CartoVox Delta V7 — full release notes

Delta V7 is the seventh Delta release, shared with invited testers on 12 October 2026. Your Delta access code carries over, and 500 new codes join it. The licence is unchanged since Delta V6 (version 2.3). Everything below is new or fixed since Delta V6. Installers are available for Apple Silicon Macs, Windows x64 and Linux x64, and they are smaller again because the offline starter worlds are gone. Worlds made with Delta V6 open as they are. This build lays out water, lakes and climate differently, so the same seed rebuilt here gets different ground: Library's Update to this version rebuilds a world with this build when you choose to, keeping the old copy beside it.

Highlights

Your brief wins over the planet type. Create composes every world as land shape, then planet type, then your brief, then Advanced. Warm, Cold, Lush, Arid and the other graded choices move the world from where the shape and planet put it, and each control's middle keeps the world's own value. Describe says where every value came from.
A chosen card keeps its promise. Candidates are drawn from the same settings their world is built from, and the built world is checked against the card you chose. Every land shape checks for what makes it that shape, Best brief match puts the safest matches first, and a Prefer box sits beside Must have and Avoid.
Lakes settle and rivers run downhill. Lakes fill, spill and merge hollow by hollow, rivers follow the actual ground, erosion follows real catchments, and moisture is carried round the globe without losing water.
Seasons belong to the place. Climate follows the finished terrain and is read at each place's own height; southern places have southern seasons, Mediterranean means a dry summer, and monsoon winds turn with the seasons.
A river through a lake is one river, named above and below the lake and ranked by its whole basin. Held passes are real cols on every road over them.
History and trade follow the ground. Living History conserves food; famine, migration and expansion follow places people can reach; goods cost what their journey costs; every port has a sea lane.
Countries, states and counties on the map. Political divisions are ready by default, with States and Counties as map layers.
Play inside CartoVox. A native tabletop with a GM desk, tokens, fog, live sight, doors, turn order, dice and a player window.
Make, a fifth workspace, with three illustrated paths to a finished piece: a playable region, maps in a chapter, or finished map artwork.
Find any map layer from one place, keep favourite layers, save named map views and keep shortlists of places.
Closing during a build asks first, and no longer freezes the app.

World generation belongs in Build. Advanced settings keeps its map preview and changed settings, without a duplicate Generate action or build estimate. Step 3 Build holds the output choices, cost and generation checks.
More room across desktop workspaces. Atlas, Travel and Campaign keep zoom and fullscreen in compact map docks, freeing the old bottom control rows. Chronicle's optional timeline, Gantt and calendar lists fold under Event views, leaving the map and era controls easier to reach.
A prominent Library resume card. The large world preview remains the Library's focal point, with World details visible beside it.
One World Library layout at every size. Saved worlds always use the map-first card grid, including one to three worlds. The former wide rows and their letterboxed thumbnails are removed.
Cleaner writing and studio controls. Manuscript has one Trash action in its visible tool tabs, with the count retained. Scene Studio groups zoom and plan/isometric controls in one canvas bar. Settlement generation and variations sit in the header, with Cancel shown while a job is active. Deep Time avoids repeating its drawing action beside the closed editor. Tabletop shows distance options when the measuring tool needs them.
A simpler Create flow. The yellow Search candidates button sits beside Reset all in Describe's header, including Advanced. The Describe footer is removed to give the main panels more room. Step 3 Build opens directly to enter a seed without searching. Phones keep the header actions together with touch-sized buttons. Candidates holds search depth and controls for first and repeat searches. Recent searches keep up to ten candidate sets, their variations, choices and comparisons in this browser; Previous, Next and the search menu let you revisit them after an accidental repeat. Settings changes keep search available and clearly mark results found with earlier settings. Phones fold view options and recent searches to leave room for maps. Deep Time uses its top tab. The separate Setup guide is removed: world settings remain in Describe and Advanced, while Build keeps generation and atlas choices and adds changed-setting, expected-output and incompatibility reviews to Run details. Time and memory estimates still refresh there.
Create's route summary sits at the right edge. The selected starting path and purpose sit close beside Change at the toolbar's right margin. The redundant question-mark control is removed from this toolbar; build information remains in the Build menu. Narrow windows keep the text and controls in view.
A new Atlas Compass identity. CartoVox's logo combines a cartographic C, a folded gold atlas V and a compass star, with custom serif lettering. The welcome screen uses the full emblem and wordmark; the header, browser tab, installable web app and desktop icons use a simplified mark that stays clear at small sizes. Light and dark welcome variants match the app's themes.
A steady map readout. The coordinate and terrain-reading box keeps its width and reserved rows as the pointer moves, readings load or the pointer leaves the map. Long descriptions use an ellipsis, with the full reading available on hover. Narrow windows and enlarged text keep a stable layout. The distance bar has the first row to itself, and the coordinates or a tool's instruction ("Click the map where the hotspot goes · Escape cancels") start the second row whole, before the reading; in a fixed coordinate column those instructions were cut off.
A cleaner World map workspace, with familiar controls. Map panels use solid, consistent surfaces and quieter borders. The Tools menu has less visual clutter, active tools and fullscreen have clearer highlights, and viewing controls have inset keyboard-focus rings that remain visible at the map edge. Inspector readings gain subtle alternating row shading. Button positions, floating-toolbar placement and map viewing behaviour remain unchanged.
More room in Render Studio. The desktop panel is wider, with fewer nested borders and padding around art direction. The four task tabs, document history, preview and export actions keep their existing behaviour.
Quicker to scan while writing. Compendium sections and Manuscript's desktop tool tabs have matching icons beside their labels. Long Compendium paragraphs use a more comfortable reading width; maps and tables retain their space.
Travel and Campaign controls closer to their work. Journey map layers sit beside map viewing controls. Campaign's sidebar has direct links to Profile, Reveals and At the table, while Player exports groups the handout, vault, atlas and token-sheet actions beside Present player view. The row of links sits above the campaign form, which scrolls beneath it on its own, so no control is ever half-covered by the row. The journey map's "Map" label uses the map bar's own ink, readable in the light themes too.
Give scene and tabletop maps more space. Inspector and GM controls can collapse to widen the canvas, then reopen with current fields and selections intact. Choosing a Scene Studio task reveals its controls again. The player display never shows the GM controls toggle.
A clearer Make and preferences layout. Make's numbered steps sit in three cards on desktop and stack on phones. App settings has a wider panel and icons for its sections. Layout keeps search and its footer in reach while presets and panel choices share a scroll area, including when a preset preview is open.
The Layout drawer names the new Candidates and Chronicle panels. View options, Recent searches and Chronicle's Event views are listed as "Candidates: view options", "Candidates: recent searches" and "Chronicle: event views" instead of names made from their internal ids, and the States and Counties maps open through the map library like every other map.
Three new one-click themes. App settings now offers Abyssal Cartographer (deep ocean, cyan and gold), Ember Forge (charcoal, ember and gold), and Sage Observatory (parchment, moss and bronze), alongside CartoVox Dark, Parchment Light and Follow system. Your choice is saved and carried into the scene, settlement, guide and tabletop windows. Themes change interface colours, surfaces and subtle textures; workspace layouts, maps, exports and saved world data stay unchanged.
A compact theme picker. The six choices sit in two columns of three, with the preview, name and radio button together in each row. A highlighted edge marks the selected theme; narrow windows keep full names and previews without the extra palette descriptions. Choosing a built-in theme also dismisses an unfinished custom-theme preview, even when reselecting the current theme.
A more expressive, compact workspace. Illustrated tools, coloured icons and large map imagery give each workspace its own character. Default interface text and buttons are smaller again: 13 px body/control text, 12 px secondary text and 32/28 px control heights. Bundled Archivo gives workspace headings and Library titles a new look; narrative and manuscript typography retain their own controls. Text scaling and compact/comfortable density remain.
More room for the work. Editing rails, scrollable inspectors and compact action rows adapt to desktop, laptop and narrow windows in both themes. Enlarged text no longer pushes floating map tools over fixed zoom controls. Unavailable actions have clearer neutral styling, warning/error states remain distinct, and newly mounted controls become usable when loading finishes.
Visible search and clearer destinations. Search has a visible label; Cmd/Ctrl-K returns to it without clearing the current query. Scene maps and settlements are distinguished from Manuscript chapter scenes, with existing encounter, battlemap, interior and settlement search terms retained. Layout gives Advanced, Deep Time plate/crust and climate/orbit sections clearer names.
Consistent viewing controls. World, scene, tabletop, import and drawing views use Zoom out, Fit and Zoom in consistently. Scene and artwork editors show the zoom percentage and offer Fit width without marking the document changed. World's former Actual size/1:1 action is labelled Fit. Fullscreen toolbars stay readable in both themes, and compact drawers have named close buttons. Setting help also explains resource planning and regional rendering.

World measurements

Moisture transport no longer loses water numerically. Humidity and its named sources use local transfers weighted by spherical cell area, with conservative mixing, recycling and rainfall smoothing. Balanced transport with fixed air columns avoids artificial humidity compression; polar transport groups explicitly trade small longitudinal detail for manageable computation. Measured drift remains reported independently of ledger consistency. Rebuilding changes rainfall, lakes and ecology and costs more at Full resolution; existing saved arrays retain their values. Generator epoch 11 explains these changes before update.

Drawn rivers follow the delivered ground. A curved river could cut through a higher dry neighbour even when its routed cells descended. Shared river artwork now keeps safe bends and replaces only offending legs with their native course. Dry salt pans and protected ground remain subject to the height check. Routes across a pole split correctly in raster, vector and export views. Native river paths, flow and lake budgets are preserved; existing saved pictures need a redraw.
Moisture accounting shows numerical drift. Ledger closure is identified as bookkeeping consistency. Signed and gross transport drift are recorded by operation, with a separate review warning and explicit limits. The eddy mixing explanation is corrected; older worlds retain diagnostics from their actual solve.
The water-balance reading counts all water reaching the sea. World health compared land runoff with discharge at each drainage basin's single mouth and converted it with an Earth year. It missed water entering the sea beside those mouths, and it misread worlds whose year is not 365 days, so most worlds read about 70% instead of about 98%. It now sums every sea cell in the world's own year, and lake overflow is credited where it enters the sea. Settled lakes and closed basins evaporate the remaining 1-4%.
Lakes settle where water can stay, and rivers run downhill. Every hollow now fills until evaporation and seepage balance what flows into it, overflows into its neighbour when full, and merges with it at their shared rim. Dry land drains down the actual ground, so a river no longer climbs out of a basin, and water that cannot cover a whole map cell is no longer painted as a 1 cm lake. A desert hollow that evaporates everything it receives shows a salt flat instead. On the default-size test world this removes 190 uphill river steps and 512 phantom lakes, and that world now passes its saved-world checks, which it previously failed. The replacement seasonal water model leaves the settled lakes unchanged for a further year. Rebuilding a world may remove tiny lakes and move rivers, basins and ecology; worlds keep their previous lake checks until rebuilt.
Worlds with a lake joined at its brim build again. A full lake made of two hollows that meet only at ground exactly at its level could stop world generation ("standing water must be one connected surface") when another lake's overflow arrived across that join. It happened on a default-size small planet. The arriving overflow now ends at the lake's level; rivers and lakes elsewhere are unchanged.
Worlds with neighbouring lakes brimming at one saddle build again. When a lake overflowed into its neighbour while the neighbour was still filling, and both then filled to the same saddle, world generation could stop ("Channel routing contains a cycle"): the first lake's outflow was routed back into itself. It now crosses the saddle and leaves with its neighbour's outflow. It happened on Quick builds of the Thin air, cold desert planet; worlds that built before are unchanged.
Shallow standing water is drawn as wetland. Where a settled hollow's water spreads over most of a map cell without the depth to draw a lake, the cell is now marsh, swamp forest or bog for its climate instead of forest, grassland or bare alpine ground. Desert hollows remain salt flats. On the default-size test world 55 such cells change; rivers, lakes and climate do not. The cell inspector says the wetland stands on a closed hollow's floor.
Verified saved water observations. The explicit finite-water API can now save a separate lossless observation archive and reopen it for lake display, relief and graph export without replaying the simulation. Bounded file checks verify native water quantities, elapsed flow rates, graph paths and shoreline clipping; conflicting saves preserve existing files. Ocean surface level and numerical-source provenance are retained explicitly. This completes developer output persistence; ordinary world generation and remaining readers still need integration. Existing annual worlds remain unchanged.
Coherent elapsed water observations. An explicit production API can now derive a finite observation from a declared starting water state and elapsed time. Lake display, relief rendering and graph export use its same native depths, lake identities and historical channel records. Lossless public fields include time, budgets, fork links and provenance; unknown salinity and permanent lake status stay unknown. Existing annual worlds remain unchanged. Ordinary generation and the remaining readers still need adoption. Saved readers avoid source replay; interactive caching and full-size costs still need qualification.
Seasonal river display groundwork. The replacement water model now draws selected channels from actual elapsed flow, clips them at observed lake and ocean shores, and preserves the complete runoff and spill evidence separately. The atlas renderer accepts this display only with matching terrain, water and source context. Existing river rendering remains unchanged. Normal generation, public saved-world fields and dependent readers still need integration.
Native seasonal channel graph groundwork. The replacement water model exposes both downhill runoff branches and dry lake-spill tails as elapsed interval volumes. Reach paths preserve forks and joins, with flow values for individual edges and mean rates based on actual season durations. Hydraulic lake connections remain separate from dry rivers. Visible river selection now has a guarded renderer path; production readers still need integration, and hillslope flow is retained evidence.
Observed seasonal lake catalogue groundwork. The replacement water model now records actual lake cells, levels, areas and volumes at elapsed seasonal endpoints. Terrain identities follow parent joins and child splits; dry basins stay dry, and incomplete years retain missing seasons. These observations prepare shared readers without assigning salinity or permanent-lake labels.
Elapsed seasonal water recovery groundwork. Exact finite-water saves can recover the river, shoreline and spill records for elapsed substeps in the current year, or the preceding year at a complete boundary. Recovery checks the exact saved water state and budgets before returning results; partial years retain their missing seasons. This prepares the replacement model's readers and does not yet replace production lakes or public river maps.
Self-contained finite-water progress groundwork. The replacement water solver saves exact terrain, cell metrics and prepared water inputs alongside pool depths, elapsed seasons and cumulative budgets. It can restore the next step with the same water and future flow records without the original inputs or rounded display fields. Repeated identical forcing steps share storage; archive and metadata size checks prevent unreadable progress saves. Changed terrain, forcing or solver context rejects incompatible progress; failed or stale saves preserve the previous snapshot. Dry runoff is measured directly so rounded wet fractions cannot invent or erase water. These standalone checkpoints prepare production integration; generated lakes, the public river catalogue and their consumers still await the complete model switch.

Finite river-graph integration groundwork. The replacement water solver now records distributed runoff stopped at moving lake shores and dry spill tails stopped at receiving water, with cell budgets and actual spill timing. Native annual transient checks preserve the solver's water state exactly. This prepares the replacement graph; production lake/river integration and public saved-world integration remain in development.

