# Language and notation validation — September 17, 2026

- 18 regional language choices, 18 local SVG flags, and 17 non-English offline dictionaries. Every inventoried UI phrase is present in each dictionary. English is the canonical source.
- Source/model validation: **56 checks passed**, including dictionary completeness, flag files, and invalid-glyph checks.
- Localization browser suite: **6 passed** with the completed packs. Covers all 18 choices, flags, translated titles, saved preferences, reversible English, simulation-state preservation, unchanged equations, German decimal commas, Arabic digits, invalid-entry atomicity, unsaved decimal edits across language changes, mobile menu bounds, localized search, explanation modes, embedded messaging, and offline standalone navigation/refresh.
- Existing Equation lab suite: **41 passed** across all 37 worlds after the locale-aware input changes. The final Arabic-edit conversion received an additional focused passing check.
- Offline `file://` sweep: all **37 labs** opened with French or German text at phone width, with no page errors, no horizontal overflow, and correctly parsed numeric fields.
- Desktop Spanish, mobile Arabic, and standalone German previews were visually inspected. See `previews/language-*.png`.
- Initial browser runs encountered local HTTP request stalls. The test server now uses a larger connection backlog; subsequent localization and equation suites passed. Historical unrelated legacy-suite limitations are preserved below.

Translations are machine-assisted, supplemented with manually written navigation, world titles, all new experiment names, and selected terminology/explanations. Coverage checks do not constitute a complete native-speaker or scientific translation review. English remains the reference. Standard equations, experiment implementations, and existing thumbnail files are preserved.

---

# Requested-topic expansion validation — September 17, 2026

Final delivery: 12 added worlds, 43 new experiments, 37 total worlds. The proposed infinite-well experiment was removed after confirming that the existing Qubit Box already implements the n = 1 and n = 2 infinite-well states.

## Passed checks

- `npm run check`: local assets, IDs, and JavaScript syntax for 42 active HTML entry points; all **55 model/runtime checks passed** (46 existing plus 9 new).
- Dedicated browser suite: **13 tests passed**, exercising **all 43 final experiments**, their parameter minima/maxima, pause, step, reset, numeric equation entry, and phone-width overflow. Rerun after removing the duplicate well experiment: 13 passed in 21.3 seconds.
- Broad Chrome suite: **182 / 188 passed** in the initial run. Three hardcoded-port failures subsequently passed after test URLs were made compatible with `PLAYWRIGHT_PORT`. Thus 185 distinct checks passed across the full run and targeted retry; this is not a claim of a clean single-run suite.
- New worlds passed the existing standalone-load, offline asset, phone-layout, and model-note tests in the broader suite. Library rendering and navigation worked with 37 catalog entries.
- Manual screenshot inspection: desktop double pendulum, atomic density, library, and full-height phone layout. Offline `file://` rendering also exercised by preview capture. Previews are saved under `previews/additions-*.png`.
- Quantitative checks include momentum/restitution/energy loss, Kepler limits, double-pendulum energy conservation, closed diffusion mass/entropy, subluminal velocity addition, harmonic-oscillator normalization/parity, standard Life patterns, lens limits, and Mandelbrot known points.

## Remaining legacy-suite failures

1. **Knot workspace pause:** the test immediately samples state after sending an asynchronous iframe pause message; the staged suite observes a small progress increment. The original app/runtime/knot source files are unchanged. The baseline attempt timed out during navigation, so the exact timing failure is not claimed to have been reproduced in that baseline run.
2. **Sound Frequency model-note disclosure:** the existing test expects a collapsible `summary`, while the original page has inline model notes. Reproduced against a separate copy of the original collection.
3. **Wind object-preview drag:** the existing test's mouse drag does not change the yaw value. Reproduced against a separate copy of the original collection.

These issues are documented rather than changing the existing experiment implementations as part of a missing-topic expansion. All new-experiment checks pass. No network is needed to run the additions; optional reference links open external sources only when clicked.

---

# Validation — updated 2026-09-17

Tested on macOS with Google Chrome 153.0.8010.36, Playwright 1.63.0, Node.js 26.7.0, and Python 3.14.

## September 17 — manual equation inputs and larger stars

- All 30 active HTML files pass local asset/link checks, unique IDs, and JavaScript syntax validation. All 46 model/runtime unit tests pass.
- 73 distinct browser checks pass across the regression run and focused reruns, including 29 new checks covering all 25 worlds, actual state changes, invalid-value rejection, mobile layout, dynamic tabs, and offline operation.
- After separating editor loading from homepage previews, all 29 equation checks and the legacy redirect check pass together (30 checks, 15.7 seconds). The redirect had timed out while an abandoned preview was fetching shared editor assets; preview frames no longer fetch those assets.
- Checked qubit mass against the inverse-mass energy law, gravity depth against the surface value, and image changes in the chaos and sound worlds. Verified native slider synchronization and validation before applying a batch of edits.
- Visually inspected the equation panel at desktop and phone widths. Homepage star radii increased by approximately 25%, with the existing density and motion preferences retained.

