Verification · 2026-09-11

MagIC doubled horizontal resolution · 2026-09-14

Both supported solver grids now double longitude and latitude counts, retaining their radial levels: 128 × 64 × 17 and 192 × 96 × 25. The default is 128 × 64. The map doubles to 768 × 384 pixels and the globe to 512 × 512, retaining their 2:1 and 1:1 display aspect ratios. The existing Wasm binary allocates both grids at runtime; no numerical Fortran source or compiler changes were required.

node --test scripts/test-magic.mjs passed all three checks. Both actual Wasm grids completed 1,000 steps and returned 25 finite, evolving frames, reaching model time 0.08. The standard run verified pause/resume; the finer run used the deep-shell preset with five density scale heights. Boundary and circulation checks passed. Local elapsed times were 37.5 and 162.3 seconds respectively; retained frame fields occupied 53.1 and 175.8 MiB (not total process memory).

All 13 UI/integration checks passed, including the doubled map and globe draw dimensions, 128-longitude/64-latitude exported coordinates, 8,192 values per shell field, replay/stop and rocky-tab regression checks. Syntax, local asset references and unique HTML IDs were checked. No browser/device performance test or numerical convergence study was performed; phone runtimes may be longer.

MagIC gas-shell integration · 2026-09-13

The unmodified vendored MagIC source matches upstream revision 0d408831fd38ac72c52d53dd091a3ccaabce5207 byte for byte. Browser integration patches and compiler/runtime ABI adapters are documented in model/magic/BROWSER.md. The 2.7 MiB Wasm engine uses double-precision calculations and single-precision G-file output. Asyncify instrumentation is restricted to the yielding call path; the pinned optimizer inlines the timestep driver and C yield hook into browser_run.

node --test scripts/test-magic.mjs passed three checks:

• Bounded input validation and native namelist generation.
• The actual MagIC worker completed the default 1,000-step 64 × 32 × 17 shell experiment, producing 25 finite, evolving field frames. Model time reached 0.08 viscous units. Pause held output and resume completed the run. The sampled outer-boundary radial velocity was zero to within 1e-5, and interior circulation was nonzero. Approximate elapsed test time: 8.3 seconds on this machine.
• A 96 × 48 × 25 deep-shell experiment with five density scale heights completed 100 steps, yielding 11 finite frames. The workers ran with all fetch and importScripts calls forbidden; engine bytes came from the parent context.

node --test scripts/test-magic-ui.mjs scripts/test-resolution.mjs passed 13 control/integration checks. These cover gas-tab selection and return to rocky controls without losing terrain; field-map/globe drawing calls; replay selection; dimensioned JSON export; pause/stop; cancelled loading and stale messages; and the existing terrain, display-resolution, palette and climate-control regressions. The final HTML has unique IDs, paired accessible tabs, and valid local assets.

These checks establish browser-compatible execution and the field/control handoff. They do not establish native/Wasm numerical equivalence, grid convergence, realistic gas-giant weather, stable behavior across the full parameter range, or iPhone/browser UI performance. No browser automation or physical-device test was performed.

This is a compiler/port check, not climate-model validation or convergence testing. No browser UI or physical iPhone test was performed. Timing below is from this Linux execution environment and must not be read as a promise of phone performance.

Original solver with display detail and cloud shell · 2026-09-11

The browser now always loads the original T21 64 × 32 solver with 30-minute steps. The temporary T42 assets were removed. Both display settings keep the same physical terrain, namelists, solver memory allocation and 2,048 values per field. High res doubles only the terrain display, map texture, globe and cubemap face dimensions. Changing resolution still resets loading, computation and replay. The retained T21 binary is unchanged from the earlier successful ten-day check.

Twenty-one application, cloud, terrain and world-generation checks passed:

• Both display selections preserve the original physical grid and timestep; terrain and climate exports distinguish display dimensions from physical data.
• Reset cancels pending downloads, running/paused workers and replay, clears results, and ignores stale callbacks. Extreme estimates remain 64 × 32 too.
• A High-res climate cloud cell controls exactly four display cells. The existing RFL 3D noise is thresholded within that group while keeping its mean coverage within 1e-7 of the raw model value. Zero coverage stays clear, full coverage stays opaque, and the source climate arrays are unchanged.
• Cloud sampling is repeatable, changes with displayed simulation time, wraps around longitude, and covers both poles. The raw scientific cloud field never receives procedural detail or the decorative shell.
• Direct pixel-array checks at 256 × 256 and 512 × 512 show clouds outside the ground silhouette and no cloud pixels beyond a radius 1.14 times the ground radius. Clear clouds leave the original background and ground unchanged; scientific layers remain unshaded with no cloud overlay.
• The two packed globe depth buffers use at most 2 MiB at High res. Its cloud alpha/shade/latitude-edge arrays use about 41 KiB. These are individual buffer checks, not measurements of total browser or device memory.
• The palette editor selects pixels, changes colors, opens a BMP, and preserves all twenty edited pixels when saving BMP or XPM. Editing during a model run leaves the run intact. Sprite row order, opacity checks and PNG dimension rejection pass. PNG decoding/encoding uses native browser image APIs and has not been exercised in an actual browser during this update.
• Existing terrain/C-noise agreement, cubemap seams, bounded replay and 5,000 seeded random-world checks pass, as do 250 draws with High display selected.

Five loader/worker checks pass: both display settings download the same original engine, invalid responses fail cleanly, cancellation releases the worker URL, and a compiled-grid mismatch is rejected before allocating output arrays. A sixth check executes the real retained engine with High display requested: 48 steps (one Earth day), 1,800-second timestep, 22 finite field updates, pause/resume and no worker networking, in 2.7 seconds on this Linux environment. The output land mask and Gaussian coordinates exactly match the original terrain; elevation versus final surface pressure has a correlation of −0.928.

App checks use a JavaScript VM with represented DOM/canvas interfaces; the globe geometry check inspects pixel arrays directly. No browser visual or physical phone test was performed. The deliberately oversized cloud sphere and its noise detail are appearance choices, not simulated cloud altitude or resolved cloud structure.

Native comparison · 2026-09-10

Compiled the same adapted model with GCC/GFortran 13.2.0 (native x86-64) and r-wasm Flang 21.1.8 + Emscripten + the descriptor-compatible runtime (WebAssembly). Default single-precision computation, T21L10, same boundary data and namelists: 1367 W/m², 360 ppm CO₂, one Earth day rotation, 30-minute timesteps, KICK=0. The native build includes the same calendar-call and hurricane initialization repairs.

After 48 timesteps (one simulated Earth day), across all 2,048 grid cells:

Field	RMS difference	Maximum absolute difference
Surface air temperature	0.00063435 K	0.00534058 K
Total cloud fraction	0.00015867	0.00484091
Lowest-level eastward wind	0.00032827 m/s	0.00583076 m/s
Lowest-level northward wind	0.00025756 m/s	0.00451183 m/s
Surface pressure	0.01747355 Pa	0.30468750 Pa

At the first timestep, temperature RMS difference was 0.00001723 K. The two builds agreed on elapsed steps, 1,800-second timestep, and shutdown flag. Small differences are expected across compilers; chaotic weather can diverge over long integrations. This check is not proof of full physical equivalence.

Worker integration

Executed the website’s worker and generated engine in a JavaScript VM with the browser worker interfaces represented locally and real bundled files served through the fetch adapter. This tests computation and messaging, not browser layout or a particular browser engine.

• Default ten-day run completed all 480 timesteps with finite output fields.
• Temperature range at completion: 227.51–300.16 K; cloud fractions: 0–1.
• Pause at step 12 held the model state; resume completed the run.
• 214 field updates; approximately 19.7 seconds of wall time on this test machine.
• Separate one-day experiment with 1500 W/m², 800 ppm CO₂, and two-day rotation completed successfully and produced different results from the default.
• Invalid configurations were rejected; stellar-flux, CO₂, and rotation controls were checked against their generated Fortran namelists.
• JavaScript syntax and local public asset references were checked.
• WebMCP is feature-detected; no supported browser context was available to validate registration. It is optional and ordinary UI operation does not depend on it.

Limits

Long climate equilibration, the full parameter range, long-run energy balance, all browser engines, memory limits, mobile suspension, and iPhone performance have not been validated. Cold-start transients can be strong. JSON exports contain only the displayed sampled fields, not a complete history or a restart checkpoint.

Rocky Atmo interface update — 2026-09-10

Checked eight planet presets, nineteen atmosphere palettes, every planet/atmosphere climate-eligibility combination, and 2,500 seeded random selections. All randomized model configurations passed the existing ExoConfig bounds validator; duration was preserved and all eight planet classes were reached. JavaScript syntax, HTML control references, unique IDs, and local asset references passed. The Fortran engine, worker and physical namelist generator are unchanged. No new browser UI or mobile-device testing was performed for this interface update.