Finite lake-storage integration groundwork. The replacement storage solver now accepts actual production rain-and-snowmelt depths, native season durations and runoff exclusions, with immutable source snapshots and budget checks. Frozen precipitation cannot sustain a liquid pool in the analytical reference. Production lakes and the public river graph still await integration; bounded transient runs are not labelled permanent or equilibrated worlds.

Lakes use rain and modeled snowmelt. Direct lake input now follows the shared rain/snow partition and sums water over the actual lake footprint. Frozen precipitation retained as snow no longer creates direct liquid outflow. New saves and GIS bundles retain annual/seasonal source depths; checks compare lake budgets with those native inputs. Rebuilding may change lake levels and downstream seasonal flow, and generator notices explain that change. Existing worlds stay unchanged. The annual lake-area approximation still needs finite storage and time integration; these source corrections do not complete it.

Aquifer recharge uses its actual water source. Aquifer reports now read infiltration-derived shallow recharge from the groundwater solver, replacing the unrelated quarter-of-runoff estimate. New saves and GIS bundles retain annual recharge and baseflow, with checks that reports, renewal classifications and graph provenance agree. Maps and settlement readings use the corrected reports. Baseflow is already included in runoff; renewal remains a model interpretation rather than measured fossil age or a pumping allowance. Existing saved worlds are preserved, and inconsistent exports leave previous downloads intact.

Height-datum migration history stays with the world. Saved checks now verify recorded author-height translations and the world's sea-zero settings. Scientific/GIS bundles include the numeric history while removing private edit IDs and author-edit fingerprints. Completion records detect a changed, missing or added report, and revision checkpoints keep its exact local copy. Later edits do not falsely invalidate historical evidence. Older completion records that omitted an existing migration report need a newly verified build; existing reports and worlds are preserved.

Erosion follows real catchments. Incision and floodplain build-up now use contributing area routed over actual ground, stopping at pits rather than escaping across an imaginary filled surface. The Erosion Lab uses the same catchment and ground-slope calculations. Rebuilding can change valleys and floodplains; this remains a terrain-shaping approximation, without a sediment budget or timed lake spill simulation.

Ground processes follow real slopes. Erosion, infiltration, wetlands, geology and settlement placement now use actual terrain descent. A filled routing surface no longer makes sloping ground inside a basin seem flat. New saves retain that measurement for independent checks, GIS export and the cell inspector, where underwater readings are labelled bed slope. Rebuilding can change local terrain, water and ecology; saved-stage reuse carries the new fields through story and realm edits. River and lake physics remain under development.

Unfinished drainage repairs retain their basins. A proposed fill or cut no longer counts as completed terrain work. Remaining depressions keep their water budgets, and rebuilding recalculates affected lakes, rivers, erosion, ecology and settlements. Saved checks compare basin footprints and routing potential with actual ground; older worlds remain readable. Compact test worlds lose their dry uphill river edges, while full-resolution river routing and time-resolved lake physics remain under development.

Drainage tries another bounded escape. If the first route cannot be cut within the existing depth, length or protection limits, repair searches for a nearby descending route. Accepted corridors remain open, and new saves record the chosen method for independent checks. Original protection and ocean heights stay intact. Rebuilding may change local ground and climate. This reduces contradictions in the full saved-terrain routing trial, but some cuts expose further unresolved basin routing; river and lake physics remain under development and strict scientific checks continue to fail them.

Automatic drainage cuts keep descending. Cuts now follow lower pockets along an escape instead of returning to the starting pit's higher level. Later fills cannot close the accepted path. New saves retain each path for independent height, neighbour and length checks; version-1 saves remain readable. The inspector shows terrain changes for both versions. Rebuilding can change local ground and its reconciled climate, so the generator notice records that change. Rejected cuts leave unresolved basins; this does not remove the remaining dry uphill rivers or qualify the lake model.

River contradictions are easier to inspect. Health and saved-world validation now check every recorded dry river edge, so a local climb cannot hide behind a reach's overall descent or a depression label. Findings show the climb and link to ground measurements at its location. Older health reports gain the check without rewriting their recorded findings. Intact worlds remain saveable with explicit physical errors in their build receipt; scientific validation and release checks still fail those contradictions. This exposes remaining uphill rivers; it does not claim to repair them.

Pressure readings use consistent units. Weather Year and Climate Lab now call the signed seasonal values "Pressure anomaly" in hPa. Map titles and saved-field descriptions agree; these values are departures, not absolute surface pressure. Existing climate arrays and Weather Year drawings stay unchanged, and old reader revisions refresh when reopened.
UV maps follow seasonal snow and rainfall. New worlds save a seasonal snow-reflection proxy, so snow that melts before year-end still affects the relevant seasons, and year-end storage no longer represents every season. Unequal season lengths now preserve the annual rainfall used for cloud; bounded sampling avoids erasing valid rain patterns when enlarging compressed climate fields. Map notes and export provenance show the reading's actual source fields. Older worlds explicitly retain the end-storage snow proxy until rebuilt; their UV drawings still receive the rainfall-reading fix.
Snow no longer loses water at an arbitrary storage limit. Seasonal melt uses a defined starting snow store instead of ten warm-up years and a 20,000 mm cap. When snowfall exceeds melting, the surplus remains in storage and is shown separately as annual accumulation. Inspectors and place reports distinguish end-of-year storage from winter snow or glacier age. Existing worlds keep their saved readings until explicitly rebuilt.
Scientific maps keep their evidence when enlarged. Higher drawing density no longer creates negative fertility/productivity or values above native maxima. Missing readings stay missing, and coastal readings do not mix with values outside their land/ocean domain. Scientific coastlines, water masks, resource winners and threshold classes keep their native footprint.
Consistent scenarios and reading keys. Sea-level scenarios and seasonal temperature ranges use native observations before enlargement. Automatic colour scales use the same native bounds as their legends. Unknown rainfall, resource prospects and aquifer measurements no longer produce misleading colours. Relief artwork retains its separate smoothing rules; higher drawing density does not add simulation detail. Cached drawings refresh automatically.

Marine boundaries respect the globe. Marine temperature mixing, coast proximity and fisheries plumes no longer connect the north edge to the south edge. Nutrient plumes also stop at intervening land. Existing saved numerical worlds can redraw these derived maps; cached drawings refresh automatically.
Honest ocean-current readings. Climate Lab, place dossiers, cell inspection and scientific exports identify relative current strength rather than claiming a calibrated speed in metres per second. Export and API metadata explain the normalization. Circulation values and wind speed units are unchanged.
Ordinary towns join the road network. Generated roads now include local town tracks while retaining links between major cities. Reachable groups connect within the configured endpoint-distance limit. Ground barriers and authored removals are respected. Travel, route exports and history use the same saved paths. Existing worlds gain the network when rebuilt.
Sea lanes consider reachable ports first. Nearby ports on a separate waterway, or connections removed by the author, no longer hide a farther valid destination. Land and lake barriers, isolated zones and the configured distance limit remain effective. Existing worlds gain this correction when rebuilt; it does not imply measured shipping traffic or capacity.
Accurate town-access diagnostics. Connectivity reports now count named towns without recorded links instead of displaying an uncomputed zero. This reports the actual network; it does not create missing roads or imply that cross-country travel is impossible.
Settlement shores agree with the saved terrain. Low coastal ground keeps its actual fractional height when measuring the shoreline, matching Render Studio's zero-elevation contour in the checked cases. High-latitude shore checks now use the actual neighboring grid cells; water several columns away no longer counts as touching the town. Recorded ports retain their separate harbor treatment. New plan metadata records the corrected reading version. Existing saved plans keep their geometry, and numerical worlds need no rebuild for a fresh plan.

Climate follows the finished terrain. After glacial and river erosion, generation recalculates climate on the final heights and routes the resulting water without eroding the land a second time. Biomes, resources and settlements then read that reconciled state. Painted climates and supplied ground remain effective. This adds calculation time, and rebuilding can change generated settlements and borders as they respond to the corrected climate.
Recorded terrain/climate agreement. New numerical saves include evidence linking final terrain, temperature and rainfall. Publication checks it against the archive and summary; scientific exports include the report. Older saves remain readable without inventing evidence for their earlier calculations.
Lakes conserve their routed water. Every preserved basin, including small lakes, participates in flow accounting. Inflow counts all tributaries; outflow follows both downstream branches without replacing other water at the outlet. Interleaved basins are intercepted before their budgets are solved. New saves carry independently checked lake budgets and discharge evidence; older saves remain readable and need re-simulation for the corrected results.
Drainage repairs can change the visible ground. Small eligible hollows are filled and shallow obstructions are cut within declared limits before the final climate and water solve. Supplied, kept and built ground is protected. New saves retain before/after evidence, and the cell inspector shows local changes and protection. Remaining uphill channel segments are reported explicitly; the bounded repair does not resolve every depression.
Terrain evidence survives cached rebuilds. Exact height differences keep their archive precision through checkpoints. Publication checks their saved surfaces, protection, bounds and volumes, and rejects missing or unsupported repair evidence. Older worlds remain readable without invented proof.
Lake maps and descriptions share the same meaning. Small freshwater lakes are no longer labelled salt pans or used as dust sources. Size alone no longer implies seasonal drying. Lake dossiers use the saved lake identity instead of borrowing facts from another lake of similar area, and distinguish a recorded outlet from a proven route to the sea. Inspectors label the model's limits.
Safer cached rebuilds. Saved ground retains the inputs needed by water and population accounting. If a late cache miss requires full generation, the rebuild continues its existing publication rather than failing at a second commit point.
Regional measurements respect the sphere. Continent temperatures, rainfall and heights, desert measurements, forest canopy and realm biome shares use the ground area each map cell represents. Realm lake areas use their actual locations; older saves without that measurement say the area was not recorded. The Worldbook, realm handbooks and Compendium share these saved measurements.
Lore geography uses the world's size and borders. Feature distances, story hooks and author questions use the planet's configured radius. Opposite polar realms no longer become neighbours through map wrapping, and narrow territories remain visible when determining shared borders.
Stronger numerical save checks. New worlds declare a complete archive format. Before replacing a numerical save, generation checks required fields, dimensions, data types, category IDs and finite values, including seasonal frames. The saved-world validator distinguishes missing modern provenance from older formats and reports corrupt data without crashing.
Honest measurement labels. The cell inspector labels melt input divided by runoff as a ratio, which may exceed one after water losses. Scientific exports explain that ratio, boundary intensity's actual range and the direction of the aridity index.
Interrupted builds cannot pass as finished worlds. Generation validates the saved world and records fingerprints of its completed maps, data and generated documents. After a force-close, startup verifies that record instead of relying on an older revision. Partly written worlds stay marked incomplete; a completed build that missed its last revision can be recovered safely.
Rebuild unfinished worlds from the Library. A compact recovery action uses recorded settings and applied edits while preserving writing and notes. Missing source maps or invalid settings are explained before rebuilding. Cancelling an incomplete world's recovery no longer incorrectly marks it ready.
Configuration migrations report rejected values. Settings loading and its migration report share the same conversion rules, so invalid values cannot silently fall back to defaults while being reported as unchanged.
Disabled exports cannot leave an older world live. Rebuilding without the optional numerical archive or JSON summary moves an earlier copy into recoverable safety storage. The completed-build check rejects leftover stale files. Author notes remain untouched, and safety copies stay in backups.
Mediterranean means a dry summer. Climate classes apply the full Köppen–Geiger summer-dry test: the driest summer month is under 40 mm, under a third of the wettest winter month and drier than every winter month, and the summer season as a whole is drier than winter. One dry shoulder month used to be enough, so monsoon coasts with rainy summers read "Mediterranean" (half of the default test world's Cs/Ds area had wetter summers than winters). Settlements classed as Mediterranean there fell from 87 to 8. Winter-dry climates use the same paired test.
Season temperatures belong to the place. Seasonal frames are saved at half resolution, so a valley town shared its season reading with the mountain beside it, and season tables could sit degrees outside the town's own coldest and warmest months. Dossier season tables, settlement context and the travel planner now move each reading from the frame's average height to the place's own with the world's lapse rate. On the default test world 77 settlements had a season colder than their coldest month or warmer than their warmest; none do now.