## September 17 — live carousel, directory, and Sound Frequency

- `npm run check`: all 30 active HTML files passed local-link, duplicate-ID, and JavaScript syntax validation; all 46 existing model/runtime tests passed.
- 44 browser checks passed across the collection regression run and targeted reruns after fixes. These include all 25 worlds on desktop and phone, offline file loading, library navigation, and six new carousel/directory/sound checks.
- Verified one visibly animated preview from each category, automatic rotation, manual arrows/dots/keyboard/native scrolling, reduced-motion behavior, preview pause and disposal, collapsed collection links, and returning from a lab during carousel navigation.
- Verified pitch and resonance change the membrane image, presets, zero drive, pause/reset, optional audio frequency updates, and silence when either pause control is used.
- Inspected desktop and phone screenshots of the library and Sound Frequency world.
- Corrected the existing touch test to use the configured server and bring the canvas into view before tapping; its targeted rerun passed.

## Results

| Check | Result |
| --- | --- |
| Higher-dimensional explorer and regressions | **18 browser checks passed**, 16.4 seconds; all 8 existing tesseract checks also passed in the earlier dedicated run |
| Hypercube geometry | **8 passed**; all **22 numerical/scheduler/mesh tests passed together** |
| Final explorer landscape refinement | **1 affected browser check passed**, including offline desktop and three phone viewport sizes |
| Wind Tunnel vehicles and import | **16 focused browser checks passed**, 21.5 seconds |
| Final model-scale and toolbar refinements | **11 affected browser checks passed**, 14.1 seconds |
| Mesh parsing and projection unit tests | **6 passed**; all 14 numerical/scheduler/mesh tests passed together |
| Time Observatory and collection integration | **10 focused browser checks passed**, 15.4 seconds |
| Time-dilation model tests | **5 passed**, including analytic and precision limits |
| Original updated-edition browser suite | **37 passed**, 35.3 seconds |
| Subsequent Wind Tunnel fidelity update | **8 focused browser checks passed**, 11.0 seconds |
| Subsequent tesseract camera update | **8 tesseract tests and 3 collection regression checks passed** |
| Individual world thumbnails | **4 existing browser checks passed**; 24 distinct illustrations, with both Wind Tunnel variants unchanged |
| Final 3D panel styling and refreshed screenshots | **3 targeted checks passed** after the complete suite |
| Scheduler regression tests | **3 passed** |
| Source, local assets, links, and duplicate IDs | **28 active HTML pages and all application scripts passed** |

The original complete browser run covered the then-registered 24 worlds. Each opened at a 1440 × 1000 desktop viewport, exercised its visible buttons, and then checked layout and control bounds at 390 × 844. External asset requests were blocked and treated as failures. No JavaScript errors or missing assets were reported. The four phone-layout failures discovered in the initial run were fixed before the complete passing run.

The Time Observatory update adds world 25. Its ten focused browser checks cover all three experiments, clock readings at known rates, rest/airliner/near-light-speed cases, Earth surface/one-meter/GPS heights, invalid input handling, reset and pause, accelerated playback, shared workspace suspension and resume, frame cleanup, Retina rendering, and offline desktop/phone layouts. They also rerun the existing new-world control smoke check, library navigation, phone/unavailable-storage behavior, all-world offline loading (now 25 worlds), and screenshot capture. The remainder of the older browser suite was not rerun for this addition.

Five tests in `tests/time-model.test.js` verify Lorentz factors at 0, 0.6c, and 0.8c; the low-speed expansion; a half-rate hovering clock at 4/3 Schwarzschild radii; near-horizon/far-away limits; equal-height and GPS-height Earth clocks; a one-meter weak-field limit; accumulated clock readings; preservation of tiny offsets; and invalid argument rejection. These passed together with the three scheduler tests. They validate the stated idealized models, not a full treatment of realistic orbital or black-hole motion.

