuous new refinery output repeatedly won the next robot. Meanwhile, other automatic consumers could take parts back out of a robot factory trying to gather its next build bill. Having plenty of settlement stock did not mean the workshop had received its input.

The correction prioritizes input deliveries above routine output storage. Construction and shared commitments retain priority. Older routine loads gain priority after two or four minutes, so production cannot permanently crowd out output clearing. Robots finish existing assignments; no cargo is teleported, cancelled or recreated. Automatic consumers retain the next queued robot bill, planned replicator bill and a spare workshop's four-part working buffer, sharing the surplus.

A replay of the same seven-player saved world on the bundled lunar elevation data ran for 30 minutes with no new orders or resource grants. The candidate produced 13 parts across the two affected workshops versus 6 with the previous runtime. All nine original workshop metal packets arrived; three remained pending with the previous runtime. The robot factory retained 10 parts toward its 12-part carrier bill versus 4 before. No new robot completed within this interval. In the longer candidate replay, the queued Good Neighbor carrier completed by the 45-minute sample, using the original colony resources and ordinary transport. These are replay measurements, not a promised completion time in the changing live world.

The Industry command notice also stays at the top of its scrolling panel, making conveyor failures visible near long lists of machines. The complete construction bill is still required before an installation starts. A proposed Rush delivery button remains an idea; this patch improves ordinary automation without adding a new command.

Eight focused regressions cover production-input priority, aging output and warehouse moves, reserved-cargo preservation, queued assembly bills, spare-workshop buffers and surplus sharing. All 393 CPU/API tests pass in the final run. The initial isolated notice test did not run; no rendered scrolling verification of that first release is claimed. V3 runtime/client a7a15ec and website 13b6f2d were verified publicly with all seven players preserved; the older V1/V2 releases remain unchanged.

Industry grows with your settlements

Industry now starts with a Seed Base selector. All my seeds shows your settlement directory, machine and construction counts, and which seeds have blocked machines. Choose a seed to open its Industry controls; your camera moves to that settlement. Visiting a neighbor still grants no operating permission.

Within a seed, filter by machine type or choose Needs attention, Waiting for inputs, Under construction, or Switched off. Search by name, ID or process. Machines are grouped by type in compact rows; expand one for its full state, physical inventory, storage capacity and existing controls. Construction rows retain Locate and Cancel & recover. Pages contain at most 40 matching machines. The all-seeds directory is an overview, not a new automated governor or a world-capacity increase.

Freight automation and precise deliveries, rocket cargo, carrier missions, bore tunnels, and the cargo ledger each have a named expandable section below the directory. Their larger forms render when opened. Open sections and machine rows survive world updates. Delivery, rocket, carrier and bore drafts stay with their seed rather than following you into a different settlement. Resource recovery links still open the corresponding delivery form for review; they do not issue a command automatically.

Validation used all 397 CPU/API regressions, plus a CPU-rendered synthetic fixture with ten seeds and 160 machines per seed. It verified bounded rows, expansion through snapshots, seed-scoped drafts and commands, lazy freight controls and a narrow viewport without horizontal overflow. This verifies the directory UI, not a new 1,600-machine server capacity.

This is a client organization change. Costs, research, machine operation, resource movement and authorization remain server-controlled.

Play V3 · AI manual · Master overview · Previous devlog

Carrier controls: interaction repair

A live snapshot could replace a button between pointer press and release, swallowing the click. Snapshots now preserve controls under the pointer and keyboard focus; explicit actions still refresh, and live refresh resumes when interaction ends. Good Neighbor and Wayfarer Crew cards now use their actual Next Horizon image files. A CPU browser test reproduced the old failure, then verified the repaired click, exactly one complete kit dispatch, successful image loads and resumed updates in a synthetic world. No player command or resource changes are part of this client fix.

A kit mission reserves receiving space. If a destination fills before dispatch, choose another depot with room; acceptance and physical delivery remain separate. Good Neighbor currently still participates in ordinary hauling. Reserving it for specialist kit/repair work is being discussed, and is not changed by this release.

Deliveries that unblock work

The earlier priority repair still let routine cargo age into the same class as waiting production inputs. Under a large backlog, the dispatcher became an old-load queue again. A parts workshop or queued robot could remain starved despite paid material waiting for collection.

Empty transports now favor missing inputs for the next usable batch. Queued robot and prefab assembly bills come first, then workshop batches, then other production. The calculation counts physical inventory and deliveries already assigned to robots or fixed lines; an unassigned reservation alone does not satisfy a need. Once one batch is covered, another robot can serve a different need. Robots finish existing carried loads; reservations and resources are preserved.

