Devlog, 13 October 2026: dunes shaped by the wind, ground that weathers and a roadmap that tells the truth

On 13 October 2026, the COLONIZED repository received 158 distinct commits. Most went into the visual review of the fifteen worlds of the Aurin system, which began on 12 October and is still running. This devlog picks four threads from that day and checks each one against the code and against the task in Jira: dune fields designed from the wind, ground that weathers by real data, a plan to make the terrain cheaper, and a roadmap that had stopped telling the truth.

The three boxes from the first devlog still apply. Main build means it is in the game’s main branch. Review branch means it is real code in the branch where the fixes from the visual review are made before they are merged. Design means it exists on paper, with a prototype at most. Only the first box counts as a feature, and nothing here is promised for the alpha.

The day in one table

Thread	What changed on 13 October	Where it lives
Dunes shaped by the wind	Full design and a prototype of the physics	Design
Ground that weathers	Tone by relief, slope, wetness and sunlight; a texture method without tiling; palettes without human marks; deposit halos only where chemistry justifies them; rocks that sit in the ground	Review branch (the first texture delivery, in the main build)
Lighter terrain	A measuring tool, a cost plan with eight proposals and a first saving from orbit	Review branch
Jira from git	A script that writes each commit into its task, and Jira brought up to date	Main build

Dunes shaped by the wind (design)

The rule on COLONIZED terrain, since 8 October, is that relief comes from real elevation data or from physics, never from noise. Dunes follow it. Their shapes will come from two sources: elevation data of real dune fields on Earth, for the large forms and the small ones, and a physical model of sand transport, the slab model Brad Werner published in 1995, for whatever the catalogue lacks.

Each world’s climate will decide the rest. Its wind bands and pressure cells give a wind regime for every season; the sediment left by erosion in dry areas gives the sand. From those two inputs comes where a dune field forms, which type of dune it grows, which way its crests face and how large they are. Dunes scale the same horizontally and vertically, because the angle at which sand slides, about 33 degrees, does not change with size.

The prototype did two useful things. Changing only the wind regime and the amount of sand, the model produced the four classic types: crescent barchans, transverse ridges, linear dunes and star dunes, with crests within 0 to 4 degrees of the orientation predicted by the Rubin and Hunter rule. Then it failed in an instructive way. With Melvara’s real climate, the wind came out reversed between solstices over 59 to 84 % of the dry zone, which yields transverse and star dunes and no linear dunes at all, the most common type on Earth. The fix is in the design: before anything goes into production, the wind model is calibrated against an Earth benchmark. The estimate is five to seven days of work. Dunes that move over time are not part of this design.

Ground that weathers

On 13 October we gave the review a rule for the ground: the tone of a texture has to change with the relief, the slope and the climate, so that planets never look like smooth balls from any distance. Five pieces of work answered it.

Weathering by real data (review branch). Each world’s ground material now darkens or lightens with what the generator already knows about the place. Up close, valley floors near watercourses get wetter and darker, shaded slopes keep more cover, sunny slopes dry out, and hollows hold shadow. From afar, a tone map of the whole body darkens exposed rock, turns alluvium darker and greener and reddens iron-rich soils. Nothing in it is noise. The cost, measured at 2128×934 pixels, about 2 megapixels, on our development PC: +0.044 ms at 2 km, +0.060 ms at 200 m and +0.116 ms at 20 km. Two earlier versions were thrown out on the way: one laid a dark veil over a whole plain, another cost too much for what it changed.

Textures without a grid (first delivery in the main build, the rest in the review branch). Seen from a few hundred metres, a repeated ground texture draws a grid. The new method breaks it up with a randomised hexagonal layout and uses the same texture at four scales that only modulate each other and never swap with distance, so nothing pops. In the test, the bands on rock between 20 and 200 m and the dotted net on snow at 20 m disappeared. It costs 0.381 ms full screen against 0.063 ms for the old method. Later the same day, all 67 ground layers got their settings measured from each texture itself instead of by hand, and a light version arrived that costs 35 to 40 % less and looks the same at 64 and 671 m. A flight from 1.5 m to 4 km of altitude showed no jumps.

No human marks (review branch). One snow texture turned out to contain tyre tracks. After that, 25 textures were vetoed (ruts, footprints, paths, quarries, construction) and 103 were reviewed by eye, and a check now rejects any palette that includes a vetoed or unreviewed texture.

