Playtest Build 11

Apogee — Update, 6 September

This one is about the two things that were costing you the most: waiting, and landing.

Time no longer has to be integrated to pass. Over a stretch of vacuum with nothing happening, the trajectory is a conic and the game now solves it in closed form — a coast that took 297 microseconds of ticking takes 0.133, with no drift at all. That is what makes the new jump buttons possible, and it is why a three-day lunar transfer is now something you press rather than something you sit through.

And the landing autopilot finally knows how fast you are actually going. It was solving a one-dimensional problem and telling you to start braking at 568 meters with two kilometers a second of sideways motion. It is not doing that any more.

The wait is over

• Jump to the maneuver, the apoapsis, the periapsis, or the arrival. Four buttons in Controls, for the four moments anybody actually waits for. All four are facts about the arc you are already on, which is exactly why jumping to one cannot change where you end up.
• It stops an hour short, on purpose, and hands over to the existing auto-warp so the approach to a burn happens with the vehicle really being flown and you still able to change your mind. A jump that landed on the event would be a jump that skipped it.
• It refuses when skipping would lie. Under thrust, on RCS, on the ground, in air, with a periapsis that could reach air or ground inside the interval, with a hot skin, or with another body's sphere of influence close enough to cross — the jump stops there and says the way ahead needs flying. Everything a normal tick advances still advances: power, heat, consumables, reliability clocks, attitude. A week that passes without the batteries discharging is a physics warp of a different kind.
• Sitting on a surface no longer crawls. A vehicle at rest read zero altitude and drew the strictest warp rung in the table — 50x — so an hour on the Moon cost twelve minutes of real time and a two-day stay the better part of an hour. Standing still cannot skip over anything, because nothing that could be skipped is happening. The ceiling at rest is 10,000x.
• The autopilot stopped re-deriving what a separation drops, sixty times a second. Two questions asked every physics sub-step cost more than the physics did. Which parts leave with a separator is a fact about the vehicle's shape and changes when the shape does; it is worked out once now instead of 64 times a second, and the sub-step is 600 times cheaper for it.

Coming down

• The braking burn is sized for the speed it has to cancel. The old solver worked along the local vertical only, so two kilometers a second of horizontal motion cost nothing and it lit the engine a few hundred meters up — pointed retrograde, which at that flight path angle is very nearly sideways. The burn is solved along the flight path now: the speed to kill is the whole velocity, and gravity is resolved onto that path. The reported case reads 2,016 m/s of braking over 207 km of altitude, which is the truth about it.
• It lights slightly early. A burn started at exactly zero margin has nothing left to pay for terrain it cannot see, a throttle floor it will pulse against, or a gravity that is not quite the constant the model assumed. Five percent of the braking distance is held back.
• "BURN NOW" instead of "TOO LATE". A lunar descent begins from a state no vehicle can hover-slam out of, and lands by burning continuously from the moment the margin runs out. Not being able to stop from here is not the same as being doomed, and the cockpit now tells those apart.
• A lander comes to rest on the ground, not on the datum. Ground elevation was surveyed once at the launch site and set back to zero on arriving at another world — so on the Moon, whose maria really do sit kilometers below the reference sphere, a lander touched down reading zero altitude while hanging 4,974 m over the surface.
• The camera stops at the ground too, instead of at a floor five kilometers above it. Zooming in on a mare landing used to stop dead at "0 km" with the craft still visibly below.
• Altitude reads down to the surface below 25 km. Above that the number that matters is the orbit's, measured against a reference sphere; below it you are on your way to a surface and the distance to that surface is the only figure worth reading. Labeled "altitude (RA)" with the radar altimeter aboard and plainly not that without one, because a readout that quietly changed which of the two it meant would be worse than either.
• The landing gear swings out. Three struts hinged near the top of the part, a drag brace back to the hull, and a second and a half of travel — in real seconds, so it does not snap through the whole motion in one frame at high warp. A torn-off leg is not drawn at all.
• A wreck is not asked whether it could have stood up. The stability check ran before the hull limits and returned "toppled but survivable" — so a legged craft arriving at 1,700 m/s across the ground scored a landing and had its velocity quietly zeroed. Both hull limits are asked first now, of every vehicle.
• Guidance waits to be asked before it lights anything. A vehicle standing still on the ground has its whole climb ahead of it, so guidance opens in Ascent rather than Landing — and it idles until you stage. That is the LAUNCH button on the pad and the STAGE button on a surface, so there is one gesture for "go" rather than a special case for the clamps. Nothing burns on the clamps any more either; the pad still previews thrust and TWR so you can see whether the stack will lift.
• A landed vessel with empty tanks can be written off where it stands. "Recover vessel (end the mission)" sits beside "Commit to orbit". Offered only when there is nothing left to burn, because ending a mission that could still continue is a decision you should have to make with the engine.