Additional browser coverage checks:

- Library search, category/favorite browsing, persistence, visit counts, old homepage redirects, and single-frame lifecycle.
- Unavailable local storage, empty search results, and a phone-sized library.
- Workspace pause/resume and frozen experiment time.
- Live Fourier waveform and pitch updates, audio suspension, and released audio nodes on stop.
- Gas temperature/remix synchronization and conservation of energy and momentum in an isolated elastic collision.
- Stable GPU geometry counts through repeated Kerr and three-body launches.
- The Kerr speed control, symmetry at zero spin, and photon capture/escape on opposite sides of the analytic critical impact parameter.
- Analytic Gaussian derivatives agreeing with numerical finite differences to less than 10⁻⁶ at sampled points.
- Every world opening directly from `file://` with browser networking disabled, plus the library opening a 3D lab from disk.
- A real emulated touchscreen gesture drawing a waveform, and a touch press changing the ant world's mass.
- Quantum Cosmos view switching and previous-frame removal.

The scheduler tests verify the 60 Hz update ceiling on a simulated 120 Hz display, hidden/paused suspension without a clock jump, and cancellation.

## Visual review

The higher-dimensional explorer adds `previews/hypercube-8d-desktop.png`, `hypercube-slice-desktop.png`, `hypercube-8d-phone.png`, and `hypercube-landscape.png`, and refreshes the two tesseract previews. These were reviewed for legible controls, visible geometry, and desktop/phone layout. A responsive-canvas overflow found during testing was fixed, and the landscape header was moved clear of the scrolling sidebar. Seven tests in `tests/hypercube.spec.js` cover all six dimensions and their vertex/edge counts, all rotation-plane choices, independent plane angles, native slider keyboard behavior, actual solid sections, empty and centered slices, stable vertex identity, reset semantics, reused GPU buffers, updates while the workspace is paused, offline Retina rendering, and control access at 390 × 844, 375 × 667, and 844 × 390. The final 18-check regression run includes those seven, six existing camera checks, Manifold and Knot control/layout smoke checks, library navigation, repeated Kerr launches, and all 25 worlds loading offline. The original eight-test camera suite also passed before the responsive-canvas refinement.

Eight tests in `tests/hypercube-model.test.js` verify 3D–8D topology and edge degree, orthogonal rotations and inverses, finite bounded perspective, aligned slices, a 45-degree tesseract section's analytic width, empty and lower-dimensional boundary contacts, the regular-octahedron diagonal section, generic rotated sections satisfying the original cube inequalities and Euler relation, and invalid arguments. They check geometry within floating-point tolerances, not every possible orientation or near-degenerate slice. The current source check covers all 28 active HTML pages and all application scripts.

The vehicle update adds `previews/wind-race-car.png`, `wind-semi-truck.png`, `wind-jet.png`, `wind-objects-desktop.png`, and `wind-import-phone.png`. Developed flow around each vehicle was reviewed. The preview lighting was adjusted to retain surface shading, and the desktop sidebar now scrolls beside the tunnel. Nine browser tests in `tests/wind-objects.spec.js` verify distinct solid masks and finite, deflected flow for all four vehicles; local OBJ and ASCII/binary STL import; explicit application; orientation and scale; a frame with a hole; invalid and edge-on imports preserving the current obstacle; mouse rotation; one live GPU geometry after repeated replacement; original shapes, painting, and resets; cancellation of stale selections; offline phone loading; and paused workspace integration. The four existing Wind Tunnel fidelity checks and three collection checks also passed. Six unit tests cover parsing, concave-face area, malformed files, limits, projection axes, fit bounds, and empty silhouettes. Other browsers and very large production meshes have not been tested.

`previews/time-hole-desktop.png`, `time-speed-desktop.png`, `time-earth-desktop.png`, and `time-earth-phone.png` show the new experiments. Desktop and phone screenshots were reviewed for legible clocks and controls, clear scene labels, and layout fit. The new Time schematic joins the existing illustrations in the refreshed 25-world gallery. All 24 previous SVG illustrations, including both uses of Wind Tunnel, are preserved.

The thumbnail update adds `previews/world-thumbnails.html`, a linked gallery of all 24 worlds, and `previews/world-thumbnails.png`. Each illustration was reviewed at card size; the desktop and phone library screenshots were refreshed. A one-off comparison against the previous illustration function verified 23 changed cards, 24 distinct SVG outputs, no invalid numeric values, and byte-for-byte preservation of both the standard and featured Wind Tunnel artwork. The library navigation, phone/unavailable-storage, all-world offline, and screenshot checks passed after integration.

Screenshots in `previews/` show the desktop library, embedded Wind Tunnel, phone library, phone Bell lab, and phone black-hole view. They were inspected for readable controls, viewport fit, rendering, and layout. The 3D phone panels were then given a more opaque background, a visible border, and independent scrolling.