Worker startup repair — 2026-09-10

The reported The climate engine could not run: undefined comes from the page’s worker error handler reading a missing event message. That message alone does not identify why the user’s browser rejected startup. Authenticated HTTP checks of the deployed page, worker, configuration, engine script, Wasm and boundary data all returned 200 with the expected content types and complete asset sizes.

The new loader removes worker-side authenticated script/data requests: the page loads the assets through its own session, constructs a Blob worker, and transfers the binary buffers. The compiled Fortran model and boundary data are unchanged.

node --test scripts/test-engine.mjs passed four regression/integration tests:

• All five assets loaded with same-origin credentials before worker creation.
• HTTP 401/404, a returned HTML login page and invalid Wasm produced actionable errors.
• Cancelled loading created no Blob URL; message-less events never displayed undefined.
• The actual compiled model completed 480 steps (ten days), with 190 finite field updates and pause/resume, in a VM worker context where fetch and importScripts throw. Transferred binary buffers were detached on the sending side.

This verifies the loading contract and model execution, not the reported device’s original failure cause. No browser UI or physical iPhone test was performed.

Constrained random worlds and extreme surface fields

node --test scripts/test-worlds.mjs passed four checks:

• All 152 planet/atmosphere pairs are classified: 3 ExoPlaSim combinations, 9 simplified surface-estimate combinations, and 140 appearance-only pairs.
• 5,000 seeded draws all selected a supported model and respected coupled bounds; all eight classes occurred within two percentage points of their authored priors.
• All extreme outputs had finite positive surface temperatures and finite nonnegative pressures. Unmodeled air temperature, cloud and wind channels stay unavailable and are omitted from exports. The discretized energy budget closes to a relative error below 1e-10 before Float32 output conversion.
• Every extreme reference, random endpoint, bounded fallback and targeted phase screen passed. Boundary testing caught and fixed a floating-point log-pressure sample just outside its lower bound. Higher sunlight raises ice temperatures; higher prescribed pressure raises the Venus-like grey greenhouse temperature.

Counts in the deterministic 5,000-draw check: temperate 1779, warm 725, cold 767, airless ice 386, volatile ice 387, Venus-like 410, steam 294, silicate 252.

node --test scripts/test-engine.mjs passed six tests, including loader regression checks and actual Wasm execution with worker networking disabled:

• Default 1367 W/m², 360 ppm, one-day rotation: all 480 steps (10 Earth days).
• Cold random boundary 1050 W/m², 400 ppm, 0.8-day rotation: 48 steps (one day).
• Warm random boundary 1500 W/m², 300 ppm, three-day rotation: 48 steps (one day).
• All runs produced finite fields and supported pause/resume.

JavaScript syntax, HTML IDs, app control references and local asset links passed. These tests check the implementation and short model runs. They do not establish extreme-climate realism, validate the grey opacity or vapor-pressure proxy, ensure long-run ExoPlaSim stability, or test a browser UI or physical phone.

RFL terrain, cubemap and playback — 2026-09-11

The engine/world suites passed 12 tests: protected asset loading (now six assets), cancellation/errors, existing baseline and cold/warm boundary runs, 5,000 random worlds, and two actual Fortran runs over RFL-generated terrain. The new default terrain completed 480 steps (ten days); seed 2468 with 5 km relief scale completed 48 steps. Both returned exactly the generated land mask at the expected Gaussian latitudes and supported pause/resume. A focused one-day rerun verified elevation versus actual surface-pressure correlation of −0.922, confirming that terrain affects physics rather than only appearance.

Five terrain/format/replay tests passed:

• 128 samples agree with original compiled RFL C noise to within 1e-7.
• Deterministic and changed-seed terrain, longitude wrap, Gaussian weights, geopotential units and bounded inputs.
• Edited BMP and XPM palettes round-trip; malformed palettes fail.
• Bounded replay keeps chronological frames and first/latest endpoints through 5,000 updates; pause, seek, restart and return-to-live state are checked.
• All twelve cube edges and eight corners meet across adjacent face directions; neighboring front/east textures use identical edge samples.

Static HTML references, paired tabs, unique IDs, balanced sections, same-origin assets and JavaScript syntax were checked. No browser UI or physical phone test was performed. Cubemap display does not increase the 64 × 32 physical resolution; long-run stability for arbitrary edited terrain is not established.