The assembly building

• "Simulate loads" flies the rocket. The load path used to ask you what dynamic pressure and what angle of attack the vehicle would meet, which is a question you had no way to answer, and then report the loads at whatever you guessed. It now flies a real gravity turn off the real launch complex, a tenth of a second at a time, and keeps the worst load every joint ever sees. On the reference vehicle: 240 s of flight, max-Q of 35,851 Pa at 12.2 km, 6.5 g peak — and the worst joint goes from 52% standing to 67% flown. Both sliders are gone. It takes a few milliseconds.
• The load table says nothing until the rocket has been flown. Its old fallback was a standing-on-the-pad estimate, which reads exactly like a measurement: a column of green percentages under a button nobody had pressed. Edit the craft and the previous numbers stay, marked as belonging to the design they were flown with.
• 80% to the red threshold now reads orange. A joint at 96% was drawn the same green as one at 4%, so the panel said "fine" about a joint one gust from being the one that fails.
• Smart Stage. One press builds the whole sequence from the vehicle: a stack separator opens a stage and leads it, an engine lights in the stage it sits in, spent boosters get a stage of their own, and chutes go last and alone.
• A "+" on each stage header inserts a stage above it. There was previously no way to make an empty stage between two others — a stage exists because something is in it, so emptying one removed it.
• Anything bolted on the side can be given a mount. "Insert radial decoupler" splices one under the whole symmetry group, so a four-fold ring does not end up with one booster that leaves and three still welded on.
• ...and decouplers stop appearing on things that were never coming off. The staging conversion put one at every radial joint — fins, thruster blocks, landing legs, antennas — and re-ran every frame on a craft with nothing to stage yet, so it decoupled its own decouplers. It only goes under something with an engine on it now.
• A placement mode selector replaces the "add above" and "radial" checkboxes, which could both be ticked while the editor only obeyed one of them.
• A Settings screen, and it belongs to you rather than to a campaign. Reached from the pause menu, with tabs, and saved to a per-user file beside your saves — starting a second program does not re-tick your boxes and deleting one does not lose them.
• Engines can be grouped by what they are for. Fuel is the right first cut when a program has three engines and the wrong one when it has thirty. Booster, upper, vacuum, descent and vernier, authored per part rather than guessed — the Kerolox Vacuum Large is bigger than several boosters and the Methalox Vernier throttles like a lander motor, and any rule written on thrust alone files both of them wrong.
• Craft files name their parts instead of counting them. Insert a part into a file by hand and every joint, engine, strut and staging action below it used to come to mean different hardware — the file still parsed, and the vehicle it described was not the one you saved. Parts carry a permanent id now; older files load by position and are upgraded on the way in.
• Part numbers count from one, and a grouped row says which parts it is: "#11-14 4x Radial Decoupler" rather than a "4x" with nothing to match the staging list against.
• A redundant igniter is a checkbox, and solids can have one. It was an unbounded free multiplier — no mass, no cost — so any ignition worry was answered by typing a bigger number. Real motors carry an initiator and a spare. Craft with taller stacks are brought back to one on load.
• Every tank states its diameter, including the spherical, toroidal and service-module ones, where the shape is not a cylinder and the diameter is the whole question.
• The new-craft button moved to the Craft File tab, where naming, saving, loading and deleting already live — and off the parts palette, where it sat one slip of the hand from the control that decides where a booster attaches.
• A category with nothing in it no longer has a tab.