Cargo also becomes urgent when removing it clears an output-full or storage-blocked machine. A full hopper with a pickup already assigned does not commandeer another empty vehicle. Every fourth demand-led assignment serves the ordinary aged queue, preserving turns for construction, shared commitments and routine storage. Power, Mind support, blocked paths and missing material remain separate causes; a warning cannot create a route or resources.

All 403 CPU/API tests passed, including failing-before/passing-after regressions for aged cargo versus waiting inputs, next robot assembly, finite batch coverage, paused machines, output clearing and fairness. In a 30-minute replay of the same observed seven-player world on the actual bundled lunar elevation data, the correction produced38parts and completed four queued robots, versus7parts and one robot with the exact pre-correction runtime logic. It left41of311original cargo packets pending versus one, a deliberate throughput tradeoff: finishing productive batches before draining all old storage moves. Concurrent player actions can change live outcomes; these are measured simulation results, not delivery promises. No new rush button or user-managed priority settings are required.

Your chosen priority, and room to see the Moon

Players can now select Priority on an installed asset in Industry or its click popup. A visible row badge and Priority assets filter make the choices easy to find. Empty robots favor useful deliveries and pickups there after finishing their current jobs. Already assigned loads count; multiple selected assets share the crew. Clearing the toggle returns that asset to automatic priorities. There is no cost, teleportation or change to production rates.

The left mission/objectives/Lunar Mind panel can be collapsed with Hide HUD, leaving a small HUD tab to restore it. The browser remembers the preference. Resources, urgent resource notices and Settlement remain accessible.

Validation includes real command authorization, read-only preview, idempotent writes and persistence across restart; dispatcher tests cover existing missions, batch coverage, output clearing, multiple priority assets and fuel delivery. A CPU-only browser fixture verifies both toggle surfaces, the filter, persisted HUD collapse, keyboard restoration and a reachable mobile handle. No tower GPU was used.

Field Trip — the Moon at wheel height

A settlement can now research and build a rover you drive yourself. Field Trip · lunar driving follows Survey and Storage, costs600 work, and needs two continuously powered Minds. The first rover is assembled at the Seed Base; additional vehicles use a Robot Foundry. Each costs24alloy,6parts,2electronics,2cells and1spare, plus240supported assembly seconds.

Open Crew → Drive · first person / chase, or click the rover. Use WASD/arrows, Space to brake, C for the camera, L for lights and Escape to park. The six mesh tires have independent suspension, steering and wheel rotation. A fixed-step physics engine gives bumps, airborne motion and landings lunar gravity. Mouse look, chase zoom and touch buttons are included.

Its12kWh battery is finite. Park near a powered Seed Base or Freight Depot for automatic charging, paid by the local grid. The exploration radius is16km around its home. A stopped rover can use its self-righting jack when it has energy and clear space. Lost controls or Mind support apply brakes; the server owns the position, and observers see the same rover. No normal freight is assigned to Field Trip.

Entering a nearby rover preserves the loaded landscape. Turning in place or changing the camera projection refreshes terrain selection; wheel-scale collision does not depend on visual detail. Existing raised logistics jobs also resolve their actual operating ports, so an older lift-fitout job can accept already-paid cargo instead of waiting at a blocked machine center.

All430 integrated CPU/API regressions pass. CPU-only rendered checks cover first-person/chase control and another player's observation; numerical tests cover motion, braking, contacts, cameras, support loss, charging, permission, stale inputs and restart. World backups and V3-only staging precede publication. This is human driving with an owner control API; autonomous visual AI driving and a general forward-simulation API remain future work.

Driving controls and API · Crew and machine gallery

Explorer: more planted, smoother to drive

Corey's first drive exposed two different problems: small grades launched the rover too easily, and server corrections snapped its orientation. Explorer's simulated mass rises from420 to520kg, suspension damping increases, and steering tapers with speed. The powered cruising speed is9m/s (32.4km/h), with the same1.62m/s² lunar gravity. Larger jumps, momentum, finite battery and physical braking remain.

Position and rotation now reconcile together. Chassis and tire presentation interpolates between the60Hz physics steps, including on faster displays. The local terrain collision patch retains2m resolution while reusing overlapping samples; a forward shift samples588new points instead of4225. Driving spends less time per frame building visual terrain, and instrument text updates ten times per second.

In a repeatable forty-second gentle-hill drive, airborne time fell from9.96% to2.92%, and the longest small-crest flight from1.08s to0.52s. On bundled lunar terrain, observed collision-rebuild spikes fell from11–26ms to about3–6ms in a CPU run, with the same measured route distance. Tests reproduce the old yaw snap and verify continuous correction and120Hz presentation. These are bounded measurements, not a promised frame rate on every device. Public release follows the full regression, disposable CPU browser and backup checks. Refresh after publication for matching client/server physics.