Deposit halos where chemistry says so (review branch). From orbit, Ithera’s desert showed hundreds of identical beige discs: the stains drawn around ore deposits. Now only ores that alter the ground leave a mark (oxidising ores leave rust, salt leaves a white crust), and twelve inert families leave none. On Ithera, 9,871 of 34,818 deposit records still paint. The stain follows the data: a cap over the ore and a tongue that spreads downhill, longer on steep ground. Beyond 16 metres per pixel it fades out, so from 60 km and from orbit no halo is drawn.

Rocks that sit in the ground. A first physical seating law for rocks and resource nodes reached the main build at midnight. In the review branch, a second version buries loose stones deeper by size, slope and biome: on average, pebbles went from 36 % to 55 % of their height underground and rocks from 26 % to 44 %, with no stone floating. On airless bodies, blocks now pile up at the foot of crater walls, as they do on the Moon.

A lighter terrain (plan)

The development target for the terrain is clear: under 10 ms per frame and under 8 GB of video memory, at native resolution and with DLSS Quality, with no visible loss. A new tool counts, without a GPU, what each block of the terrain shader costs. The answers were blunt:

Measurement	Figure
Terrain shader, total	36,600 instructions and 600 texture reads
Share of the desert layer stack	63 %
Far version of the shader	Already folds down to 11,900 instructions (−67 %)
Terrain base pass at about 2 megapixels	4.9 to 12.8 ms
Proposed budget for that pass	2.5 ms
Each world’s own video memory	483 to 1,976 MiB
Editor session, whole scene	6.7 to 8.2 GB, with three worlds above 8 GB

To meet the budget, the terrain has to get two to six times cheaper, and small tweaks will not do it. The plan ranks eight proposals by milliseconds saved and risk: compiling the middle-distance tiers by footprint (in a static emulation, 39 % fewer instructions on the desert stack), sharing work the three near layers now repeat, removing six engine noise nodes that cost 48 reads per pixel and draw seams every 655.36 m, and, as the only order-of-magnitude lever, a virtual shading cache.

One saving already landed in the review branch. Seen from orbit, the scene requested about 2,000 distant sectors that were built in full only to end up empty. They are no longer requested, the image is identical, and the orbital view now settles in 84 seconds, where before it had not settled after 200.

A roadmap that tells the truth (main build)

Our public roadmap reads Jira every 15 minutes. On 13 October it looked as if nothing had moved since September, and the reason was ours: since 25 September, 1,933 of 2,304 commits carried the same catch-all task key, so Jira never moved. That night the work was assigned to its real tasks: 56 open tasks moved to done, 3 epics were closed and 9 tasks were created for work that had none, from the generation of the fifteen worlds to the faction stations.

So that it does not happen again, a script now runs on every commit and on a schedule. It leaves one comment per task and day with the commits that name it, and applies the done or in-progress marks written in the commit subject. Until its access key is installed, which is one manual step, nothing breaks: commits wait in a queue and are posted once it exists.

Also on 13 October

• The faction stations got their anchors (main build). Delvar and Trasvar, the trojans at Ithera’s L4 and L5 points, stopped being surface bodies and became anchors with no ground, with their faction station at the centre. Visiting them is not promised.
• Ten effects that had never fired (main build for the dust). Seven dust events and three weather effects (splashes, near dust and lightning flashes) were saved without their emitter and had never emitted a single particle. The generators were fixed. On Ashkar, the dust curtain of an ejecta event now rises in a parabola, with 2,500 particles in three seconds.
• Trees that move (main build). One family of trees ignored the wind; now it sways with the same rule as the rest.
• Three worlds regenerated after the review: Tirane, Pyra and Cinera.

What it means

None of this is a new feature for the alpha. It is the unglamorous part: making fifteen worlds look like places, measuring what each decision costs, and making sure the roadmap shows the work as it happens. The review of the worlds and the ground textures are still open tasks in Jira, and they will move on the roadmap as they close. How the system fits in the engine is in our guide to the Aurin system, and what we already learned about video memory is in the PC optimisation guide.

If you want to see it first-hand when it is ready, apply for the closed alpha: the first wave is planned for winter 2026-2027.