The later Wind Tunnel update adds `previews/wind-tunnel-retina.png`, captured after 2600 solver steps at device pixel ratio 2. A before/after visual review confirmed smooth obstacle edges and sharper labels and particle marks. Its dedicated tests check 390-, 1440-, and 1920-pixel viewports; drawing and erasing alignment; contours with holes; fullscreen pause/resume; color and flow-line controls; and preservation of solver state during rendering and resizing. The existing Wind Tunnel mobile/control test, library lifecycle, all-world offline checks, and screenshot capture were also rerun successfully.

`previews/wind-tunnel-rendering.json` records a 2104 × 1052 backing canvas for the 1052 × 526 CSS view and a median renderer-call duration of about 0.8 ms across 25 calls on this machine. This measures rendering submission time, not total experiment time or end-to-end GPU frame latency.

The tesseract update adds `previews/tesseract-controls-desktop.png` and `previews/tesseract-controls-phone.png`, both visually reviewed. Eight tests in `tests/manifold.spec.js` cover wheel/keyboard/button zoom, two-axis mouse orbit, camera stability across animation frames, paused 4D geometry with an interactive camera, surface/tesseract transitions, camera reset and limits, offline startup, emulated phone drag/pinch, and camera manipulation in the paused library workspace. The existing Manifold Playground mobile/control test, library lifecycle test, and all-world offline test also passed.

## Scope

The browser checks cover startup and selected interactions, not every possible parameter combination or a full scientific validation of all numerical models. Other browsers and physical mobile devices were not tested. Touch checks use Chrome's device emulation. Demanding experiments can still run below 60 updates per second on slower hardware.

The historical files in `archive/` are unchanged source snapshots and are deliberately excluded from active-page validation. In particular, the MHTML snapshot is not an executable experiment.

Run `npm run check` and `npm test` to repeat the checks. Browser reports are written to `test-results/` and excluded from the distributable ZIP.


## Equations and model notes — September 16, 2026

- Audited all 25 catalog worlds. Time Observatory retains its existing notes; 23 standalone worlds now use `assets/world-notes.js`. Quantum Cosmos exposes the notes belonging to each of its three embedded views.
- Each new disclosure includes equations, symbol/unit definitions, the implemented method, and assumptions/limitations. Notes describe the current experiments, including the Bell gauge/label sign mismatch, GPS contribution toggles, and fixed Roche reference. Experiment calculations were not changed.
- Shared keyboard handling now preserves native summary operation. Manifold Playground moves its single disclosure between the surface footer and the cube explorer controls.
- `npm run check` passed: all active HTML/scripts, local links, unique IDs, and 22 model/runtime tests.
- Browser verification through the in-app browser over a local HTTP server passed for all 23 new standalone disclosures (mouse expansion, Space collapse), phone-width checks (390 × 844, no horizontal overflow), all three Cosmos views, Manifold view switching, and the existing Time Observatory disclosure. Turing/Fourier Enter expansion was confirmed after rendering. Representative desktop and phone views were visually inspected.
- Added `tests/model-notes.spec.js` for automated disclosure coverage. The Playwright CLI suite could not execute in this session because the sandboxed Chrome subprocess aborted before opening pages; the browser checks above were run through the available browser tool instead.


## Observatory-style redesign — September 16, 2026

- Completed all 24 other catalog worlds: consistent headers, grid-backed scene cards, separate live readouts, condition controls, and explanation/model-note sections. Quantum Cosmos uses the same presentation for all three embedded worlds, with frame height following the active content.
- Time Observatory's page and dedicated assets, the catalog, and the collection homepage are unchanged. All original controls with IDs are retained. Manifold's single model-note disclosure now stays in the explanation section across surface and cube views.
- Canvas backing sizes adapt to display density while logical experiment and pointer coordinates stay fixed. Paused images survive resizes without applying their drawing opacity twice. WebGL cameras and renderers fit the scene card. Double Slit's wave image uses full logical resolution, and Gravity uses a clean coordinate grid.
- `npm run check` passed: local references, IDs, and syntax across 28 active HTML pages, plus all 26 automated model/runtime/rendering tests. Four new rendering tests cover logical coordinates through repeated resizing, paused-image state, detached accumulation canvases, and WebGL card resizing.
- In-app browser checks at desktop (1440 × 1000) and phone width (390 × 844) covered startup and layout across the redesigned worlds. Every standalone model-note disclosure opened by mouse and closed with Space; phone checks found no horizontal overflow or clipped controls. Representative scene/control screenshots were visually reviewed.
- Focused interactions checked Qubit measurement, Double Slit wavelength/mode/particle detection, Fourier presets and harmonics, Wind Tunnel vehicle selection/front projection/application/pause and its phone editor, Manifold 8D slicing and returning to surfaces, all three Cosmos views and their notes, and collection-level embedded pause/resume. These are selected interaction checks, not exhaustive scientific validation.
- The Playwright CLI browser suite was not rerun because its Chrome subprocess was blocked by the sandbox earlier in this session. Browser verification used the available in-app browser. Existing historical test results above describe their original versions; they are not a claim that the entire old browser suite was rerun after this redesign.
- Originals of changed files are backed up in the Codex workspace under `backups/before-observatory-redesign`; `changes.json` records new files and hashes.