Terrain Solver integration — 2026-09-11

• The C99 algorithm-2 core/header/knob definition match the previously delivered Terrain Solver source byte for byte. Emscripten 3.1.6 + LLVM 15 compile the core and a thin browser ABI, independently of the existing Fortran engine.
• Native C and Wasm agree at 64 world directions for both Balanced and Regional: heights within 1e-6 m, exact water/biome/palette classifications, and moisture within 1e-12. Identical browser inputs reproduce all sampled arrays and metadata.
• Balanced, Archipelago, Alpine and Regional produce finite samples, both land and water for seed 42, valid palette indices, normalized Gaussian weights, correct low/high/square dimensions and macro slope excess below 1e-6 m. Changing the seed changes the land mask; malformed configurations and boundary arrays are rejected.
• The actual solver worker instantiates Wasm and transfers its generated arrays in a VM context with worker downloads/importScripts forbidden. This is a numerical and worker-protocol check, not a physical-device browser test.
• Control integration verifies independent workspace/generator tabs, low/high display projection, original 64 × 32 climate inputs, cached solver reuse, restoring Quick, cancelled/late/error results, blocked stale climate starts, invalidation of completed/replayed fields, and unchanged palette/cloud controls.
• The actual bundled ExoPlaSim ran 48 steps (one day) over Balanced seed 42, with 17 finite field updates and working pause/resume. The output land mask exactly matches Terrain Solver input and Gaussian coordinates agree within 1e-7. Elevation/surface-pressure correlation: -0.899. This validates the terrain handoff; it does not establish long-run stability for every procedural configuration.

Saturn cutaway (2026-09-14)

node --test scripts/test-saturn.mjs verifies occlusion and section-wall depth, all five exposed layers, rotation, both polygons through the water deck, flowing textures, XPM geometry/defaults, responsive glass/reflection settings, context controls, retained edits, reduced motion, inactive-tab scheduling, and animated remove/restore returning to the identical paused image. Rendered pixel outputs were inspected directly without running a browser. HTML IDs, local assets and module imports were checked, alongside JavaScript syntax and the unchanged shared MagIC UI checks. No live browser QA was performed.

Jupiter layer preset — September 14, 2026

node --test scripts/test-saturn.mjs: all nine tests pass, including the five existing Saturn checks and four Jupiter checks. They exercise the real renderer and controller with a DOM adapter, without starting a browser or server.

• Every one of Jupiter’s six layers is visible in the open cutaway. Closed, opaque outer clouds completely occlude changes to internal emission.
• The closed blue-green glass shell visibly transmits red-orange emission at the default maximum IOR (2.5) and midpoint scatter (0.5). Pixel-region checks distinguish the warm center, cool glass edge, and emission-off state.
• Jupiter textures animate without a longitude seam; the Great Red Spot has an independent strength control.
• Switching Saturn/Jupiter preserves independent layer edits, core emission, zoom and selected layer. Remove/restore recovers the same paused Jupiter view. Resetting Jupiter does not discard Saturn’s edits. Hidden views stop drawing.

Direct renderer images at 512 × 448 were inspected for whole Jupiter, its six layer cutaway, exposed glass, and reflective glowing center. Existing mobile layout and controls are reused. No browser QA or numerical-model convergence claim is made by these appearance checks.

2026-10-10 · Extreme-world weather and tidally locked super-Earths

Added diagnostic thermal-gradient/drag/Coriolis surface winds and condensable-supply cloud potential to atmospheric extreme estimates; airless fields remain unavailable. Wet and dry locked rocky presets expose radius, gravity, rotation/orbital period, pressure, drag, fixed substellar longitude and condensable supply. The cloud shell, streamlines, field units and exports preserve estimated-model provenance.

node --test scripts/test-extreme-weather.mjs scripts/test-worlds.mjs scripts/test-streamlines.mjs scripts/test-clouds.mjs scripts/test-resolution.mjs passed 37 tests, including 5,000 seeded random-world draws, seam/pole finiteness, locked illumination shifts, wind response to drag/rotation, dry-cloud suppression, equal-area cloud alignment, field rendering, export availability and existing streamline transitions. Application integration uses a Node DOM/canvas harness. No browser UI or physical-device QA was available. These checks verify implementation, not physical calibration: diagnostic wind/cloud patterns are not a GCM solution.