Rocket loading: next free transport, visible supply progress

An ordered Kestrel flight could wait behind ordinary deliveries even after its cargo and propellant had been reserved. Launch inputs now get the next suitable free transport ahead of routine work and asset Priority. Current trips finish. Assigned material counts toward the finite launch bill, so surplus robots can help elsewhere. Supported conveyors and pipelines keep serving their routes. Cancellation or loss of ground support removes the automatic launch boost.

The loading card separates cargo from each propellant, shows actual arrival versus reservation, lists assigned robots or lines, and identifies unreserved shortages. Travel estimates are approximate and only appear after assignment. The final supported loading countdown is separate. The Guide sees the same structured report. A further accounting correction prevents aluminium being shipped as cargo from also satisfying the fuel reservation, for both robot and fixed-line requests.

Repeated flights reward built logistics: an aluminium conveyor and a liquid-oxygen pipeline can maintain existing20-unit pad buffers between missions. Coldwell still has to liquefy actual oxygen; no new propellant, free infrastructure, faster robots or flight-cost changes were added. Idle pads do not request automatic robot stockpiling.

Regression coverage includes launch priority, completing existing rides, finite assigned coverage, cancellation/support loss, working fixed routes, priority packet packing, cargo/fuel separation and the loading display. All445CPU/API tests passed, and a disposable CPU-only browser verified waiting supplies, assigned robots, loading countdown and narrow layout with no errors. Publication verification is recorded in project operations. Driving frame-time investigation remains a separate follow-up; a new native client or revised simulation architecture has not been implemented in this release.

Kestrel prepares for launch

The flight model now has a recessed cargo bay, a movable door and retracting lift hardware. During the existing loading countdown, the elevator raises the cargo, draws it inside, retracts and closes the door before liftoff. Cargo and the lift stay out of sight throughout the outbound and return flights, including a loaded return after an unavailable receiver. The receiving pad reverses the sequence to unload, then secures the ship again.

The motion follows actual flight progress and pauses with ground operations. It no longer repeats a showcase loop or freezes a loaded elevator outside the flying hull. Refresh to load the versioned Kestrel model. Flight bills, travel time, delivery timing and saved progress are unchanged.

Explorer: less work competing for each frame

The next frame-time investigation reproduced two avoidable CPU costs in a captured large colony. The HUD repeatedly rebuilt resource diagnostics that the observation already supplied. The chase camera searched distant detailed factory meshes twice per frame while checking for nearby walls.

The client now reuses diagnostics only when both the observation tick and selected Seed Base match, and retains derived HUD data until the snapshot changes. Older servers and visits to other seeds use a fresh local report. The chase camera first excludes equipment outside a conservative nearby region, then checks the remaining detailed meshes exactly. Nearby-wall retraction, the previous camera position, lunar motion and collision rules remain covered. Sampled animation poses for all current machine and robot models fit within the conservative region.

A fixed captured-world browser comparison isolated CPU work by disabling render submission, while retaining the actual driving, scene and snapshot paths. During chase driving, median HUD update cost fell from15.1ms to1.6ms; median driver update cost fell from6.9ms to0.5ms. The driver’s95th-percentile cost fell from52.1ms to2.0ms. First-person HUD work also decreased. These are bounded VPS CPU measurements, not the player's GPU frame rate. Longer pauses still occurred, and first-person frame-tail measurements did not materially improve. Scene traversal, graphics load and occasional terrain work remain follow-up targets.

The full-world software-rendered attempt could not sustain driving control and is retained as a failed profiling route. A separate small-world rendered check passed with 22.14 metres of server-observed keyboard travel, first-person/chase views and another player’s observation. All456CPU/API regressions passed, including new stale-report, seed-switch, distant-camera-mesh and asset-bound checks. No tower GPU was used.

The whole-day whitepaper refresh

The master overview now has a dated September10 entrance, actual Explorer chase/first-person verification images, three Kestrel02 preparation poses, the current34-model gallery and a library that includes both recent journals, driving and transport. Transport and driving belong in the narrative before the closing research library. The previous storage chapters and measured experiments keep their dates and scope.

Five proposals connect the physical game to cooperative learning: sunward manufacturing slots, radiator gardens, joint crisis rehearsals, rescue beacons and a constellation of recorded contributions. They remain ideas. General forward simulations, autonomous Guide actions, manual rocket landings and in-game screenshot/video capture are not newly shipped by this update. Corey deferred the old laptop video trim and plans new footage after the motion work.