## Manifold consolidation and 4D–8D knot explorer — September 16, 2026

- Collection reduced from 25 to 23 worlds: Quantum Cosmos retired; the separate knot moved into Manifold Playground. Its three experiment selectors are Curved surfaces, Cubes · 3D–8D, and Knots · 3D–8D. Old standalone files are compatibility redirects. The removed Cosmos controller is backed up before removal. Catalog search, descriptions, counts, gallery links, and saved knot IDs are updated.
- The knot starts paused at a trefoil, with play, scrub, re-tie, midway, and fourth-direction peek controls. An optional explorer adds dimensions 4–8, all 6/10/15/21/28 coordinate rotation planes, composed plane angles, automatic rotation, sequential perspective, slice coordinates, zero/positive slice thickness, a labeled whole-loop guide, and camera reset/zoom. Each experiment retains its settings when switching tabs; inactive knot and surface/cube rendering is suspended.
- The revised analytic knot centerline has an explicit injectivity and regularity argument for its interior deformation in the notes. All display coordinates, the additional harmonic coordinates, projection bound, and sampled-polyline slice limitations are documented. This is not a rope dynamics model or a sequence of different knot types.
- `npm run check`: 28 HTML entry points (including redirects) and application scripts passed reference/ID/syntax validation; all 33 numerical, rendering, mesh, and runtime tests passed. Seven added knot tests check endpoints/closure, the equal-W candidate separation used by the injectivity argument, rotations and bounded projection, analytic single-plane intersections, simultaneous higher-dimensional bands, empty slices, and invalid input handling.
- In-app browser verification over local HTTP checked demo endpoints, keyboard scrubbing, 4D–8D dimension/plane counts, angle changes, populated/empty and zero-thickness slices, mouse/Space disclosure operation, switching to 8D cubes and back, preserved knot settings, embedded workspace pause and manual controls while paused, retired-file redirects, and library knot search. Desktop visuals and 390 × 844 controls/expanded notes were reviewed; no horizontal overflow or clipped controls were found.
- Added four repeatable browser regressions in `tests/knot-explorer.spec.js` and updated old Cosmos/model-note coverage. These Playwright CLI browser tests were not executed in this session; Chrome subprocess launch was blocked by the sandbox in the earlier work. The browser checks reported above used the available in-app browser. Historical results earlier in this file refer to the versions tested then.

## Electromagnetic Fields — September 16, 2026

- Added world 24 in Particles & Fields, with a custom gold/cyan field schematic, catalog search/selection, thumbnail gallery entry, and updated collection counts. Existing experiment pages are unchanged.
- Three top-level experiments contain electrostatics, wire/loop/coil magnetostatics and uniform-B particle motion, plus separate induction and Maxwell-wave views. Each view has units, equations, methods, limitations, and primary-source links in its model-note disclosure.
- `npm run check` passes all 29 HTML entry points and application script/reference/ID checks, plus 42 numerical/runtime/rendering tests. Nine added model tests cover E = −∇V, superposition and Coulomb limits, wire and loop analytic fields, coil summation, magnetic handedness and speed/radius preservation over 10,000 steps, electric trajectory energy, Faraday/Lenz signs and scaling, Maxwell stability/finite propagation, absorption, screen/gap transmission, and reset.
- In-app browser verification at 1440 × 1000 and 390 × 844 checked charge dragging and numeric positions, six-source limit and empty-field behavior, potential toggle, test-particle launch, probe positions, current reversal, all magnetic geometries, uniform-particle pause and zero-B readout, automatic/stationary induction, and its model notes. Wave checks covered the conducting gap, continuous source, visible wavefronts, pause, clear, and pulse injection while paused. Keyboard Space operates the notes. Phone controls and expanded notes fit without horizontal overflow.
- Collection integration verified the 24-world count, four Particles & Fields entries, the custom thumbnail, search, embedded loading, and workspace pause/resume. Solver time remained exactly frozen while paused even when a display control was changed, then resumed. No lab-error banner appeared in these checks.
- Added three repeatable Playwright browser regressions in `tests/electromagnetism.spec.js`. These CLI browser tests were not executed: Chrome subprocess launch was blocked by the sandbox earlier in this session. Reported browser checks used the available in-app browser; they are selected interaction checks, not exhaustive validation of every parameter combination.
- Original changed files and a hash receipt are backed up in the Codex workspace under `backups/before-electromagnetic-world`. New files are listed separately in the receipt. The release checksum manifest is refreshed when applying the change.