World evidence and lore

Rivers and lakes are counted as the water itself runs. Two rivers that reach the sea at the same spot are now two rivers; one used to vanish into the other, keeping only the larger basin. Branches that end in a lake which overflows to the sea now count as reaching the sea (the default test world read 77% of branches reaching the ocean; it is about 98%), and every lake is counted as open or closed. A lake whose water touches only corner to corner is one lake with one name, not two. The Worldbook's lake count and table follow the same lakes, and its salt-pan figure says how much is lakes that dry to salt and how much is dry desert floors. River reach exports add the discharge at each reach's end and say that their mean is the largest along the reach. The Worldbook's river table and each river's dossier say whether it ends at the sea, in a named lake (which may overflow on as another river) or in a closed basin, instead of giving a lake shore as its mouth.
A river that runs through a lake is one river. The river feeding a lake and the river leaving it now share one name and number, measured from the farthest source across its lakes. On the default test world the largest river reads 3,771 km instead of 1,023 km, the river that read "174 km" is now its upper course, and 2 of the twenty largest rivers end in a lake instead of 7. Rivers above those lakes give up their own names (19 of 60); every other name stays.
River pages say where a river rises and which lakes it runs through. The Worldbook table gains a "runs through" column; a river page gives its source, its lakes (linked when named) and, for a river ending in a lake, the river it drains on as. The inspector and Why here? name the river a channel belongs to above a lake as below it, exported river reaches carry that name, and town rivers and map lettering follow the river's own course.
Your river names survive the join. Rebuilding keeps every river's number and name; a name you gave a river that is now another river's upper course is held under Released in the rebuild review, never silently dropped.
A river through a lake is ranked by its whole basin. A river whose lake spills straight to its end was ranked by the stretch above the lake; it now ranks, and is numbered, by the whole river. The help for year and day length now lists the water stage among their readers, since each river's mean flow at its end is worked out from the world's own year.
Realms see the river they stand by. A capital's nearby river is measured to the river's course, not its mouth: 26 of 42 realms on the compact test world name one (was 18), and a capital on its river is said to stand on it.
A river is named above its lakes too. Atlas plates, the vector sheet and the SVG export set the name again above and below a lake where there is room, after every other name, and the lettering list offers those stretches. A border river's far-bank name is judged along its whole course, and vector sheets read the river data once instead of on every export.
Exports and dossiers report the measure they name. Capitals in the GeoJSON export carry their saved height instead of 100 m for every capital. An ocean's dossier gives its farthest distance from land instead of calling that distance an area; an enclosed sea's longest span is its greatest extent, not its shoreline. Lake kinds use one name everywhere. The Worldbook lists ranges as largest first, says when the island table is cut short, and names the largest rivers with deltas. The codex lists the named continents by the Worldbook's rule.
Held passes are real cols. A keep now marks a road's crossing only where land rises on both sides of the road. A road's high point on a hillside or an open crest is no longer called a pass. The default test world keeps 18 of its 27 keeps. Each pass belongs to the range whose high ground flanks it, and the Worldbook's pass table names that range; range dossiers and realm prose count those passes instead of guessing from distance or assuming "a small number". Realm prose no longer calls unmapped wells strategic, and the Worldbook measures each climb from the road's lower approach.
Held passes say whether they can be gone round. Each pass is closed, with the ground across its road for 120 km either side, and the cheapest overland route between the road's approaches is found on the same terrain the roads follow. A pass is the only practical crossing when no way round costs less than twice as much, easily bypassed when one costs at most a quarter more, and otherwise has a dearer alternative. On the default test world 1 of 18 passes is the only practical crossing, 6 have a dearer alternative (32–95% more) and 11 are easily bypassed (5–24% more). The Worldbook's pass table gains an "if it is closed" column; generation takes about 0.1 s longer.
Realm prose calls a pass a chokepoint only when it is one. Realms cited every nearby held pass as a weakness. A pass is now named the only practical crossing only where that was measured, a dearer way round is given with its cost, and passes that are easily bypassed are said to be. Range dossiers and the map inspector show each pass's crossing; worlds saved before this read "not measured".
A held pass is measured on every road over it. Only the road it was found on was tested; now every road through the col is, and the pass reads as the one that is hardest to go round. On the default test world one pass becomes the only practical crossing for one of its five roads, and another costs up to 53% more to bypass than before.
Scientific downloads explain their measurements. Global summaries use spherical cell areas and state their scope and missing coverage. Category IDs and circular bearings are no longer averaged. Receiver indices stay exact integers in TIFF exports; the original numerical archive remains unchanged.
Harbours, plans and summits agree with the map. A capital whose sea lane leaves from a distant shore says so instead of being silently "not a port" with passages. City plans keep the town's own port status and record separately whether the harbour falls on the sheet. A range's summit snow comes from the saved climate at its highest point, so a summit with a +9 °C warmest month no longer "carries snow all year".
Island harbours get a sea lane. Generated sea lanes joined only the founding and secondary cities, so a harbour town alone on its island had no recorded route anywhere. Each road network that no lane reaches now gets one lane from its best harbour, within the usual lane reach and respecting removed lanes and cut-off zones. On the default-size test world every one of the 12 unlinked places is now connected, and 16 separate networks become 2.
Dossiers credit only worked ground. "The ground gives…" named the mineral province under a town even when no worked deposit supplied its market; 228 of 274 such claims on the default test world did not match that settlement's market. Dossiers now say what the ground could give, with its prospect, and that no worked deposit supplies the market yet. A place without a dominant prospect names its strongest family instead of quoting a contradictory percentage, and farming crops come from "the farming cradle whose reach covers this place" rather than a mislabelled nearest one.
Roads out tell land from sea. A road was dropped from a place's routes whenever the same pair also had a sea lane, and travel times offered a ship where no lane existed and days on foot between separate landmasses. Every route is listed now; a ship appears only with a lane, overland times follow the road or railway's own length, and no overland time is given across sea.
Southern places have southern seasons. River flow notes, the "river that roars" detail and Foundry season labels named northern seasons for southern places; they use the local season, as the season table does. The season-card note says daylight is measured on the solstice or equinox that names the season, not on the quarter's longest day.
Winds that turn with the seasons. Prevailing wind was the year's average vector, which cancels where monsoon winds reverse, so places with 7 m/s seasonal winds read "light air". Where that average is under half the seasonal speed, dossiers now say the wind turns with the seasons and describe it by its seasonal strength.
Open country is not closed canopy. Tree cover in savanna and other open biomes is described in that biome's terms (well-wooded savanna, scattered trees) instead of as a closed forest canopy.
One storm vocabulary. Dossiers and the storm hazard map share their thresholds for frequent, common and occasional storms.
Region dossiers say when they read one cell. Continent, desert and physiographic-region dossiers that follow an area average with the centre cell's climate now say so. Continent land cover adds each biome's actual area from the archive instead of counting each region wholly as its dominant biome.
Only kept features are counted. Names were made for early river, mountain and desert candidates that later passes dropped (60 rivers named, 20 kept on the default test world). World dossier counts and the Codex's Tongues page use only the features the map publishes, and continent counts use the map's own minimum area.
A monsoon delta has both. A culture is a monsoon-delta people only when a delta the hydrology built lies within reach and its rain comes in the warm season. A seasonal river mouth in a dry-summer or desert climate used to qualify. Founding cradles are picked one per kind, so where a false monsoon delta founded a realm, rebuilding moves that cradle and nearby borders: on the compact test world (seed 576601) two of seven cradles move, monsoon-delta towns fall from 12 to 2 and 42 realms form instead of 45. Ground, water and climate are unchanged.
Living History starts where it started. The starting capital, and every town's rank in the dated-state reader, came from the end of the run; both are rebuilt from each town's own history at the year asked. Realms no longer come to the aid of their own rivals.
Smaller agreements. A culture within 120 km of the sea is coastal in both its qualifier and its situation line. The settlement catalogue lists each place once (441 rows for 441 places, was 446) and travel catalogue rows carry their realm. Interpretation goods follow the realm's current government alternatives. Chronicle panels link realms by their own id, and Chronicle event types match whole words, so "Warwick" is no longer a war.
Every realm's capital is a capital everywhere. Most realms are seated in a regional centre or a town, but the settlement builder, Compendium, city plans, GeoJSON export and lore map treated only the founding hearths as capitals: those capitals were built with no citadel, their pages said "Secondary centre" or were missing, and the export listed 7 capitals for 38 realms on the default test world. All 38 now read as capitals, and each town a realm is seated in has a Compendium page its realm links to.
A famine is dated where it struck. A small island famine could absorb a later mainland one and date it to the island's first lean year. A famine now runs on only while most of its hunger stays where it was, and describes all the ground it covered. On the default test world misdated famines fall from 12 to 1 and mismatched descriptions from 84 to 0, and abandoned towns cite only the famines that reached them (13 of 31 corrected).
A harbour closes with its last port town. In Living History an abandoned or sacked port no longer keeps "with a harbour", its trade or its plague harbour count (14 provinces on the default test world).
History says what it counts. Land shares say they exclude lakes, a town founded during the run grows from its founding rather than from year 0, turnover counts realms founded during the run, and plagues count routes, not map steps.
A famine opens a realm to war only as far as it hurt. Any recent famine made war on a realm 2.5 times as likely, even one that killed a single province's handful. The opening now grows with the famine's toll on the whole realm, and the war says how much it counted: over ten runs of the default test world, wars owed to famines fall from about 90 to 28. A famine that killed under 0.1% of the realm in its worst year no longer opens it at all, and no war names it: over ten runs, wars naming a famine fall from about 99 to 52 per run, with no other cause invented in its place. Such a famine costs the realm no stability either and is no longer listed among the shocks behind a collapse or breakaway; its entry says stability held.
Unclaimed land starts with food in store. Unclaimed provinces began Living History with empty granaries and starved in the first lean year (99.8% of year 0's hunger on the default test world). Every province now starts with a quarter-year of food.
New towns earn their growth. A town founded during a run became a regional centre once its province reached 1.35 times its year-0 people, which its settlers alone usually achieved. It now grows from its founding: growth events on the default test world fall from 117 to 28.
Realm shares leave lakes out everywhere. Lore, dossiers, handbooks, the Compendium, the World data chart and the Worldbook count a realm's share of land outside lakes, as Living History does; 10 of 38 realms on the default test world move by 0.1 point.
Famines say what made them one. A famine states the one-province rule behind it, a war names the famine that helped open it, and the history's health report says how many hunger deaths the famines tell.
Hunger on unclaimed ground is told. People starving where no realm holds the land died without an event; each decade with such hunger now has a "Hunger on unclaimed ground" event giving the dead, where and when, and the health report checks that every hunger death is told by an event or counted as untold (untold deaths fall from 16.6% to 2.3% on the compact test world).
Living History tells lean years. When a realm's harvest falls short everywhere without any province losing 1.5% of its people, yet the realm still loses 0.1% or more, the run records a "Lean year in …" event with the food, the deaths and the worst-hit province, and collapses and secessions name it among their shocks (hunger told by no event falls from about 1.9 million to 0.04 million per run on the compact test worlds).
Abandoned towns blame only what hit them. A famine is named for a town's abandonment only where it killed at least 1.5% of the town's own province in a year, and a plague only if the province belonged to the realm when it struck.
Dossiers no longer call farming towns herders. "Farming 5% potential — herding and gathering" measured the wild crops a first farming people could domesticate, not how well fields grow; 184 of 441 settlements read "herding", most with good growing seasons. It is now labelled Wild crop package.
A Harbour Compact needs a harbour. Realms with no port on their ground are no longer offered one: 3 of 38 realms on the default test world, 10 of 42 on the compact one.
Specialty descriptions fit their towns. No more "monsoon hills" where no monsoon was asked about; highlands, uplands and plains follow the town's height, and coastal wording needs a port. 18 of 19 contradicting lines are gone on the compact test world.
Towns trade where their roads lead. Towns used to join their realm's nearest market as the crow flies, even across a sea with no lane. They now join the market they reach most cheaply by road, sea lane or railway, their own realm's first, and a town no route brings within 5,000 km of a market says so. On the default test world towns beyond that reach fell from 15 to 2, and every town on another landmass now reaches its market by a sea lane.
Goods cost what their journey costs. Anything a market's towns made was sold as cheap and local, even from 26,000 road-km away. Each good is now priced along the route to the nearest town that makes it, so 119 such goods on the default test world no longer read as cheap; a good whose maker has no route reads as unknown.
Bronze where copper meets tin. Bronze needed two market towns within 1,500 road-km of each other; it now needs the copper and tin workings themselves within reach, so mines 1,129 km apart make bronze even when their markets are 2,217 km apart (four markets on the compact test world).
Clearer market and scenario notes. Town pages say how a town's goods reach its market and how far, market resource rows say how many towns their goods come from, and the Bronze and Iron Age maps say they count every deposit, worked or not.
Bought goods take the direct road. A good from another market was priced to that market's town and back out to its maker, sometimes doubling back. It now follows the route straight to the nearest town that makes it: 185 overpriced goods on the default test world are gone.
Furs need a winter, pulp needs a chemical. Furs come only where the coldest month freezes, not from 437 of 441 towns, and wood pulp needs sulphur, lime or potash as well as timber, so paper is no longer at every market.
Market goods lists show what sets each market apart. Its famous good comes first, then what few markets hold; shared staples are summarised. Forest towns no longer say the land gives charcoal.
Goods rows say whose they are. A market's row names what its seat itself makes. A realm's exports come from its own ground; what towns outside its borders bring is listed as passed on, with those towns named.
Sea-ruled realms keep a fleet they can crew. A realm whose government rests on ships (Merchant League, Admiralty Republic, Thalassocratic Court) keeps one squadron whenever its whole army budget can crew it (on the default test world 5 of 5 now have a fleet, was 2), and a realm that still cannot says why.
Armies stand where the people are. Regiments are spread over a realm's provinces by how many people live there, not by area: realms whose largest army was outside their most populous province fall from 11 to 0 on the default test world. A realm too small for provinces says its armies stand at the capital (in a larger one, at the capital and its next town of note).
Every port has a sea lane. A port the generated lanes passed by gets one lane to the nearest harbour already on the sea network, so ships call where the dossier says they do (ports with no lane fall from 112 to 18 on the compact test world); every earlier lane is kept exactly, no price gets dearer and many along the coast get cheaper. A port on a pocket of sea no other harbour shares reads "a sheltered harbour, but no recorded sea lane calls here" instead of "a working port; ships call here".
Sea lanes no longer sail round a continent to reach a neighbour. Every generated lane, between founding cities, island feeders or ports, sails at most twice the straight distance between its quays, or goes to the next nearest harbour instead: the longest default-world lane falls from 10,229 km to 4,664 km. A port whose only nearby sea is shared with shore towns gets a lane to one of them.
A town on a scrap of sea cut off from the ocean is no longer a port. It stays a shore town, without pearls, a fleet station or a harbour line in its dossier (15 and 20 such ports on the two test worlds), so every remaining port ends a sea lane.
Inland markets no longer sell what the sea gives. An inland market counted pearls, coral, whale oil and salt fish among the goods it collects (17 of 29 inland markets on the compact test world). A coastal town's catch now lands at the coastal market it reaches most cheaply, and an inland market buys it from there along its own road ("from Pearlhithe, which lands its catch at Seagate"); the town's dossier says where its catch goes. Where no coastal market is near enough to land a town's catch, it is landed at the town itself and carried inland: its own market, and any other that reaches the town by road, buys it there ("landed at Pearlhithe and carried inland"), still never counted as collected at an inland seat, and a realm lists what its own towns carry inland among its possible exports. A catch is said to be not traded only when no market reaches its town by any recorded route, and every output gives that reason (one town on the compact test world, down from 13 across the two test worlds).
Crafts and bronze take the road to where they are made. A craft or bronze bought from another market is priced along the route to its workings, not to that market's seat and back.
One supplier, one name. The codex and a market's dossier name the same outside supplier: the town that makes the good, then the market it trades at. A specialty named in the town's own tongue reads as one name in lists ("Skjoldeford (the cacao of Skjoldanes)"), and a market with one town no longer says "1 towns".
Settlements too cold to farm are no longer farmland. A settlement's character ignored temperature: a regional centre on tundra at 81.8° S, warmest month 3 °C, was a "rain-fed agrarian heartland". Where no month reaches 10 °C a settlement is now a cold-margin people of hunters and herders (unless it is a harbour, an upland or dry country), and its culture, suggested governments, goods and market say so. One place changes on the compact test world and none on the default world; on a Dim Star world 33 of 441 do. No cradle or border moves, and ground, water and climate are unchanged.
A borrowed culture says how the realm lives. Most realms borrow the culture of a founding hearth, so a tundra realm of a river-valley people read only "river-valley farmers". Where a realm is seated in other country than its culture's hearth, its dossiers add how its people live there (for example "in this realm they live as hunters and herders of the cold margin").
Cold towns keep yards, not gardens. A cold-margin town's city plan has no Garden Ward and no fields along its roads; its outer ward is The Yards, and the sheet says it is too cold to farm.
Cold towns grow no crops. A town too cold to farm, by its warmest month or because the map draws it as tundra, polar desert or ice, no longer sells grain, beer, potato spirit or mead from warmer ground nearby; herds, furs and wild berries stay. On a Dim Star world 13 towns change.
Tundra is never farmland. A town the map draws as tundra is no longer a farming people even when its own warmest month is just above 10 °C, and its dossier explains the difference.
City plans describe the ground they stand on. Tundra and icy towns of any kind have no gardens or fields, and their sheet no longer says "the woods press close"; ground above the tree line says so instead.
Distances to the sea are measured to the shore. A town on the coast says "on the coast" instead of 33–69 km, and cultures, coats of arms and the Planet Advisor share one coastal line.
The settlement builder sizes fields by how well they grow. It uses the best growing season and draws no fields where it is too cold, and the founder-crop package is labelled as such in the builder and dossiers.
The biome map keeps to the polar line. Cleaning up stray cells and small deserts no longer draws tundra or ice where a month reaches 10 °C, or forest and desert where none does, so the map, the dossier and "too cold to farm" agree. 93 to 159 cells change on the test worlds; ground, water and climate are unchanged.
The Agricultural calendar grows nothing where it is too cold to farm. Where no month reaches 10 °C, or the map draws tundra, polar desert or ice, no season is a growing window, matching the town's kind, market and plan; Living History food follows it.
A cool spring no longer counts as a growing season. The Agricultural calendar, town dossiers and Living History's harvests need a season of at least 10 °C, the line below which a town is too cold to farm (seasons scoring a window below it fall from 3,414 to 141 on the compact test world). Each place's seasons are read at its own height, so a valley beside a range no longer takes the range's cold, and a town's season table matches the map.
Refined child maps keep to the polar line too. A refined region no longer draws tundra or ice where its own warmest month reaches 10 °C, or forest and desert where none does (1.8% to 4.1% of fine land cells on the test worlds were on the wrong side). Saved refinements re-simulate once.
Towns whose only wet seasons are too cold no longer farm. Where a town's rainy seasons are all below 10 °C and its warm season is too dry, the calendar gives it no growing window; its kind, market, city plan, town builder and dossier now agree (on the Dim Star test world eleven farming towns become cold-margin herders and five markets stop selling grain grown there). On a refined child map the builder decides a town's fields from the town itself, so finer ground drawn a little colder no longer takes them away.
The Köppen and storm maps read each season at the cell's own height. About 120 cells per test world change Köppen code, closer to the climate the world solved, and the thunderstorm, blizzard and fog maps agree better with it.
Every map reads the seasons the same, more accurate way. The calendar, food, Köppen, storm and snow maps, a town's season table, the town builder and the travel planner take each season's fine detail from the annual temperature, as the snow map did: about twice as close to what the climate solved (season samples on the wrong side of 10 °C fall from about 200 to about 100 per test world), and quicker to read.
Desert towns on reliable rivers farm their banks. Within 35 km of a river that flows all year and still carries at least 10 m³/s in its driest season (enough for a town's fields, however small its catchment), or of a fresh lake, the Agricultural calendar waters any season short of rain: a Nile-like valley gets growing windows, Living History feeds it, and its market keeps the rice and dates the river grows. The calendar, a town's "too dry to farm" verdict and the crops its market sells read this one test, so big rivers that run dry in a season no longer water fields, and paddy, cane, cotton and dates no longer grow beside seasonal rivers or salt lakes.
A town too dry to farm says so everywhere. Where no season can ripen a crop and no river or lake is near enough to water fields, the town is a herding town: its market lists no crops, its plan and town builder draw no fields, and its dossier says why.
The vegetation map reads each season's rain as the calendar does. It no longer invents a rainless season beside a wet one, so a few dozen broadleaf forests per world are correctly evergreen.
Monsoon deltas use the coastal line. A delta town must be within 60 km of the shore, like every other coastal reading.
Route tools share the saved transport record. Roads, sea lanes and railways remain available without an illustrated atlas. Names and endpoints agree with lore, GIS and market evidence; open ends stay unassigned. Older inferred endpoints are identified as such. Connectivity checks include towns and rail, explaining disconnected networks without inventing links.
Mineral access follows complete road segments. Access estimates use native route cells, including drawn roads and seam crossings. Removed roads cannot reappear from an older vector map. Revised rules invalidate dependent caches; older goods evidence remains readable and is marked for recalculation.
History detects changed starting conditions. Rainfall, resources, seasonal fields, planet size, settlement details, diplomacy, routes and naming choices now participate in the history run's input check. An older run cannot resume or be accepted against changed inputs. Cosmetic notes and atlas redraws do not invalidate it. Earlier logs and author edits remain readable; older incomplete input proofs require a fresh reviewed run for current-state acceptance.
History follows recorded routes. Complete native roads and railways now participate in the history network. Sea crossings use recorded landings; nearby ports no longer invent a connection, and long mapped crossings are retained. Open or unrepresented ends are explained. Changed route access can change a newly run history; existing saved logs and author writing remain.
Political reach follows controlled paths. Capital influence uses paths through the realm's own provinces, with the length of recorded routes and their local access. Foreign corridors cannot grant a shortcut, and an isolated province cannot be assigned to a new capital merely because it is nearby. Disconnected residents can still defend locally. Starting conditions and event causes explain the scenario costs and limits; these are not Travel itineraries.
Border explanations separate facts from causes. The inspector measures distance on the world's sphere, including poles and map-seam crossings. Suitability is measured from saved cells instead of inferred from capacity. A river crossing no longer becomes a claim that the whole border follows a river, and initial ownership does not invent a story of how the border formed. Missing or changed terrain evidence is explained while saved events remain readable. Older coarse histories retain an explicit approximation notice.
Inspectable history assumptions. Starting conditions expose route coverage and distinguish suggested, authored and unverified political relationships. Event causes retain that provenance. History checks describe scenario ranges rather than claiming calibrated realism, and older model versions require a fresh reviewed run before current-state acceptance.
History keeps the world's actual ground. Small islands, narrow realms and lake boundaries survive in the history map at the saved world's resolution. Disconnected pieces and corner-only contacts no longer become one province. Area and capacity are summed from original cells, avoiding inflated coastlines and capacity estimates distorted by averaging first. The capacity formula and any missing-data assumptions are available under Starting conditions.
Detailed histories stay portable. Accepted histories store their parcel grid losslessly in a compact snapshot. Older snapshots remain readable, and a settlement's historical owner is read from its actual saved cell. These corrections can change new histories; previously accepted writing is retained.
Food is conserved through history. Reserves are actual stored quantities, with fixed provincial storage ceilings. Population loss, growth, a new crown or an author pin cannot silently resize the food stock. Yearly ledgers account for harvest, consumption, remaining stores and discarded surplus.
Crop supply follows growing conditions. New histories distinguish places people want to live from land that supports the scenario's crops. Food uses the Agricultural calendar's saved temperatures, soils, water and season lengths; harbour appeal and mineral prospects cannot create crop harvests. Residents can still use food carried over a connected realm's recorded routes. The assumptions and older saves' missing-data fallback appear in Starting conditions. This is a generic agrarian scenario; fisheries, foraging, livestock and technology-specific food systems remain outside it. Newly run histories can change substantially; existing accepted histories remain readable.
Famine and migration follow reachable places. Food pools within connected regions of one realm, using land and recorded crossings. An isolated island can face a shortage even when stores elsewhere are full. Refugees leave the affected region for a connected, fed destination with spare capacity; distant surpluses no longer invent a migration route. Unrelated island famines remain separate, and annual mortality percentages are no longer incorrectly added.
Starting populations use the same food rules as history. The agrarian reference now begins with residents its connected regions can feed in a normal year. Changing the story date or random variant no longer invents a different starting population. Starting conditions explains the calculation and its limits; saved settlements and author writing remain intact.
Expansion respects people, supply and space. Colonists come from connected territory and fit the receiving province's residence capacity. A move must leave enough reference-year food, including the labour lost at its source. Population peaks and growth diagnostics now use the correct population.
Army suggestions share history's population reference. Troops and naval crews now fit one declared personnel budget. Small realms no longer acquire oversized minimum regiments or uncrewed fleets, and rounding cannot create extra soldiers. Government and culture edits feed the next suggestion. Fleet crews are editable; older unrecorded crews stay explicitly unknown. The editor and realm pages explain the assumptions. Author formations can exceed the suggestion and survive when a new reference supports fewer people. These are scenario suggestions; equipment, recruitment, military labour and campaign supply are not simulated. Saved-world checks compare the population, military budgets, unit totals and author choices before declaring agreement.
Food assumptions and evidence stay inspectable. Starting conditions shows the scenario's food rules and latest balance. Conservation checks run alongside history diagnostics, and accepted snapshots retain their assumptions and food ledgers even after a new preview. These remain fictional scenario estimates, not calibrated yields, shipment quantities or historical mortality rates.
Descriptions follow mapped connections. Roads, sea lanes and railways in lore share endpoint identities, names and lengths with GIS exports. Open-ended drawings are not silently assigned to nearby capitals. A connection alone does not invent cargo, traffic or hazards.
Route geometry agrees across views. Vector routes use the same cell centres as their saved paths, and short crossings at the map seam retain both halves. Travel, map overlays, place dossiers and newly accepted history read native road, sea-lane and railway records even without an optional vector export. Dated routes retain their kind, and authored openings preserve their alignment and endpoint evidence through later map edits and closures.
Generated transport respects the ground. Roads follow the actual terrain cells, reach their named settlements and cannot cross ocean, lakes or blocked corners. Sea lanes stay on marine water with recorded shore approaches to their harbours. Narrow land bridges and map-edge cells survive routing, and drawn routes retain every bend and the author's intentional crossings. New publication checks paths against saved ground; legacy lines stay readable without being certified under the new routing model.
Journeys respect narrow passages. Travel uses the saved terrain cells, preserving narrow straits, islands and land corridors. Walking endpoints stay at their settlements, blocked corners cannot become shortcuts, and displayed routes retain every bend. Ship journeys record reachable harbour approaches separately and explain that those approaches are outside the sailing time.
Access maps respect the same ground. Sailing catchments and caravan access preserve narrow channels and land corridors, exclude lakes and blocked corners, and leave unreachable areas without a time estimate. Roads remain available to trade and flood-exposure maps when optional vector exports are absent. Port and resource influence now spreads by distance rather than a fixed number of pixels.
Decision-map values stay meaningful when enlarged. Drawing no longer invents negative access costs, out-of-range indices or missing coastal strips. Legends explain the maps' comparative cost model, its difference from Travel itineraries, and missing lake information in older saves. Native calculations take longer on large maps; existing cached drawings remain reusable.
Travel estimates use the same journey. Fastest follows the actual movement costs, including current assistance and fast roads. Road and hazard shares measure distance; dry stretches, sea legs and daily distances use the solved elapsed time. Current percentages describe the actual time effect; wind is clearly shown as direction context rather than an unmodelled speed bonus. These remain story estimates, with vessel and resupply assumptions.
Neighbours are geography; diplomacy is a choice. Shared borders have recorded map evidence. Rivalries, alliances and disputes are labelled as suggestions unless authored. Border prompts focus an actual border sample and no longer invent contested rivers or mountain passes between capitals.
Clearer provenance across views. Compendium, handbooks, dossiers and the relations graph distinguish suggested goods and politics from author canon. Turning generated lore off preserves measured connections and neighbours.
Cross-output checks before completion. New builds retain compact route evidence even without an illustrated atlas. Publication checks descriptions against paths, GIS and territory grids; older descriptions remain readable and report that their evidence has not been verified.
One source for local goods. Resource descriptions, market goods, author questions and realm handbooks use the same local market outputs. Realms seated in smaller towns receive a local market, keeping their goods within their own collection area. Renaming a settlement refreshes its goods evidence in the same save.
Trade estimates follow recorded routes. Outside suppliers and imported bronze ingredients require connected roads, sea lanes or railways. Physical travel distance stays separate from assumed transport costs. Nearby ports alone no longer establish a shipping connection or export demand.
Goods edits update dependent crafts. Removing iron removes the steel it supported; adding ingredients recalculates craft potential. Explicit author choices remain possible, and restoring the world's goods restores the original ingredients and dependent products.
Potential is clearer than certainty. Markets, Compendium and dossiers distinguish modelled goods, suggested specialties and author choices. Expand a market's evidence section to see assumptions and missing readings. New publication checks compare resources, markets, route costs and saved summaries; failed ground readings stop a fresh build instead of silently keeping older goods. These are availability and cost estimates, not measured production, demand or actual shipments.
Habitat readings use the world's actual size. Species ranges, biodiversity and local goods now share spherical distances and the recorded planet radius. Nearby habitat across the date line or near a pole is sampled correctly; inland lakes no longer qualify as marine habitat. Land goods use a stated 50-km collection radius, with marine habitat sampled within 250 km of ports.
Ground evidence takes priority over labels. Grain, livestock, fruit and forest goods follow suitable catalogue habitats; iron needs a worked deposit when those readings are available. A port alone no longer guarantees fish. Charcoal follows timber availability. Markets explain their sampling limits; habitat remains potential, not proof of harvests or actual production.
Fresher maps with lower habitat-reading memory. Changed calculation recipes invalidate scaled drawings, and preserved file timestamps cannot hide replaced habitat or deposit inputs. Catalogue reads stream species one at a time. Older goods rules remain readable and are explicitly marked as needing a refresh before verification against the current model.