The parts catalog

• A nosecone is worth fitting, and a blunt front is not free. Body drag was one constant for every vehicle ever built, so a nosecone was thirty kilograms that did nothing. It is per-part now: a bare stack flies at 0.60 and an ogive at 0.28. Only the foremost part counts, blended by how squarely it meets the airflow — so a cone buried mid-stack shapes nothing, and a capsule re-entering blunt-end-first is correctly draggy.
• A part that is beaten on every axis stops being offered. Research the Descent engine and the Basic one leaves the palette: same vacuum thrust, better Isp, a 10% throttle floor against 60%, and lighter. It stays in the catalog and every craft built around one still loads and flies — only the offer goes, and only for the strict case where one part wins on everything.
• Mounts and separators are sized by what they carry. The sized stack separators fit identically and differed only in mass, with the heavy ones behind later tech — a tree selling a 180 kg downgrade. The sized radial decouplers held identically too, measured byte-for-byte. There is one of each now, and its mass follows the diameter it spans or the load it holds: a 388 t booster costs 1.2 t to strap on, where the "heavy" part charged 95 kg for it.
• Parts that did nothing are gone. Spacers, interstages and the four "adapters" (the taper between two diameters is derived from the joint and always was — there was never a part behind it), the strut connectors (bracing is a control on a joint), the backshell, the heavy chute mount, and the probe heat shield, which could not raise the shield allowance it was sold to raise. Craft carrying any of them load with the part dropped and say so.

The research tree

• The tree is redrawn. The start node is vertically centered at the far left, the tree's first fork becomes the vertical axis, and a node sits level with the average of the parents it is drawn to — whole chains now run as one horizontal line across five columns. Basic Recovery sits above Liquid Propulsion I, which by itself took the shipped tree from ten edge crossings to five.
• A line that crosses a column claims a lane in it, instead of sweeping over whatever boxes happened to be in the way.
• A prerequisite the picture already reaches is not drawn again. Five redundant edges are gone, and they were the long ones — a line leaving a node several columns back to say what a shorter route already said.
• Quick Research: what is next, cheapest first. A column down the left listing every node whose prerequisites are met, whether or not you can afford it — a node you are saving for is exactly the thing worth showing. The tree is the right picture for "how do I get to nuclear thermal" and the wrong one for "what can I buy with the science I just landed with".
• A node says what it stands on, bought or not. The prerequisite list used to appear only while something was missing, and listed only the missing ones — so a node ready to buy said nothing about its footing. Every prerequisite now, by name, with the flight milestone last, each marked as met or not.
• Twenty nodes cost less, and none costs more. The prices had stopped tracking what a node hands over: tier 5 asked 300 science for Interplanetary Navigation I, which grants one part, while tier 3 asked 160 for Orbital Maneuvering Systems, which grants eight. The tree falls from 5,685 to 4,345. The deepest cuts are the ones furthest out of line with their tier — Halo Orbits 340 to 170, Relay Networks 280 to 150, Lunar Surface Science 240 to 130, Cryogenic Handling I 200 to 110.
• Four nodes fold into the ones behind them. Airframe Basics was two nosecones and a toll booth in front of the engines, so the nosecones come free and what stood behind it hangs off the start node directly — the opening move is now a choice between four things that actually differ. High-Energy Re-entry granted one shield and sat behind the node granting the other. Advanced Recovery Systems and Structural Reinforcement were left granting nothing once their dead parts were retired. Not one capability was lost: everything they gave that worked is still given, by fewer purchases.
• Fourteen grants that gated nothing are off the nodes that advertised them. A node listing "Transfer Planner" among what it sells is telling you that you cannot plan a transfer without it, which was not true — the planner is gated once, as a panel, and everything inside it came along with that. Same for the burn queue, the docking HUD, the link and network readouts, the shield sizer, the corridor planner and the rest. Nothing moves; the tree just stops promising things it was already handing out.
• SAS is what Guidance & Avionics I actually sells. The other side of that audit: sas_hold was granted by that node and enforced nowhere, so 40 science and a flight to 80 km bought nothing. Every SAS mode but Off is behind it now — an attitude held for you is the vehicle flying itself, and the rule this tree runs on is that arithmetic is never gated and automation always is. Before the node, a rocket is flown on the rotation keys, which is what the first sounding flights were anyway.

In flight

• Engines burn where you can see it. A plume per firing nozzle, at that nozzle, along that engine's own thrust axis, with its length and brightness set by what that engine is producing right now — so a cluster with one motor out is three flames and a dark bell. The shape is a real plume: it flares out of the bell, necks down, and carries a shock train along its length.
• The camera goes over the top. Elevation was clamped short of overhead, so a drag upward simply ran out of travel and there was no way round to the other side at all. It now passes through the zenith and comes down the other side, and below the local horizon too, which in orbit is what continuing to rotate means.
• Every panel you can open, you can close from its own corner. All twelve have an X now, instead of sending you back to find the right icon among a dozen to get rid of the panel already under your cursor.
• One progress bar for one burn, and it is the clock. The delta-v bar ran ahead of it on purpose — a burning stack sheds mass — and two bars at two lengths for one burn reads as one of them being wrong with no way to tell which.
• The window starts maximized — genuinely maximized, not merely sized to fill the work area, so the maximize button, snapping and the resize border all behave the way Windows behaves.