## Homepage appearance update — September 16, 2026

`npm run check`: HTML/assets/JavaScript validation passed; all 46 Node tests passed, including new appearance preference tests covering System mode changes, custom color persistence, invalid or unavailable storage, density bounds, and contrast handling for black and white accents.

Manually verified in the Codex browser at desktop and 390 × 844 phone sizes: live Wind Tunnel preview, dark and light presentation, System selection, saved settings after reload, gold/violet/mint accents, a live accent change on the Wind Tunnel slider, search, navigation into and back out of a lab, and home-animation pause/resume. No browser errors were reported. The phone layout had no horizontal page overflow.

The existing Playwright suite was not rerun: sandboxed headless Chrome could not launch. Browser UI verification used the in-app browser instead. The existing library test now scopes its “no lab iframe” assertion to the lab workspace because the home page intentionally contains a preview iframe.


## Fields & Spacetime 3.0 — September 17, 2026

The home page now indexes 75 experiments within the existing 37 collections. No duplicate worlds were added. Cards use the actual solver canvases; only visible previews are mounted. The featured investigation is driven resonance.

Validation for this edition:
- Source/reference validation and all 56 numerical/runtime Node tests passed.
- All 75 preview URLs loaded with a visible selected canvas, correct added-collection mode, and no reported JavaScript/runtime error. Verified a canvas changed over time, the motion switch paused every mounted preview, and entering a lab removed catalog frames.
- All 41 equation-editor browser checks passed (40 on the initial run; Manifold passed after fixing investigation-panel insertion into its nested layout).
- Six redesign regressions passed: the Bell statistic reaches the ideal 2√2 and all 16 deterministic local cards stay at 2; changing preparation clears counts; Rutherford uses equal force coefficients and v=√(2E); impossible wormhole launches are rejected and valid speed is preserved; collider idle has no beam; settings restore and CSV download work; live preview lifecycle works.
- Five homepage/sound checks passed, including the featured actual solver, reduced motion, subject navigation, opt-in audio, and phone layout.
- Locale checks cover 18 flags/languages, reversible selection, preserved paused state and equations, decimal conventions, Arabic digits, RTL/mobile fit, embedded propagation, and offline links.
- Desktop (1440×1000) and phone (390×844) homepage smoke checks found no horizontal overflow or page errors.

Targeted corrections from the prior audit: aligned CHSH signs and preparation handling; shared Rutherford charge scale and kinetic-energy control; wormhole feasibility and critical case; collider idle labeling; qualified Roche outcome, GPS range-equivalent wording, capped fission expectation, correct flyby direction, and reference-mass lensing angle. The three-body time limit no longer claims stability. The original audit is retained in MATH_AUDIT.md.

These are selected implementation and mathematical-limit checks, not a complete convergence study or research-grade validation. Machine-assisted translations retain English as reference. Full historical browser suites were not rerun; results above identify the coverage actually executed.


## Live-preview motion corrections — September 17, 2026

Reviewed all 29 previews named by the user. Projectile & drag remains unchanged.

- Electric fields now release a test charge; the magnetic preview launches a Lorentz-force trajectory; induction moves its magnet; Maxwell waves use the continuous source.
- Earth clocks use the existing minute-per-second playback option to avoid a 30 Hz alias of the day-per-second setting. Kerr launches paired photons, and knots play and replay their existing untying demonstration.
- Fourier silently varies its real harmonic-count control. Static analytical/geometric views demonstrate labeled input changes: flyby impact, refraction angle, lens distance, prism apex, nodal mode, uncertainty width, orbital state, relativistic speed, event time, lensing source offset, stellar stage, scale factor, CMB display gain, timeline age, particle-family highlight, and Mandelbrot boundary zoom. These are control demonstrations, not new physical evolution laws.
- RLC and heat-transfer trials repeat; diffusion uses an existing higher diffusion-coefficient setting and repeats. Polarization's existing evolving fields were verified.
- All automation uses the shared runtime scheduler and stops on homepage pause, reduced motion, or hidden/offscreen preview teardown. Normal standalone labs retain their manual controls and default states.