Make

Make is a fifth primary workspace. It opens its own page alongside Create, World, Write and Library, with the same active navigation and arrow-key behaviour. Its selected guide remains selected when leaving and returning; Escape does not dismiss the workspace.
Three illustrated paths to a finished piece. Make a playable region, put maps in a chapter, or finish map artwork. Each path has three steps that open the relevant existing editor, with book export and Atlas PDF routes available too. Missing worlds or books lead to their starting controls.
Guides preserve the work in progress. Handoffs save pending writing or artwork first, retain child-map drafts and preselect the current child map for hex/VTT export. Delayed loads or failed saves cannot pull you away from a newer workspace. Opening a step does not automatically create content, export a file or mark the task complete.

Library

Chosen world names survive rebuilds. Metadata renames now update the same canonical name used by lore and text exports. Ordinary rebuilds preserve older Library-only names without replacing an existing canonical rename. Worldbook titles also respect the chosen name during generation and rebuilding. Failed derivative refreshes are reported while the saved name remains recoverable.
The world takes centre stage. Resume pairs a large, full-proportion map with the world's story and measured details, always visible below Resume. Removed the details toggle, decorative gold background and the map's black inset, border and rounded frame. Compact creation buttons with coloured icons stay above the saved projects. Book covers retain their saved proportions.
Browse a growing collection. One to three worlds use compact project rows; larger collections use a card grid. Search matches world names, seeds and stories, including accented names. Filter favourites, unfinished projects or worlds needing a version update; sort by recent save, creation or name. Reset view restores the complete shelf without changing the saved collection.
Recovery is easier to find. A visible Trash shortcut shows the recoverable count and opens the existing recovery controls. Open, favourite, update and deletion actions remain available on each project.
Library actions share one row. Tabletop now has a dice icon alongside Scenes, Map gallery and Import project backup. The row uses the full width below a saved world's summary; narrow windows scroll horizontally while keeping every button's label and keyboard access.
A smaller bundle for authors creating their own worlds. Removed the offline starter-world gallery and its prebuilt worlds, freeing about 227 MiB of bundled resources. Any starter world you already saved remains an ordinary editable project in your Library, including its notes and backup support.