Fixes

• A rocket that broke apart the moment stage 1 fired. 25 parts to 2 in a single tick, leaving the capsule falling from 40 km. The stage sequence separated four radial decouplers and the strap-on watchdog independently named the same four by position — but staging drops eight parts and renumbers everything above them, so by the time the second list was applied those numbers meant the life support, the capsule, and the two stack separators under it. Parts are named rather than counted now, before anything can renumber.
• Landing legs were being staged as separators. Four of them on a four-legged lander, shown as a "4x separate" row naming a part that is not a separator, on a vehicle with no decoupler in it. Not cosmetic: a leg mounted under the stack would have released everything above it — the top of the rocket coming off when the gear was meant to come down.
• Craft are repaired on the way in. Joints, struts, chutes and surfaces naming parts that are not there are dropped, the staging list is reconciled against the parts that exist, and the VAB says what it fixed on the status line. A repair is not a refusal.
• Stages reading 0 m/s with a full tank and a lit engine. A parallel-burn liftoff charges the core's early draw to the segment that spends it, and the segment that ended by dropping the core was then charged the whole tank again — far enough over and the burnout mass goes negative. On the reported craft the five stages now read 2,048 / 2,664 / 3,432 / 3,475 / 4,492 m/s, and the burns sum to within a kilogram of the propellant aboard.
• TWR after the boosters drop. The core engine burns alongside them and drinks from the core tank the whole time, and none of that was recorded anywhere — so the stage beginning at separation was priced against a full tank the vehicle had not had since liftoff. Measured on a 60 t core with two boosters: 1.85 became 2.73, and 2.73 is what the arithmetic gives for the mass the vehicle actually has.
• The Moon was in the wrong place. The flight read the ephemeris from mission time zero while the renderer drew every body at the campaign clock, which is however many months into your program you are — and the Moon goes round in 27.3 days. The intercept marker pointed at empty space and the craft teleported across the gap at capture. Worth being plain about: a transfer solved in month eight was solved against the Moon of month zero.
• The cheapest transfer is the cheapest one found. Two faults. The "direct" rows were the "cheapest" rows with another label on them — measured across 648 scenarios, the flag changed the answer in zero of them — so the panel offers four real options instead of six, two of which were duplicates. And the wide search could return a plan more expensive than the narrow one it contains, because its departure-window scoring never visited the first orbit; the earliest departure is always considered now.
• Corrections solve to 400 m/s. Both correction solvers bounded their search at four times a linear guess taken from a half-m/s probe, on a response that is nowhere near linear — so where the leverage was high the scan stopped short of answers it was entitled to find. And the return trim had no ceiling at all, so it would quote 965 m/s as a "correction" against a lunar-return injection of about 800. Those refuse now, and say to plan a fresh return.
• There is no atmosphere at the Moon. Two places said otherwise. The deorbit planner solved an entry corridor with Earth's atmosphere against the Moon's radius and gravity, printing a survivable band and an arrival speed, in confident kilometers, for a descent through vacuum. And the map's orbit warning tested a flat 100 km — a 100 km lunar orbit is as permanent as a 400 km one, and the only thing that can end it is the ground, which the warning now says instead.
• Life support runs while you sit on the Moon, and the endurance readout stays on screen while you do. It was the last row of a panel that grows when you land, and an auto-resizing window is clamped to the screen — so whatever sat at the end fell off the bottom. It reads with the mass and the propellant now. The crew count behind it was also read once at launch and never again, so a vehicle that dropped its cabin still claimed a crew.
• Throttle snaps to full when the clamps go, whichever control staged. The space bar was handled somewhere the new ground-idle gate never saw, so launching with the key released the clamps with the ascent unarmed and walked the throttle up from zero over about three seconds.
• An orbit can be committed at exactly 140 km. The game's own definition puts that altitude in vacuum, and the gate was refusing it by one. Below that it stays where it is, and the measurement is why: a 120 km periapsis is destroyed inside a day, and a 125 km one loses 58 km of apoapsis a day.

As always: the core simulation is deterministic and dependency-free, and every build is verified bit-identical across two compilers before it ships. This update runs 812 checks on each.