Verification: 29 homepage browser checks compared actual canvas screenshots for change, pause stability, and resume; three further checks covered knot replay, RLC reintegration, and unchanged standalone behavior. All 32 passed. The existing 56 source/runtime/model tests passed. Additional offline Chrome checks confirmed visible captions on desktop and mobile, no phone-width overflow, and reduced-motion freezing of preset changes. Earlier load-only preview tests were insufficient to detect stationary models; the new regression checks test visible motion.


## Fullscreen compatibility fix — September 17, 2026

Fullscreen now expands the complete workspace, including its exit and pause controls. It attempts the standard or WebKit native API, and retains a viewport-sized in-page fallback if that API is missing or rejects the request. Escape from within the embedded lab exits the expanded view, and navigating away clears the layout/scroll lock. The iframe is retained when toggling fullscreen. Mobile layouts use dynamic viewport height and safe-area padding.

Three Chrome browser regressions passed at 390 × 844: native fullscreen, a simulated absent API, and a simulated denied request. Each verified viewport bounds, retained lab settings/iframe, exit button, Escape from the iframe, and navigation cleanup. All three shared runtime tests passed. Real iPhone Safari was not directly tested; the fallback does not promise to hide browser chrome.


## Simulation-first carousel — September 17, 2026

Moved the live showcase before the introduction. Slides are Qubit Box, Sound Frequency, Double Slit Lab, Ring Collider, then Double pendulum. Rotation uses the existing 6.5-second delay and respects pause/reduced motion, visibility, and keyboard focus. Touch devices no longer inherit a persistent mouse-hover pause. Only the active slide mounts a solver iframe. Mobile cards show the visual before a short title and launch link; descriptive paragraphs remain on desktop.

Two mobile browser regressions passed: ordering and top placement, real automatic advancement, all five live canvases and source URLs, the Double pendulum launch destination, reduced motion, manual controls, RTL scrolling, and no horizontal overflow. The 390 × 844 layout was visually inspected.


## Mobile carousel touch and playback — September 17, 2026

The paused carousel now offers an enabled Play button, including when reduced motion requested the initial pause. Explicit playback overrides that preference for this page session; a subsequent pause or device preference change cancels the override. Initial reduced-motion behavior remains paused. Frame readiness is resynchronized on load and the page resumes its activity state after pageshow/focus.

Horizontal pointer gestures over the visual drag the carousel and snap to the next/previous world. Vertical touch scrolling and pinch zoom remain native; dragging does not activate an experiment link. Rotation waits until a drag ends.

Three mobile Chrome regressions passed: CDP touchStart/touchMove/touchEnd gestures for left/right swipes and vertical scrolling; manual live playback under reduced motion; screenshot changes in all five carousel canvases. These used emulated touch input, not a physical iPhone Safari session.

## September 18 — focused teaching redesign

The four teaching labs were checked for desktop/phone layout, page errors, equation-input compatibility, silent Fourier previews, touch drawing, audio updates/pause, saved advanced settings, and view restoration. Existing gas collision conservation checks pass. Known-sine reconstruction, logistic fixed-point/sensitivity, and deterministic seed checks pass.

An initial mature-pattern check exposed a disappearing splitting preset; the final presets were adjusted and rendered after 12,000 updates each. All remained finite and nonnegative for the checked seed. The Gray–Scott explicit timestep is now 0.25 after the original timestep produced nonfinite concentrations in preset testing. This is targeted numerical validation, not a full convergence study or validation over every parameter and initial condition.

Screenshots are in `previews/*-teaching-{desktop,phone}.png` and `previews/pattern-{spots,stripes,splitting}.png`. New English teaching copy falls back to English in locales without matching translations.

## September 18 — reference-inspired hypercube diagrams

- The default cube display is now a flat orthographic projection with a selectable 2D–8D comparison gallery. Existing perspective and solid 3D-section modes remain available.
- Nine focused browser checks passed, covering the dimension sequence, section geometry, accumulated plane rotations, tracked vertices, buffer reuse, paused-workspace updates, offline/phone rendering, flat dragging/construction colors, and perspective orbit.
- Nine hypercube model tests passed, including 2D topology and an independent check that the flat projection basis is orthonormal, preserves parallel edges, and stays inside the n-cube radius bound.
- The source/model suite passed 60 tests. Desktop comparison, construction, and phone previews are saved in `previews/hypercube-comparison*.png` and `previews/tesseract-flat-construction.png`.
- Gallery cards use individually fitted uniform display scales; their sizes are not measurements on a shared axis. New text uses the existing English fallback where untranslated.