Create

Your brief's choices win over the planet type. A world is composed as land shape, then planet type, then your brief, then Advanced, in one place the search, previews, reading and build all share. Lush on a thin, cold planet makes that planet wetter (340 to 464 mm a year) instead of being overruled, and a Land coverage band you pick is the share you get; the planet fills only what you leave on its middle setting. With explicit choices the planet used to overrule at least one of them on 8 of 11 planet types; now on none.
Warm, Cold, Lush, Arid and the other graded choices move your world. Each is a step from the land shape and planet, not a fixed value: Warm on the tropics is warmer than the tropics, where it used to cool their poles, and Cold on an Ice age world is colder. On an ordinary world every choice gives what it always did.
Each graded control has a middle that keeps the world's own value. Climate, Rainfall and Seasons read "From the planet"; Land coverage, Coasts, Mountains, Geological age and Great rivers read "From the world". Land coverage starts there, so a shape's or planet's own share holds.
Describe says where each value comes from, and Advanced opens on your world. "Where the numbers come from" reads, for example, "Rainfall 340 mm a year — set by Thin air, cold desert"; the Advanced sheet starts from the combined values rather than the bare preset, says who set each, and your own edits still win. World recipes remember the planet type.
A clearer first visit. Describe identifies its pictures as preset illustrations. Candidates opens with Ready to search, explains how the brief is compared with measured geography, and says where earlier searches remain available. Opening it does not start a search or add another search button. Searches with conflicting checks still offer Review the brief.
Every land shape checks for what makes it that shape. Rifted, Drowned, Fjordland and Equatorial belt were judged as Balanced worlds, so most rifted candidates showed a 0% miss. A Rifted world now needs the one great continent its rift tears through, a Drowned world shallow seas covering at least 30% as much as its land, a Fjordland a coast complexity of 48 or more, and an Equatorial belt at least 45% of its land in the tropics with at most 12% beyond 55°. Ice age planets ask for at least 8% of their land beyond 55° again, and Lush and warm planets at least 35% in the tropics.
"Best brief match" puts the safest matches first. Candidates that pass the same checks are ordered by how far their figures sit from each limit, measured against how much that figure moves between the card and the built world, a must-have counting twice a preference. It used to put the most ragged coastline first; checks now change answer at the build about a third less often.
An inland sea is asked for, not assumed. "An inland sea" and the Inland sea shape ask for 3 million km² of enclosed sea instead of 0.35, which every world has, and Twin hemispheres prefers its continents at least 120° apart. A coast kept unbroken ("Avoid: broken coast", "no fjords") is left to you, as a smooth coast already was: a coastline reads far more broken at the detail a world is built at than on its card.
Rules no longer change your rainfall behind your back. "Avoid: jungles" or "Must have: a great desert" used to make the world drier or wetter while Rainfall still said Earth-like. The rules leave Rainfall alone now, and the reading beside the brief offers "Rainfall: Arid (from 'Avoid: jungles')" with an Apply button.
A Prefer box beside Must have and Avoid. Things you would like but can live without raise a world's match score without ruling any world out, appear in what the search checks, and save with the brief and recipes.
Changing the land shape brings that shape's own settings and keeps yours. Picking Fjordland after starting on Balanced gives its fjorded coasts and sea-going peoples, not only on a fresh sheet; anything you set yourself stays, and a line under the shape's name says what changed. "Use an example" knows your planet type too, so all nineteen examples can be reached.
Find 24 more worlds says what it does. "Show 24 variations" never made variations: it ran a second search from the chosen card's seed, and its worlds were as unlike that card as the twelve are. The button, heading, Recent searches and manual call them 24 more worlds, named on from the twelve (World M, N, ...).
Search again is a new search every time. It reused the seed field, so it returned the same twelve if no card was chosen and an unrelated set if one was. Each Search again uses a fresh seed, kept with the set and on every card; the first search of a brief still starts from the seed on Build, and Recent searches keeps your last ten sets.
A card scrolled under the pointer shows its larger look. Scrolling the candidates under a resting pointer used to hide the card's larger preview until the pointer moved again.
Each kind of "brief" has its own name. The realms you name before a build are the realm roster (it was a "nation brief", or "Nations, cultures and tongues"), a shape drawn with Describe a region is a described region, and what the Regional lens measures is a lens summary; only the World Brief is called a brief. Verdicts, saved files and behaviour are unchanged.
A candidate is drawn from the same settings its world is built from. Cards, variations and every preview use your planet type, Advanced settings and tectonic drawing, as the build does: a thin, cold planet's card no longer shows six continents and builds one (card-to-build shape difference fell from 0.28 to 0.07, about what the finer grid alone makes). Changing the planet, a setting or the tectonic drawing marks the shortlist out of date, cards no longer count a thesis idea you already applied, and "None: flat" mountains are flat on the cards too.
Choose shows the checks your Quick build will measure. It redraws the chosen world at the size a Quick build uses and lists what changed since the card ("on the card: 3; at this detail: 2").
A world built from a chosen card keeps that card's promise. The World panel re-checks every check on the saved world and says how many of the card's promises held, with what the card showed where one changed.
Every check says what it measured. On Candidates and Choose each check shows whether it is a Must, Prefer or Avoid, what the world measured and what you asked for; a world that meets every must-have but misses most of what you preferred is marked to check, not as a fit.
Must have, Prefer and Avoid read your words the right way round. "No inland sea", "not tropical", "not far apart" and "no fjords" ask for the opposite of what they name; "polar land", "ice-covered poles", "mostly land" and "an overwhelming landmass" ask for more of it, not less; "subtropical deserts" is left for you to judge rather than read as a tropical world.
One island count everywhere. An island is a landmass of 40,000 km² up to 2 million km², as the brief reading always said, in the Library, the Compendium and the cards alike (one world that showed 111 islands in the Library now shows 55).
Searches are lighter and keep up. The shortlist updates as worlds are found without resending every picture: a Thorough search moves about 0.4 MB instead of 6.6 MB, and ranking keeps pace even at Deep depth.
A brief that contradicts itself says so while you write it. Warnings and errors appear under the brief on Describe as you type, not only after a search: land coverage that runs backwards, high mountains on flat land, several continents on One Great Continent, No peoples with only the realms you draw. A brief with an error is no longer built; Build says what to fix.
The World Brief accepts only choices its menus offer. A saved brief or recipe with a value no menu has (a misspelt climate, an unknown magic source) uses that control's default and names the field, instead of being stored and quietly built as the default. A recipe with a damaged brief opens and loses only the broken field, and the brief no longer carries fields nothing read.
A Delta V6 brief opens on Describe without warnings. Delta V6 saved the graded controls' middles as words ("prominent" mountains, a "temperate" climate). Describe now reads them as From the world or From the planet, saves the brief that way, and lists nothing under your story about it; before, it showed one warning per control until you touched it.
Rewording a built world's thesis no longer re-simulates the planet. Of the brief, only the climate zones you place change the simulation, so after an edit to the thesis, title or rules the next rebuild starts at the names: about 17 seconds instead of about 110 on a compact world. The brief lives in one file, so the Library, search, atlas cover and recipe export can no longer show a different thesis from the one the world uses.
A smooth coast is left to you rather than failing every candidate. The search cannot measure a smooth coastline at its preview size, so a Must have of one marked every candidate a miss; it is now listed as yours to judge.
Drowned Interior and Dim Star start from choices the menus offer. Drowned Interior asks for old, subtle mountains (it asked for a value no menu had and got prominent ones); Dim Star asks for gentle seasons. The land-coverage phrase previews ("Oceanic", "Land-rich", "Inland seas only", "No oceans") now build the coverage they describe.
Icons bring the Describe sections to life. Story, Planet, Land, Peoples, Tongues, Tectonics, Limits and Advanced each have a distinct line icon beside their label. The selected icon follows the theme accent; on narrow screens, icons sit above their labels. Text labels and keyboard navigation are retained, with light/dark themes and enlarged text checked across five screen sizes.
The drawing holds still when you change tools. Draw your world kept brush settings at the end of its tool strip, so picking a brush wrapped the strip and moved the drawing by up to 48 px under the pointer. Brush settings now sit in a one-row options bar of their own, and tools without settings say there how they are used. On phones the bars still wrap, and the drawing stays centred on the stage.
The drawing holds still on narrow windows too. Between a phone and 1100 px some tools wrapped the tool strip or options bar to a second row and others did not, so changing tools moved the drawing by up to 25 px. Both bars now keep two rows there for every tool, with a brush's sliders above its switches, and the drawing stays put. Just above that, from 1100 to 1122 px, opening the Relief tools pushed Weather onto a second line and moved the drawing about 15 px; the strip's buttons are a little narrower up to 1200 px so every group fits on one line. With text at 110% or 125% the drawing moved up to 26 px on desks from 761 to 1520 px wide; where the larger words need it the tool strip and options bar now keep two rows (three for the strip at 761-820 px at 125%), and the drawing holds still. At larger text on a wide window the bar along the bottom of the canvas keeps one height instead of growing over the south of the drawing; on a small phone Save drawing and Use this drawing stay on screen with a relief brush in hand; and the canvas's menus and Detect coastline show their full words (the save format's "Land outlines GeoJSON" is now "Outlines GeoJSON").
Numeric heightfields can declare their source waterline with terrain_source_sea_level_m. Heights are converted to metres above world sea level before simulation, so a file's vertical offset no longer changes its climate altitude or turns its ocean into land during sculpting. The original file and its offset are retained for consistent rebuilds.
Existing numeric-import saves with a nonzero world datum can use Update to this version when their sculpt edits support an exact height translation. The update keeps the previous world, converts absolute targets and preserved heights together, and fully rebuilds dependent information. Coast-sensitive or otherwise unsupported edits receive an explanation before copying or converting anything. Ordinary rebuilds cannot silently mix the two datums.

Brief sections open in view on phones. After the first questions on a narrow screen, choosing a brief section (Advanced included) left the page where those questions were, with the section's first settings behind the search footer. Choosing a section now brings that section to the top of the page; the section tabs are one scroll up.
Candidates can be picked while the search runs. The candidate grid redrew every card twice a second during a search, so a click that straddled a redraw was lost and an open More menu closed. Cards that have not changed are kept as they are; only new or changed ones are drawn.
Larger candidate previews with room to compare. Desktop and laptop screens show three worlds per row, putting twelve matches into four rows. Maps keep their full 2:1 shape and more space around them; the candidate pane scrolls to later rows while the action bar stays visible. Phones retain two columns, and hover previews move above or below a card when needed.
See what sets each candidate apart. Cards show a short description of measured geography, such as a dominant landmass, substantial islands or broken coastlines, alongside their existing statistics. Repeated passing check counts move to the selected world's details; missed checks still appear on the card. Selection, comparison and the three-column grid remain.
More room to draw tectonics. The plate map fills the editor's width. Shields, Plates, Margins, Hotspots, Motion and Conflicts sit in a compact row below it, with one collapsible section open at a time. Draft checks open Conflicts and show a count when problems are found. Drawing, Undo, Reset draft and Apply to brief keep their existing behaviour.
Preview and build in one step. Create now runs Describe → Candidates → Build. The chosen world and nearby candidates sit beside its seed, statistics, output choices and estimates, with Create always available in the footer. Sketch, Quick, Full and Publication atlas retain their existing settings. Known seeds and imported or drawn maps use the same final view; attached maps show their own preview instead of an earlier candidate. Narrow screens stack the preview and options, and saved sessions on Choose reopen on Build.
A shorter route from an idea to candidates. Describe your world keeps Story, Planet and Land together, with Peoples, Tongues, Tectonics, Limits and Advanced directly available. Each section's empty/default/configured status appears on hover or keyboard focus. A smaller header and footer leave more space for the story and settings. Candidate names have their own row, with room for the candidate grid on shorter laptop screens.
More visible Create buttons. "I have a seed" and "Start from Deep Time" have compact bordered icon buttons, the selected search depth has a clear accent, and Find matches uses stronger contrast. The footer keeps its original quiet surface and compact spacing. Keyboard focus is more visible, with larger primary touch targets on phones.
Quieter Create feedback. Removed the technical notice about unused fields from older World Briefs.
See what the search will actually check. Open "What the search checks" in the right-side world panel; the duplicate footer button is removed. It distinguishes active requirements from suggestions you have not applied, includes land-shape checks, and shows how many active checks a candidate passed. The searched brief is saved with the results. Failed readings can be retried in that panel; late replies cannot overwrite a changed brief, and editing a land shape refreshes its reading.
Different creation recipes get distinct automatic world names. New names use the seed together with the generation settings, brief and starting terrain. Duplicate automatic Library names receive a numeric suffix. Existing saved world names stay unchanged.
Name a drawn or imported world before building it. The generation dialog asks for a world name; leaving it blank chooses an automatic one. Cancel starts no job. An entered name is saved before generation so the naming pass cannot replace it.
Fast builds can use spare CPU and RAM beyond atlas rendering. Build → Build performance → Fast evaluates up to four climate seasons concurrently. The chosen detail, solver budgets, seasonal sampling and physical results stay the same. Free memory and CPU count limit concurrency; Standard, Low memory, Quick and Sketch remain available. Fast stays selected through preferences and recipes. The build log reports the workers actually used, and timing estimates distinguish worker counts and deferred city-map drawing.
Measured climate-stage improvement. On the local Quick-size fixture, serial climate took 31.06 seconds and four workers took 14.02 seconds, with identical output hashes. This measures that stage on macOS; whole-world runtime and other platforms depend on their remaining work and hardware.

Deep Time

A visible cursor when placing plates. Choosing Place plates now shows a crosshair inside the starting-crust canvas. The cursor previously disappeared when entering the canvas, making placement and dragging harder to follow.
Start from your own work. Opening Deep Time restores saved history or a saved crust drawing. With neither present, it offers a blank start and a direct drawing action instead of generating a starting continent. Generating an outline remains an explicit option in the drawing editor. Restoration now happens before saving an initial state, preserving the drawing and its Undo history.
A clearer drawing and history view. Crust previews use their full saved 1200 × 600 resolution, smoother display sampling and a faint graticule. Era images use thin, antialiased plate boundaries instead of enlarged black cell edges. Older saved frames refresh their presentation when needed without rerunning simulation, retaining saved ice, sea level and isostatic displacement. Simulation resolution and saved drawing coordinates stay unchanged.
Fullscreen leaves the picture and caption visible. The caption and fullscreen action share a footer outside the drawing, including in fullscreen, so the button no longer covers information. Specialist plate and orbital settings are grouped separately from the main starting controls.
An empty canvas says so. After Clear, after erasing every coast, or when a generated outline fails to load, the Starting crust summary reads No drawing yet instead of Custom drawing, matching the drawing prompt beside it.