## September 18 — Curvature & Geodesics

Five numerical tests check analytic gradients/Hessians, orthonormal frames and known Gaussian curvatures, plane lines/sphere great circles/cylinder helices, constraint and speed preservation, timestep refinement, and tangent-patch error against the sphere formula. Five browser tests check the six surfaces, lesson controls, picking/orbit, bounded comparison trails, geometry-tab isolation, Equation lab integration, and desktop/phone layouts. The nine hypercube browser regressions also pass. Desktop and phone screenshots are saved as `previews/surface-*.png`.

Numerical checks cover selected initial conditions and lengths, not every possible trajectory. Patch maxima are sampled estimates; color saturation is fixed at Gaussian curvature ±4 model units⁻².

## September 18 — quantum accuracy upgrades

Nine pure-model tests cover closed-system probability conservation, free-packet group velocity, absorbed-probability accounting, packet-weighted rectangular-barrier transmission, threshold/resonance limits, grid refinement, Fresnel symmetry/coherence and far-field behavior, Zeno ensemble statistics, and CHSH predictions/bounds. The complete source/model suite passes 74 tests.

Focused browser checks exercise all four desktop and phone layouts, independent tunneling controls, detection sampling, Bell marginal statistics and preparation resets, Zeno fixed-duration trials, global preview pause, and automatic replay. Screenshots are saved as `previews/*-accurate-{desktop,phone}.png`. An older homepage test still expected a separate Earth catalog card; it now checks the grouped Time Observatory card and independently retains Earth-route coverage.

The numerical checks cover selected parameters, not every drawn potential or every control combination. Tunneling retains finite-grid and absorber errors. The double-slit model is paraxial and conditions detections on a finite screen. Bell intervals are conservative fixed-sample bounds, not repeated-peeking significance tests. Zeno measurements remain ideal projective measurements. New explanatory copy uses the existing English fallback where untranslated.

## September 18 — eight dynamics/relativity/detector labs

The source/model suite passes 87 tests. Thirteen new pure-model checks cover adaptive integration; GPS rate derivation and clock-bias separation; Rutherford angles, energy, and distributed-charge matching; full three-body figure-eight closure and invariants; restricted Jacobi conservation; Schwarzschild symmetry/capture and high-spin prograde capture bracketing; wormhole speed/angular-momentum conservation and turning radius; twin proper times; decay four-momentum/mass reconstruction; rubble internal momentum, initial binding, finiteness, and timestep refinement.

Twenty-five targeted browser regressions passed across the new interfaces and retained Kerr/wormhole/Three-Body functionality, including repeated-launch GPU resource disposal. A subsequent 23-check browser run passed desktop/phone layouts, advanced branches, Equation lab compatibility, and actual homepage motion/pause for the rebuilt labs. Screenshots are in `previews/*-precision-{desktop,phone}.png`, `previews/roche-live-aggregate.png`, and `previews/collider-missing-momentum.png`.

Kerr capture cutoff was reduced to r₊ + 0.001 so it lies inside the prograde photon orbit even at a = 0.99. Existing Roche/GPS/Rutherford/collider explanatory notes were replaced to match the new models.

These are targeted checks, not validation over all initial conditions. The 123-sphere rubble model uses uncalibrated damped contacts/bonds and nominal density; its fluid reference is not a failure prediction. The GPS solver is 2D and fixed-epoch. Detector events are selected decays with simplified response, not physical event-production rates. Twin pulses are resolved to the 0.01-year worldline samples.

## Bell / Schwarzschild / retarded-field rebuild — September 18

- `npm run check`: active HTML/scripts and links pass; 100 model/runtime checks pass, including 13 new tests for these labs.
- Targeted interface regression run: 66 browser tests passed across the new labs, quantum upgrades, equation controls, and earlier redesign checks.
- Final new-lab and shared regressions: 12 browser tests passed, including offline opening of every world, phone touch controls, exports, previews, and pause behavior.
- Existing orbital/scattering/detector lab regression suite: all 15 tests passed.
- Desktop and phone screenshots were generated for all three rebuilt labs. Bell desktop, gravity phone, and Light Factory desktop were visually inspected; spacing and plot labels were adjusted during review.

Detailed physical assumptions and sampled numerical benchmarks are in `ACCURACY_REPORT.md`. These checks do not constitute independent scientific certification or exhaustive validation over every possible setting.

## Approved Qubit Box promoted to the library — September 18

The concept is now the main experiment, with separate production model/controller/styles. The original HTML backup matches its pre-replacement SHA-256. `npm run check` passed all 104 model/runtime tests and source/link checks. The 44-test browser run passed equation-entry regressions and production controls, measurements, ensemble sampling, downloads, mobile/offline loading, preview looping, and shared workspace pause. Production screenshots are in `previews/qubit-box-main-desktop.png` and `previews/qubit-box-main-phone.png`.