World exploration

Check scores have a clear scope. World overview and Planet Advisor label scores as summaries of available checks. A clear report no longer declares physical plausibility; independent calibration is still required.
States and Counties are map layers. Find them under Life & settlement beside Realm territories. Each colours the saved divisions at its own level, with names, borders, places and individual territory filters. The layers follow your custom political titles and names, edited in Political hierarchy. Each map remembers its own border and place choices.
Political divisions are ready by default. New worlds generate both subdivision levels during creation. Opening an older world fills in missing divisions from its saved realm territories, reusing existing data and keeping names, custom titles and deliberate hierarchy settings. An author can still choose provinces only; very small territories may remain undivided.
Countries, states and counties on the map. Political hierarchy is accessible from Realms & settlements and Map appearance's Borders and places. Choose country → state → county, kingdom → duchy → county, or custom titles for one realm or the world's defaults, with up to four subdivision levels below a realm. Each division stays inside its parent; names, descriptions, child maps, dossiers and exported notes agree.
Realm territories shows the people and borders. The Life & settlement map opens with realm names, distinct capitals, towns, ports, routes and the first two subdivision levels when available. Crowded labels wait for a closer view. Toggle individual levels in Borders and places; political map choices are remembered separately from other maps and in saved views.
Maps sharpen when opened at a busy moment. A world opened while the page was still laying out its map could keep the blurred overview until you zoomed or panned. Sharpening now waits for the map to take its place (up to five seconds) before settling for the overview, including when the map picture itself has not been laid out yet (on a busy machine about 1 opening in 30 still stayed blurred that way; now none in 90).
Planet card buttons are easy to tap on phones. Fold and Close each keep a full 24-pixel target when the card's header closes up, instead of a 22-pixel Fold squeezed beside Close.
A quieter favourite control. Map layers keeps a compact star in the lower action row beside Original and Download, with its action named on hover.
Visible exits for map panels. World overview, Map layers and the floating Planet card now have their own close buttons. Keyboard users can close them with Escape and return to the launcher; search still clears its query first.
Place highlights are removable. Show on map adds a small place-name chip with a clear button. Clearing it or closing the cell inspector removes the temporary marker without changing map layers, moving the camera or deleting the saved place. This also covers points located from other world tools.
The cell inspector names the whole place. Its heading was cut short beside Pin, Report this place and close ("Freshwater lake · la…") at every window size. The name now has its own line under those actions and wraps when it is long; close stays in the top-right corner.
Map menus stay clear of the workspace rail. Appearance, Tools and Export now use the available height below the rail, including short and narrow windows. Escape still closes the menu and returns focus to its launcher.
Split, Stack and Minimap stay usable at smaller sizes. Comparison controls wrap together below the rail; the planet card moves below them. Narrow windows keep their comparison controls and minimap, with close buttons inside the visible frame. Opacity, Multiply and the second map remain reachable. The comparison image refreshes with the current projection, meridian and look.
Erosion previews follow the map. Overlay colours now use the selected projection and meridian, fit the geographic picture and resize with it. Switching tools clears the overlay; a pending redraw hides the old one. Independent preview requests no longer overwrite a shared image file.
HD regions use the matching map geometry. From a curved projection or a turned map, Use Stretched & draw opens the supported flat view at its original meridian. Changing projection or turning afterward removes the old HD overlay instead of leaving refined terrain over the wrong location. Regional relief, temperature, rainfall, drainage and PNG export remain available.
Lab readings and control names stay visible and accurate. Climate Lab's seasonal readings remain on screen as the season or layer changes. Its inputs and Erosion Lab's sliders have explicit labels. Temporary loading explanations no longer replace the names of Map look and Shallows after loading finishes.
Find any map layer from one place. Map layers offers Search all map layers and All, Favourites and Recent views. Search covers names, descriptions and categories, including advanced layers without changing the Essentials setting. Enter selects the first available match; Escape clears the local search first. Empty results leave the current map intact. Category headings show available and total counts again, and failed catalogue loads retain a retry action.
Keep favourite and recent layers. Star the active layer beside its title; each world remembers favourites and its last 12 distinct layers locally. If local storage is unavailable, the app explains that these shortcuts last only for the session. Resetting these preferences leaves world data intact.
Save named map views. Saved views on the World map can save, open, rename, replace and delete a view. Views include layer, pan/zoom, projection, meridian, appearance, dynamic filters, overlays, borders, places and grid spacing. They travel in project backups. Missing layers or obsolete filters are explained rather than silently replaced; revision checks protect changes made in another window. Temporary Compare, Time strip and HD region views must be closed before saving a main-map view.
Closed history overlays stay closed. A delayed Time strip response can no longer redraw its overlay after closure, and hidden SVG overlays honour their hidden state.
Find places using the world's saved evidence. Combine name, place kind, realm, biome, elevation, annual temperature/rainfall, population, distance to sea and port status. Numbered results open the map location or place dossier. Coverage counts show which readings exist; missing values remain unknown, and incomplete matches are counted separately. Regional readings describe the map pin, not every part of a realm, river or region.
Keep named place shortlists. Save up to 30 lists of 200 places, with rename, update, restore-search and delete controls. Draft selections survive filtering and dossier visits; saved lists survive reload and project backup. Missing places remain labelled and removable. Concurrent-save conflicts keep the draft and offer to reload the saved lists. Compact layouts put results first and keep filters behind a visible Criteria button.
Lakes show their water budget. Inspecting a lake lists the water arriving, rain on its surface, evaporation and seepage, and any overflow at its rim, in m³/s from the world's own year. A closed hollow without a drawn lake explains where its water goes: a desert hollow evaporates it into a salt flat, and a flat floor spreads it thinly over a stated share. Why here? explains a lake by the same budget: whether its losses use everything that arrives below the rim, or the rest overflows, and why a closed lake is salt.
Lakes are named for their water. The biome maps and legends call the lake class Lake, since a closed desert lake is salt. The inspector names each lake Freshwater lake or Salt lake from its saved regime, and a salt lake's place prompts describe brine rather than reeds and frogs.
Why here? explains rivers. A river cell now has its own subject: the upstream drainage area, mean and seasonal discharge, whether it runs all year and its channel form, with what the map does not resolve. It is offered first on a river cell unless a settlement stands there.
Why here? explains only routes that cross the cell. A road or sea lane within a few cells can still be selected and edited from the inspector, but only a route whose line runs through the inspected cell is offered as its explanation. A lake beside a road no longer claims the road crosses it.

Sculpt terrain

Sculpt owns its lifecycle. Repeated initialization binds once. Disposing releases listeners, observers, timers and browser requests; reopening keeps staged edits and undo history while ignoring stale callbacks. Rebuilt controls replace their old listeners without retaining discarded targets.
A rebuilt copy that fails to save says so. Saving a comparison draft as a new world in the background (with name transfers) used to leave the status reading "Saving the rebuilt copy in the background" for ever when the save failed. The status now reports whether the copy was saved or why it was not, and the draft stays ready to retry.
More map to sculpt on phones. With the Sculptor open on a narrow screen, Saved views and Find places step aside, as the globe does, so the map band above the workbench is no longer squeezed: the whole world draws at 179×86 instead of 104×50 pixels on a 390-pixel-wide phone.
An illustrated palette for all 31 tools. Coloured tool illustrations, the original selector and searchable Browse tools make every sculpting tool reachable. Settings, Preview, Edits, Compare and Help open directly; active settings precede explanatory hints. Search matches names and descriptions.
Less crowding beside the map. Smaller icons and descriptions, reduced nested padding and a wider desktop settings area give the right-hand controls more room. Compact layouts separate the map from scrolling settings. Undo, Redo, Review & re-simulate and Discard stay reachable, with the editing scope and pending changes visible.
Keyboard editing returns to the canvas. Undo and finishing a gesture work after using settings fields. Naming a realm returns focus to the map so Enter can continue to its capital selection.

Maps

Unavailable ground estimates have a reason. Inspector and dossier readings distinguish missing climate, invalid inputs and a heat cycle that did not settle. Travel keeps valid weather when its optional ground estimate fails; the saved campaign and Markdown timeline retain that reason. Failed readings do not become zero scorching days or reuse temperatures from an earlier plan.
Ground estimates respond to pressure and settle before being shown. Inspector, dossier and journey readings use the planet's pressure and gravity in a compact dry-air atmosphere model. High-altitude sunlight stays bounded. The soil column now checks its repeating temperature cycle and heat budget; the former fixed three-day warm-up could leave a biased reading. Estimates identify their atmosphere inputs. Existing worlds refresh these derived readings on request; saved campaign plans retain their original conditions.
Ground heat follows the saved orbit. Cell inspections, dossiers and journey readings now use the climate archive's orbital geometry and stellar flux. Eccentric orbits change sunlight through the year, using the same equations as the climate model. Older saves fall back to their forcing report or normalized planet settings. Readings include their forcing source and clarify that cell summer/winter estimates pair air-temperature extrema with solstice sunlight.
Journey weather agrees with its saved climate. Route distances, calendars and seeded storms use the same normalized planet settings. A rebuild or weather sheet change during calculation requests a refresh. Campaign journeys preserve their storm settings and climate reading context, and their expanded timelines and Markdown downloads identify the weather as a recorded plan. Older journeys remain readable without invented source information.
Tides and storms keep the world's own clock. The calendar gives the moon's orbit in local days, but tides read it in Earth days, so a world with a long day got the wrong tidal rhythm. Journey weather used Earth's spring-to-neap ratio on every world (2.7, against 1.39 on the default test world); it now uses the world's own, reads a port's tide at its nearest coastal cell instead of the highest within about 80 km, and moves storms by the world's own day length and radius.
Saved journeys retain their original readings. Campaign timelines and Markdown downloads keep Travel's midday/predawn ground temperatures, materials and sunlight inputs. The original planned route stays with the weather record; editing its map path shows a visible mismatch notice. Older records identify their missing route context. Complete 600-point planner paths are preserved, and city pins no longer block dragging journey endpoints.
Weather Year refreshes saved climate reliably. Reopening the player reads current data even when replacement files keep their timestamps. Map frames, wind/current vectors and seeded storms use the climate revision supplied by the player's calendar. A changed source requests a refresh instead of mixing revisions; a failed frame clears playback and overlays. Climate equations, map colours and seeded storm behaviour are unchanged.
Seasonal saves keep the whole world. New odd-sized worlds no longer lose their last row or column when climate frames are compressed. Weather Year follows the saved grid method for land, currents and tides, and particle-grid reduction also retains edge flow. Even-sized worlds keep their existing frame values. Older cropped frames stay readable and identify their limited coverage; re-simulation creates complete coverage. Incomplete or damaged seasonal years are reported instead of being animated as a complete year.
Climate Lab readings describe their real scope. Seasonal global means now account for cell area. Resolution notes report the actual saved and simulation grids, including irregular layouts. Rainfall rates use explicit mm/year or in/year units, and point rainfall agrees with native-cell hazard sampling. Missing or invalid seasons receive an explanation; failed reads cannot bring back values from a previous layer. Readings include grid and source evidence.
Seasonal rain keeps its annual total. Climate classification, snow, storm interpretation, crop suitability, growing-season findings and history food capacity now share one rainfall normalization. Tiny valid samples no longer lose rainfall to a denominator floor. Bounded rainfall interpolation prevents false dry cells between positive samples. A wet annual field with no supporting seasonal pattern receives an explicit error; existing pictures remain readable. Affected map caches refresh, and previous history runs are marked stale against the revised food calculation. Saved simulation data does not need rewriting.
Tiny seasonal rain survives saving. Some imported worlds could fail to finish because compact seasonal saves rounded real rain samples to zero while annual rainfall stayed positive. Affected rain frames now retain float32 at the same resolution; ordinary frames keep their compact format. The original seasonal pattern survives, and checks still reject genuinely missing evidence. Re-simulation can regenerate older erased samples from their original inputs.
Damaged seasons no longer break the map library. Seasonal redraws reject malformed, non-finite or mismatched saved grids instead of crashing the whole catalog or interpreting bad values as climate. Existing pictures stay available; affected redraws explain the missing or invalid source. Other maps remain usable, and repairing the saved fields restores the affected controls without restarting.
Monsoon maps and readings agree. Valid irregular seasonal grids now work, including odd dimensions. Bounded interpolation wraps at the date line, narrow islands retain their land support, and wind arrows follow both grid dimensions. Rainfall per day uses the world's year length. Inspection reports the same measures used to classify the map, with source/version evidence; existing Monsoon render caches refresh for the revised interpretation.
Choose whether to generate city maps. Build has a separate Generate city maps switch. Turning it off keeps towns, realms and lore while deferring their drawn plans, including plans made by child-map refinement. The choice stays in preferences and recipes and applies on rebuilds. Draw city plans later from Atlas when you want them.
Nest your own city or encounter map. Scene Studio's Add your map image accepts PNG and JPEG. Child maps → Connections offers the same path and saves the imported scene linked to that world and child map. Images keep their proportions, remain embedded through scene backups, and can gain editable pins, lighting and a grid without replacing the world's terrain.

Drag out a polygon with the sides you choose. Polygon, in the Sculptor's Stroke choice and in Describe a region's Shape, now asks how many sides (3 to 64), then you press where its centre goes and drag outwards, the way a described circle is drawn. With Equilateral ticked every side is the same length and a corner follows the pointer, so the drag sets the size and the turn (Shift snaps the turn to 15°). Untick it and the shape stretches to fill the box you drag out (Shift keeps that box square). A click without a drag uses the brush size. Clicking corner by corner is still there as Freeform polygon.

Find saved child maps from World → Tools → Child maps. The entry sits next to Regional lens and opens this world's saved maps without search. Nested maps show their parents, and breadcrumbs take you back through the map hierarchy to World. Reopening a map keeps its identity, pins, routes, hotspots and recoverable editing drafts.
A clearer child-map desk. Map, Notes and Connections separate map tools and refinement from writing and linked content. Title and Save stay outside the section switcher, with a clear saved or unsaved status. Compact layouts put the map first. Connections shows smaller maps inside the current map and its linked scenes; opening a scene and returning keeps that child map selected after reload. Existing place, Compendium and manuscript hotspots continue using the same saved links.
Keep your place while moving between maps. Unsaved titles, purposes, notes and route drafts remain tied to their own world and child map. Late map reads cannot replace the map data after you switch worlds. Maps with an unavailable parent remain reachable in the saved directory, and the current map is marked in both the directory and its breadcrumbs.
Read the connections for this map. The Connections section lists maps nested directly inside it and scenes linked to this world and region. Linked scenes can be refreshed, with retry guidance if they cannot be read; a delayed response from another map cannot replace the current list. Returning from a scene restores the saved map selection without overriding a newer navigation choice.
Start another child map from the directory. An empty directory explains how to create one, with a shortcut back to the regional lens. Returning to a lens already ready to choose its centre keeps it ready. On compact screens, choosing a map tool brings the canvas into view for the next click. Map, Notes and Connections also support keyboard section navigation.

Scene maps and settlements

An illustrated scene library. More compact starting cards introduce a blank canvas, settlement planning or bringing a map. Saved scenes retain their open and play actions, with visible scene and recoverable-trash counts. Saved-scene import options open in a dedicated dialog with keyboard focus kept inside it and returned to the trigger on close.
A drawing desk built around the canvas. Scene Studio has a labelled tool rail, an independently scrolling inspector and direct access to canvas/grid, text, appearance, connections, exports and saved-map history. Save, Undo, Redo, preview, map selection and tool search stay reachable. Compact windows keep the document bar visible and the action row outside the drawing.
Preview and drawing stay connected. Preview remains beside hotspot and fog controls. Returning restores drawing keyboard access, including after a delayed preview. Isometric pointer edits focus the canvas so Undo works after inspector changes. Scene links, geometry, lighting, fog and exports retain their existing editing and saving behaviour.
Choosing a world place keeps the list it is reading. Opening Choose a world place while Settlement Studio was still reading the saved worlds cancelled that read and started it again; it now waits for the one under way.
Viewpoint pins show their view. Following a viewpoint pin from the scene preview, or Go to in Viewpoints, closes the preview, so the saved plan or isometric view is on screen instead of hidden behind it.
Settlement planning gives the map more space. Plan settings sit beside the drawing where room permits; naming and saving sit near the result. Planning reports and drawing/tabletop exports follow the map, with export grid settings beside those exports. The world picker opens deliberately or from a linked world/place.

Atlas

Rivers are named on plates drawn later. Plates drawn after the build with Create draft plates or Create publication atlas drew every river but named none. They now read the rivers the build saved and letter them as plates drawn during the build do, above and below their lakes too: 55 river names across the compact test world's 12 plates, where there were none.
Atlas actions wait for a build that is still opening. Create draft plates, Render the publication atlas and Draw city plans did nothing at all when clicked while a just-finished build was still opening its world. They now wait for it, showing so on the button, and then start.
Choose the world-sheet format. Draw projected world sheets, full-pole rectangular sheets, or both, alongside selected continent and regional sheets. Named and unlabelled variants remain available. Existing worlds can draw the selected sheets from saved geography without regenerating terrain; unselected sheets remain untouched and interrupted plans can be resumed.
Cleaner illustrated sheets and lettering. Lighter land washes, quieter texture and relief, prominent continent names and smaller realm names improve the hierarchy. Water and lake labels have clearer contrast, while settlement and mountain-range halos are narrower. Projected labels follow their actual geography and are drawn crisply after projection. World and continent sheets omit the former compass and scale decoration. Existing saved pictures change when explicitly redrawn.
Review the exact sheet that needs attention. Needs review filters sheets; Needs refresh and Unverified badges explain their cause and lead to the relevant refresh or drawing choice. Freshness follows the displayed named or unlabelled variant, including fallbacks, so a rename need not invalidate an unlabelled sheet. Changing a sheet's shape retires an incompatible old companion instead of offering a mismatched image. Missing or deferred sheets name what is absent and offer Choose sheets to draw.
Regional plates show the whole region, with its border drawn. A regional plate is named for a realm but was framed on one fixed window around its capital, so it often showed only part of that realm — between a fifth and two thirds of each realm on a measured world. Each regional plate is now framed on all of its realm, and the realm is outlined in red with a tinted edge while the land beyond it fades toward the paper, so a reader can tell at a glance what belongs to the plate's region and what is a neighbour. Plates that follow your drawn realms are outlined along the border you drew. Region frame (Whole region or Around capital) and Region border (Drawn or None) sit with the other atlas choices in Create, in Generate atlas sheets on the Atlas tab, and beside plates waiting to be drawn; the sea, the names and every non-regional sheet are drawn as before.

Map artwork and exports

Library artwork stays with the scene that asked for it. Artwork from the shared library that finished loading after you pressed Back could be added to the scene you had left, unsaved, while the library said it was added. The library now says the map changed and asks you to choose the action again.
The Export menu scrolls again. On shorter windows the format list no longer snaps back to the top as you scroll it, so the last formats, including the complete project backup, can be reached.
Older worlds redraw their flat sheet again. Redrawing a saved world's vector sheet no longer fails when its numerical archive predates the saved land mask; such worlds use land above sea level, as they did when saved.
Render Studio is easier to find. Its Tools and Map appearance entrances now say Render Studio instead of Edit artwork, matching the editor's heading. Search also recognises Edit artwork, oblique, relief and colour terms; Make's artwork guide uses the same name.
Adjust the Oblique view. A saved render bundle's Oblique view opens a dedicated preview with camera tilt (0–75°), relief height (0–4× the automatic height for that sheet), and side shading. Update preview redraws the view; Reset restores the defaults. Save PNG downloads that preview, or View on map opens it in the main viewer. Settings are remembered per bundle during the session, and the PNG records the camera settings and scale explanation.
Keep the artwork's look while changing its relief. Oblique carries the saved bundle's colours and terrain layers through the tilt, without changing the artwork document or simulated ground. Zero relief produces a flat view. Pending changes are clearly marked before downloading, rendering errors can be retried, and a late result cannot reopen a closed window or replace another world's view. The preview fits both themes and compact windows, with keyboard focus returned to its opener on close.
Illustrated art directions across the map editors. World artwork, Scene maps, native Settlement Studio and child maps show the three existing art families as compact illustrated choices. The regional lens offers them too. World previews appear beside the choice and clear when the selection changes; the cover illustrations are examples, while Preview uses your actual map.
Keep individual style adjustments. The preservation option is visible and checked by default. Custom inks, grid/caption settings, atmosphere and vector styling remain while untouched values follow the next family. Custom assets, colours, text, light sources and geometry retain their owners. Full supported-style replacement remains available. Reapplying a family no longer repeatedly increases vector line widths; settlement palette edits use the existing Undo/Redo, recovery, saving and export controls.
Regional artwork uses the saved ground. Parent and refined child maps, including sculpted ground, can use those families without terrain generation. Rivers, lakes, salt flats, features, pins and routes remain. Original scientific views stay available. A base-layer change remains a recoverable draft until Save child map; restored drafts now show their selected layer correctly.
The chosen map appearance travels with exports. Open base PNG and child-map Foundry, Universal VTT and Atlas VTT exports use the selected saved art, including available refinement data. Export does not rerun simulation. Missing refinement asks for restoration first. Base PNG remains the base image; annotations follow each format's existing support.
World artwork controls follow the task. Look, Layers, Artwork and Export show one task at a time, with preview and bundle actions kept at the bottom and Undo, Redo and Save always in sight above the tabs. Map document names, copies, saved versions and detailed resolution help can be expanded when needed. Keyboard navigation works across task tabs, and a successful composition preview opens Export.

Travel

Read the journey before choosing the next step. Time, distance and key facts appear above the route map. Write scene at destination, Save to Campaign and CSV/GPX exports sit beside the result. Route alternatives, supplies/trade, the journal and departure weather are directly reachable.
Changed plans cannot reuse an old result silently. Changing endpoints, transport, season, objective or party marks the route stale and asks for a recalculation before saving or exporting it. Late calculations cannot replace a newer world or journey. Selecting the same historical year no longer needlessly reloads and discards the route. Compact layouts bring results into view and keep fields clear of enlarged headings.
One season for the whole journey. The chosen season's runoff and mud were read by calendar date, but its temperature as the northern midwinter or midsummer everywhere. Temperature now comes from the same seasonal frame, at each cell's own height, so a southern road is not frozen in its summer.
Dry stretches can be dry. The supply plan's "dry stretch" threshold was a percentile over sea and land together; the sea's zero runoff made it 0 mm, so no stretch ever counted. It is measured over land.

Manuscript and Chronicle

Chapter rows show whole titles. In the chapter rail a long title wraps instead of being cut to "A rather long cha…", and Open beside appears as a small icon in a space the row keeps for it, so a click meant for a chapter no longer lands on a button that covered half the row.
The plotline grid has room on phones again. Opening Plotlines in a narrow window shows the grid below its filters instead of a 24-pixel sliver.
More writing space and direct panels. Plan, Draft, World, Find, Style, Insights and Trash have visible tabs above a shorter toolbar. Closing a panel keeps its destinations available. Chapters has a labelled button and a narrower laptop rail; side panels dock only where there is room. Save stays on the first row while secondary fields wrap. Focus and Read remain clear.
Compact rulebooks remain editable. Very short windows use a compact Chapters/Panel toolbar while retaining the source editor and preview. Brew map/snippet choices show their complete labels. Existing formatting, Markdown, versions, split views, autosave, conflict recovery, trash and exports remain.
Chapter maps are available in both editors. Chapter tools offers the map chooser for rich writing, Markdown and rulebooks. Rich chapters retain caption and placement controls; source chapters insert portable map references. Conflicts and draft recovery prevent insertion until resolved.
Chronicle years say what they mean. An event's end year is exclusive, so "Year 20–23" for a siege that ended before year 23 now reads "Year 20–22", in the Chronicle and in reader packets; a one-year event shows one year.
Move from a Chronicle event to its places and writing. Places & writing is available in Read, Timeline, Gantt and Calendar. It shows the event's date, history branch and evidence, with direct edit, mapped-place, dossier and Write scene actions. Missing historical location evidence is stated explicitly.
Keep the exact event attached to a scene. Linked manuscript scenes retain the event and history-branch identity separately from their scene setting. Plan can reopen the Chronicle event or unlink it; Chronicle reopens linked writing. Changing a scene's year does not silently retarget the event, and unlinking keeps the prose and place details. Pending writing saves before handoffs; missing references are shown without substituting another event.

Compendium and Lore

Early-exchange proposals cite the mapped route. The suggested early corridor in a world's timeline now lists the route's kind and the length of its mapped path beside its endpoints. Traffic, cargo and age remain the author's to decide.
Market evidence stays valid when a world moves. Goods evidence now records the saved archive and the planet radius it actually reads, instead of the whole settings file. Duplicating, moving or restoring a world, or building it with more workers, no longer makes its markets look stale. Changing the planet still does.
Name and lore edits work on worlds saved without realms. Renaming places or reapplying lore no longer fails on an older or minimal world whose saved data has no realm map. Its recorded population estimate is kept and labelled as not recomputed, rather than blocking the edit.
Sections open at the top. Choosing or reselecting a Compendium section resets its reading pane, so opening Realms after scrolling Land starts at the beginning. Refreshing the same content preserves its reading position.
A clearer Lore editing desk. A section chooser leads to world details, realm records, arms, flags, diplomacy, provinces, armies, markets, events, hooks, notes and language/religion trees. Undo/Redo remains reachable; existing drafts and selections are retained. Tree diagrams sit beside their fields when space permits and stack in narrow windows.
Arms and flags have room to show their design. Larger named previews sit above grouped shield/field, ordinary, charge, pattern, palette and emblem controls. Keep/save actions and feedback have a clear footer, with flag downloads grouped separately. Intermediate-width Compendium layouts also leave more room for the record being read or edited.

Review

One desk for unfinished threads. Review is visible in the main and compact navigation. Four illustrated tabs gather Atlas freshness, naming conflicts and famous-name decisions, book continuity, and unanswered author questions. Search and paginated results keep larger projects browsable.
Go straight to the relevant editor. Open the exact Atlas sheet/lettering, Naming Studio record, book finding or inline answer. Saved answers leave the unanswered list; accepted findings stay settled. Similar names remain an author decision and low-confidence findings remain labelled.
Report gaps stay visible. Refresh reads the current reports, completed sources remain usable while others load, and unavailable/stale sources or damaged book metadata are explained. Pending writing and inline answers are preserved through handoffs. An empty queue describes the available reports, rather than certifying that the entire project has no problems.

Virtual tabletops

A more compact GM desk. Movement, fog and view controls are grouped; token position, combat details and GM notes expand separately, with Save and Cancel reachable. Connections separates scene context, session details, pins and backup. Narrow windows and enlarged text wrap the controls, and the empty table's create/import action stays clear of the inspector.
Scene entry points. Scenes offers Open tabletop even before a map has been created. Set up a blank scene brings its form into view, selects the scene name and explains the next step; Create scene opens the new canvas.

Play inside CartoVox. Open a scene → Open tabletop, choose Play in the scene library, or use Tabletop in the World Library or Campaign workspace. Child maps' linked scenes also offer Play scene. The GM desk opens in the app's workspace with its normal navigation and keeps the table when visiting Library or scene tools. It has movable tokens, grid snapping, measurement, area radius, HP and conditions, turn order, dice, fog, live token sight and doors. Open player window for a second local display that follows scene and floor changes. Connections includes parent/child scenes, map pins, linked world/lore, campaign markers, character sources and imported Atlas notes. Export/restore table state or retain it in a scene backup.

Native tabletop engine. Saved scenes now have independent table state for tokens, hit points, initiative, dice, fog and door overrides, per map and floor. Live sight uses the scene's walls, doors and token vision range. Player data omits GM notes, private rolls and unrevealed tokens. Revision checks protect edits from another window. New table state travels in scene backups and recoverable scene trash; existing Studio fog stays separate.
Connected table recovery. Quitting CartoVox checks the embedded GM desk and saves complete token and session-note edits before closing its player displays. Scene, character and world links open within the app after saving. The player display stays in a separate window for the table. Campaign and character references keep their source-world identity; linked-world backups reconnect that identity, while independent scene imports detach it. Explicit party copying carries current token details between scenes while retaining existing positions. Clear fog undo history after a backup to reclaim space without covering the map.

Atlas VTT native collections. Scene Studio and Settlement Studio can download an .atlas-collection.zip containing the background, editable Atlas walls and door states, calibrated square grid, lights and linked GM notes. Import it through Atlas VTT's Asset Manager in Obsidian desktop. Lights start hidden because CartoVox's picture already contains lighting.
Bring Atlas edits back. Scene Studio → Bring your map → Import Atlas VTT collection creates an independent scene. Choose the scene when a collection contains several maps. Supported walls, doors and radial lights become editable; imported notes and the original collection are available under Connect scenes. The original collection travels in scene, scene-tree and linked-world backups. Unsupported Atlas elements remain recoverable in that original, with conversion limits shown on the imported scene.
Atlas maps from saved geography. Hexcrawl export offers Atlas VTT for illustrated hex maps and saved child-map references, with linked Markdown notes and GM pins. These native scenes stay gridless, retaining the original geographic/hex scale in their notes; terrain supplies no invented walls, doors or lights. Atlas remains an Obsidian plugin; this integration exchanges its editable native collections and does not bundle its runtime.

Desktop

Smaller local transfers. Scripts, styles and JSON negotiate gzip when it saves bytes, with bounded reuse of encoded static assets. Decoded content and browser parsing budgets stay the same; downloads and media retain their existing delivery behavior.
Safer local HTTP access and downloads. Application API reads and writes require the desktop launch credential alongside existing write safeguards; native health and close probes remain available. Download names reject header controls and retain international names through UTF-8 encoding.
Build-time estimates learn from completed builds again. Fixed a dropped timing record by passing the actual climate worker count into the record. Estimates can distinguish serial/parallel climate and deferred city-map work.
Closing during a build asks first, and no longer freezes the app. If a world is being built (or rebuilt, or its maps drawn) when you close the window or quit, CartoVox now says so, with how far along it is, and asks whether you really want to close: Cancel keeps it open so the build can finish. Before, that question could freeze the window so it had to be force quit, and a busy build could even be missed and the app closed without asking. If CartoVox cannot tell whether a build is running, it asks rather than assuming nothing is.
500 more Delta access codes. Delta V7 accepts 551 codes, up from 51 (D-052 to D-551 are new), so Delta can open to the whole server. Every earlier code still works.

Development validation

Moisture correction has paired scientific evidence. Six ordinary compact worlds, Quick seed 991 and Full seed 60613 were generated from pinned before/after sources. Saved-field integrals, source attribution, dry channels and lake checks are recorded alongside every seed, rainfall contact sheets, material biome drift, timing and peak memory. Full seed 60613 takes 20.3 minutes versus 9.4 before, with peak memory 4.54 versus 4.31 GiB on this machine; these are bounded measurements, not universal estimates. The first mass-conserving proposal was rejected for extreme rainfall, and the slower corrected implementation's interrupted Full run is retained. All 444 focused checks passed, including parallel-season equivalence; contracts remain current. Conserving a fixed-column proxy does not establish empirical rainfall or recycling accuracy. Full-suite, native packaging and release gates were not rerun for this change.

Scientific-map endpoint coverage is restored. Direct request fixtures now include their server and HTTP-method context after the desktop credential gate changed. All 30 dynamic-map checks pass, including the two previously failing fixtures; production access checks remain enforced.
River ground and artwork agree in the tested cohort. Six compact ordinary worlds, Quick seed 991 and Full seed 60613 retain their native paths and pass dry-channel and lake checks. Corrected artwork has no measured uphill dry cell crossings in this cohort. Reference, saved replay and rendering checks passed 185 tests; two endpoint mocks that already fail at the pinned baseline are recorded separately. Paired measurements and contact sheets retain all seeds and the drawing tradeoff. This is bounded cohort evidence, not proof of every seed or lake equilibrium.
Editor API routes remain in the contracts. The endpoint scanner follows Sculpt's lifecycle-owned request wrapper only when it delegates to fetch. All 13 existing calls remain tracked, with no retirements or waivers. Contract, endpoint and feature-preservation checks passed all 44 tests; generated manifests remain current.
First-use Create guidance. Focused controller and startup checks cover one render owner, initialization without a search, and recovery after conflicting checks. Desktop and laptop browser journeys check loaded illustrations, the first Candidates visit, repeat searches and retained search history in dark and light themes. The full release and native packaging gates were not rerun for this change.
Startup budget runs without a browser. The default source gate measures eager JavaScript from parsed script tags, rejects missing or ambiguous assets, and shares the existing 3.2 MB ceiling with runtime browser checks.
Political-map dependency coverage. Render checks distinguish saved picture inputs from hierarchy preparation reads. Geometry and naming edits invalidate their views; publication bookkeeping avoids unrelated redraws.
Linux visual baselines are current again. All 75 linux-x86_64 screens are recorded (21 were stale since 28 September and 54 had never been recorded), in a Docker image proved to render as the retired ubuntu-24.04 runner did: run on that runner's own commit it matched all 33 screens recorded there. scripts/linux_visual_baselines.sh checks or re-records them from a Mac.
The map readout fix is covered by missing, loading, short, oversized and cleared-reading layout checks, plus real cursor and full-text tooltip journeys at normal and enlarged text sizes. Checks verify the readout frame, map viewport, toolbar and corner controls stay steady, and the distance ruler keeps its measured width. The full release and packaging gates were not rerun for this fix.
The World surface polish retained the same control positions and sizes in 117 before/after comparisons across map, Tools and inspector states in dark, parchment and abyssal themes. Focused layout, theme and browser checks cover shared drawer contrast, zoom/Fit, fullscreen restoration, toolbar positioning and enlarged text. No control contracts changed; this pass did not rerun the full release or native packaging gates.
The workspace layout pass passed focused controller, theme, layout, reachability and feature-preservation checks. Browser coverage includes Render Studio preview/export, Compendium navigation, Manuscript editing, Travel results, Campaign exports and player visibility, Scene Studio and tabletop inspector toggles, Make journeys, and settings/Layout keyboard access across desktop, laptop and phone sizes, including enlarged text. A phone writing-space regression found during the pass was fixed; decorative Manuscript tab icons stay on wider screens. Public control contracts are current. The full release gate and native packaging were not rerun.
The tectonics layout and candidate descriptions passed focused controller, browser layout, theme, reachability and feature-preservation checks. Browser coverage includes keyboard section switching, enlarged text, drawing, Undo, conflict reporting and Apply. A local application check exercised drawing and applying plate seeds, plus a real candidate search in dark and parchment themes. Public control contracts remain current; the full release gate was not rerun for these changes.
The larger candidate grid and combined Build view passed focused layout, controller, reachability and feature-preservation checks. Browser coverage includes desktop and phone layouts, seed correction, imported-map previews, candidate switching and restored sessions. Three application journeys passed: a candidate search through a complete Quick-world build, a known-seed preview with atlas options, and an imported-map flow. Public control contracts were regenerated and checked; the full release gate was not rerun for these changes.
The full unit suite is green again. Make is registered as a routed workspace, the Deep Time lifecycle test expects the empty starting canvas the flow now opens on, and a manuscript test no longer fails when a random note id happens to begin with the id it was counting.
Every preset and five extremes were built at the shipping size with settled water: no river climbs uphill and every saved world validates. Two cases (a small planet and an ocean world) exposed a lake-join crash, fixed before release.
The browser release gate was repaired and run four times in full. About 180 journeys had fallen behind the 7–8 October interface changes (Render Studio's task tabs, the Lore editor's sections, moved scene and Sculpt controls); each was updated without loosening what it checks. The repair found more than a dozen real interface faults, listed above, from a viewpoint pin hidden by the preview to candidates that could not be picked during a search. The last run passed 2,321 journeys with two failures: one journey needed a longer search and is fixed; the other has not recurred in five reruns. The full unit suite passed (14,309 tests), and the 21 visual baselines changed by deliberate design were re-recorded.
The fast full-world calibration baseline was re-recorded for Delta V7. Every material drift since the V6 baseline is attributed to specific commits in docs/RELEASE-READINESS.md (area-weighted means, climate on the finished terrain, settled lakes, the corrected water-balance reading); the structural baseline is unchanged. Hack's exponent at default size now reads above its Earth band and stays visible as a health warning.
The freshwater flow reference now handles direct ice sublimation, alongside liquid evaporation and received vapor. Both outgoing pathways use the same explicit vapor enthalpy convention, with separate mass and heat accounting. Ice can lose mass while cooling without first melting; liquid cannot satisfy an ice-loss request until freezing forms ice. The previous mode remains exact. Validation includes 91 focused cases, a broader 1,785-test run, and 56 retained short intervals at three subdivisions. The new annual reference completes 18 of 28 scenarios; ten exceed its supported thermal domain, with every failure retained. This is reference groundwork: modeled weather/material inputs, atmospheric availability, surface/phase/area feedback and production lake/river integration remain open.
The freshwater flow reference now accepts received vapor with explicit mass and full enthalpy. Condensation heat is counted once, incoming water can warm or melt ice, and dry condensation requires an explicit heat sink. Analytical checks, 48 focused vapor cases and 56 retained short intervals at three time subdivisions verify the accounting. All previous no-vapor results and rejected boiling cases remain exact. This is reference groundwork; atmospheric water availability, actual heat forcing, surface feedback and annual production lake integration remain open.
The surface-heat reference now has explicit ideal-mixture humidity and air density conversions, plus separate liquid-water and ice saturation fits. Absolute local pressure, vapor partial pressure and gas properties are required; the climate's normalized moisture proxy is never treated as specific humidity. Published reference values and the retained native/cold scenarios verify units and phase distinctions. Signed inward vapor now has a received-mass/enthalpy reference operation; production condensation and deposition still need weather and lake integration.
A new explicit surface-energy reference separates absorbed sunlight, incoming/outgoing longwave radiation, sensible heat and signed potential vapor transfer. All atmospheric/material inputs are required; missing humidity or surface radiation is never replaced by normalized climate proxies. Analytical checks and 56 short intervals across the retained native/cold cohort verify components and water/heat budgets, including full resolution. This is fixed-coefficient reference work; physical weather prediction, phase feedback and annual production integration remain open.
The freshwater flow reference now mixes incoming liquid/snow, evaporation and heat-carrying liquid withdrawal on one clock. Cold stores cannot supply liquid prematurely, and outgoing water carries its changing temperature. Analytical and retained native scenarios verify the accounting. Five supplied- heat scenarios reject unsupported boiling, exposing why potential evaporation alone cannot set atmospheric heat. This remains development reference work; production lake integration and physical heat forcing are still open.
The freshwater surface reference now advances prescribed heat and evaporation together. Vaporization pays its heat cost, warm water can cool and freeze, and cold ice warms before melt-mediated vapor loss. Dry columns report heat they cannot absorb. The original forcing clock survives subdivisions and drying boundaries. Analytical checks and 28 retained native surface scenarios verify water/energy accounting; atmospheric forcing, moving lake geometry and production integration still require qualification. Existing worlds are unchanged.
The replacement lake model now has an explicit freshwater phase and energy reference. Frozen water remains in storage, warming and melting require sensible and latent heat, and liquid withdrawal carries its heat without spending frozen mass. Analytical checks and fourteen retained native column scenarios, including the full-size world, verify conservation and time subdivision. These scenarios prescribe initial energy and external heat; atmospheric heat, moving shorelines, groundwater return and production lake integration remain open. This does not add predicted lake ice to saved worlds.
Seasonal UV reading 2 and optional snow-reflection model 1 pass analytical growth/thaw, seasonal sunlight, unequal-orbit rainfall, damaged-save and legacy checks. Fourteen native comparisons and fourteen complete compact builds preserve all existing hydrological/simulation fields, while recording local changes in UV. Both variants render from actual saves, and six browser journeys cover both themes and three window sizes. Derived drawing caches refresh for the new reading. UV remains a heuristic with Earth-like ozone, a Sun-like spectrum and a bounded snow-reflection proxy; this does not validate physical albedo, glacier age or lake ice.
Snow model 2 records initial storage and retained annual accumulation, with archive checks for consistent end storage. Analytical extremes, fourteen paired native calculations and fourteen complete compact builds verify the conservation change. Runoff and the other existing saved fields remain exact across these builds. Seasonal reflection fields now keep new UV readings aligned with this snow cycle; older worlds disclose their end-storage proxy until rebuilt. Physical albedo and lake ice remain further validation work. This is not glacier calibration.
Repeated basin preparation in the experimental lake integrator is now reused within each calculation. The same full-size recorded interval took about 105 seconds before and 44 seconds after, preserving depths, water budgets and all spill contacts. Changing inflow, water state and forcing still receive validation. Seasonal settling, broader performance and production integration remain open; this does not change the application's current lake model.
Shallow groundwater now starts from the repeating seasonal recharge state rather than an arbitrary ten-year warm-up. This corrects understated storage and baseflow at slow drainage settings, while keeping the existing snow, infiltration and recession equations. Saved metadata distinguishes the new groundwater model from legacy results. Paired native and complete compact world builds verify water accounting and downstream changes; existing worlds retain their saved values until an explicit rebuild.
Longer fixed-forcing checks now establish numerical annual repetition in all six retained compact native lake cases, after 246–9,377 modeled years. The original shorter-bound failures are preserved. The coupled reference also removes the actual retained two-year shoreline cycle; finer seasonal steps and 100-year continuations remain stable at numerical precision. These checks qualify the experimental calculation, not physical lake equilibrium. Full-size annual validation, physical forcing and production integration remain open.
The experimental lake model now integrates actual nested basins on a shared event clock: overflow arrives while losses occur, full basins join at saddles, and receding lakes separate again. Receiving-water contacts follow shoreline changes. An exact-head check also fixes a numerical event stall. Analytical, subdivision and seven retained native-world comparisons pass. All six compact annual experiments still accumulate water after 40 years, and full-size performance remains an open gate. This advances the earlier reference work; production lakes and rivers have not switched to it.
The replacement lake model now has a tested drainage core that divides water along actual downhill ground into real basins. Focused checks and seven native terrain comparisons include a full-size world. Seasonal experiments still fail to reach a repeating annual state within the tested bound, so this remains internal groundwork; it has not replaced production lakes or river routing.
The experimental lake model now removes evaporation and seepage from each connected pool as its level falls, separating basins when their saddle dries. This corrects an artificial wet equilibrium caused by capping losses in each cell before redistributing water. Analytical and native-world checks are recorded; production integration and seasonal qualification remain unfinished.
A separate reservoir exchange calculation now follows shoreline changes within an interval and spills only net incoming water. It removes the observed two-year numerical cycle in a verified isolated test basin and agrees under time subdivision. Coupled lake networks and production routing still need integration; this is not a replacement for the application's lake model yet.
Experimental reservoir networks now preserve when upstream overflow reaches downstream lakes. Analytical checks and 856 eligible native basins verify water accounting and time subdivision. Merging lakes, backwater and full production integration remain unfinished; this is internal model groundwork.
The reference reservoir calculation can now pause at an exact saddle level, returning the actual elapsed time and water budget. Analytical join/split and native leaf checks verify pause/resume; automatic merging lake networks and production integration remain unfinished.
Reference spill traces now distinguish dry channels from submerged paths, marking where they meet a receiving lake or the ocean. All 1,100 checked native transfers preserve their water and original routes. Contacts describe the start of each transfer; moving shorelines and public river integration still require the remaining model work.
The Render Studio/Oblique tester follow-up passed 110 focused source and module checks, 85 shared layout/stylesheet/theme checks, and 15 browser journeys with six intentional viewport skips. Coverage includes actual PNG downloads, zero and raised relief, camera/shading changes, unchanged source terrain and artwork, retry, delayed-response ownership, session settings, keyboard focus and desktop/laptop/compact layouts in both themes. These are development checks; no installer was built for this follow-up.
The application overhaul has a section-by-section implementation and evidence record in docs/application-review-implementation.txt. The earlier consistency pass is recorded in docs/visual-consistency-pass.md. Checks used disposable worlds and documents, not edits to the author's saved projects.
The main review passed 222 distinct browser integration cases and 70 visual checks against reviewed Darwin/arm64 captures, alongside focused source, contract, theme, stylesheet, calibration and determinism checks. The earlier consistency pass also reported zero violations in 24 automated accessibility probes across 12 surfaces and both themes at 125% text; this is bounded evidence, not complete screen-reader qualification.
The corrected full-source run reported 9,834 passed, 14 failed and 2,322 skipped across 12,170 tests, plus 85 passing subtests. All 14 failures passed a fresh rerun at review runtime commit 3cc11565 after fixes. Fixes landed during the broad run, so it is not recorded as a clean full-suite run on one final commit. Archived review test copies were removed from normal test discovery.
Later feedback has separate focused coverage: Library frame removal passed 12 visual comparisons; Deep Time/Create/Sculpt passed 15. Make's own workspace passed 161 focused checks including 24 guide journeys, then all 20 expanded Make module checks and 30 visual comparisons across the five workspaces. Real small-world builds verified recipe-dependent names and preservation of supplied names; a full drawn-world journey retained its named towns and road.
These records cover source and browser development work. They do not replace final-candidate release checks, native installer verification, Linux visual qualification or human usability feedback. Delta V7 remains in development.
The starter-pack removal passed 247 focused checks and three responsive Library journeys. Existing adopted worlds retain their metadata and numerical data through opening and backup/restore. About 227 MiB of bundled resources were removed; compressed installer savings await a new desktop build.
The child-map changes passed focused source checks and browser journeys on desktop, laptop and compact layouts, including existing nested counties, refinement, sculpting and regional hotspots. The two-hour full suite was not run during these implementation sections; full V7 release qualification remains a separate step before release.
The tester-feedback sections passed focused source and responsive browser checks. Fast generation matched every numerical archive member, names and lore in a small real-world comparison. Atlas collections passed the pinned upstream manifest validator and CartoVox download/import journeys; running the collections inside Obsidian/Atlas itself remains unqualified.
The native tabletop passed focused engine, desktop bridge/close, backup, preference and compatibility checks, plus 18 browser journeys across three layouts covering play, connected worlds, legacy fog/library and Child maps. Installed desktop and installer qualification remain release checks.
The in-app tabletop correction passed host/navigation, desktop close and feature-preservation checks, plus 18 responsive browser journeys covering blank-canvas setup, in-app launch/resume, play and Child maps connections.
The installer check presents the app's window key. Delta V7's API asks for the per-launch window key, and the packaged smoke test sent it only for the interface's own files, so every installer's verification stopped at its first API request. It now sends the key on every request, as the app's window does.