About this tiny world

Slice & Seek: sections and evidence

This puzzle is inspired by destructive FIB-SEM serial imaging: material is milled away and the newly exposed face is imaged. Saved images preserve observations after the material is removed. The game uses original stylised geometric inclusions, a fixed depth grid and a cut budget. There is no physical scale, milling-rate model or predictive electron contrast. Grey regions distinguish materials. The full teal answer is known model geometry, including unsampled regions; it is not a reconstruction from the player’s sections.

Shape profiles supply the candidate models, recorded sections and final reveal. Confirmation requires a unique match among the three candidates using the acquired sections. Hints use only the candidates consistent with those observations. Links select one of six authored version-1 mysteries; there is no daily challenge, leaderboard or stored result. Inspecting client-side code can reveal answers.

• ZEISS · Volume EM with FIB-SEM — serial milling, imaging the exposed face, image-stack processing and segmented volumes.
• Xu et al. (2017), Enhanced FIB-SEM systems for large-volume 3D imaging, eLife — the serial imaging method and acquisition trade-offs. Historical performance limits in this paper are not current instrument specifications.

References checked 23 September 2026. Links provide background, not endorsement. Play Slice & Seek.

Bump or Pit? Contrast and shape

These are original, dimensionless synthetic surfaces. Each feature has a compact smooth signed height profile: a positive height makes a bump; a negative height makes a pit. The image uses a bounded directional term from the local slope, plus a small slope-magnitude term. The same height field supplies the image, model cross-section and answers. Detector changes preserve all geometry. No noise, charging, material contrast or electron transport is modelled.

This is an SEM-inspired contrast analogy, not a calibrated secondary-electron image. Real detectors collect electrons rather than illuminating the specimen, and their collection fields mean the process is not equivalent to a lamp. Contrast depends on more than height. The free cross-section clue reveals the known synthetic geometry; it is not recovered from image brightness or stereo data. No tilt-pair reconstruction is implemented.

There are three generated surfaces per version-1 seed, with four independently signed features on each. There is no fixed number of bumps or pits. All choices remain editable until submission; one submission reveals every answer and ends the round. Clues carry no penalty. Counts reflect matching model labels, not demonstrated SEM operating or interpretation skill. Links repeat the same surfaces. Answers are not stored. Client-side code can reveal the geometry, so this game is unsuitable for secure assessment.

• JEOL · Secondary electron image — surface inclination, topography and edge effects.
• JEOL · Scanning Electron Microscope A to Z — detector position, topographic contrast and curved secondary-electron trajectories.

References checked 23 September 2026. Links provide background, not endorsement. Play Bump or Pit?.

Invisible Swimmers: contrast reveals detail

The seven shape-based names describe original invented objects, not biological species. Each object has a fixed synthetic absorption and phase field. Motion follows a bounded artificial path. These are not experimental images, true species identification, calibrated sizes or a biological motility model.

Brightfield uses weak absorption. Darkfield emphasizes changes in the phase field and some absorption-related structure; it is an illustration of redirected light against a dark background, not fluorescence. Phase contrast uses an illustrative interior and surrounding-halo transform, not a wave-optical phase-ring calculation. Oblique lighting uses a directional phase gradient in the field's coordinate frame; its apparent relief is not measured height. These simple image transforms do not predict real microscope performance. No aperture or resolution model is included.

Geometry, internal features, position and the movement clock are shared across views. All seven objects give a signal in every view. The close-up enlarges the same model display and does not represent additional optical resolution. The discovery thumbnails use a standard darkfield-style view for recognition. The game shows no physical scale.

A version-1 seed recreates the same seven objects and starting arrangement. Inspecting and confirming a distinct object adds it to a temporary field guide; repeated inspection does not increase the count. Guided discovery pauses and marks an unfound object, with a described route for inspection. There is no timer, health model, drying threat, assessment or saved score. Client-side code and guided clues can reveal locations.

Microscope Detective uses version-2 seeds for three cases with seven lookalike candidates each. The clues concern outline, the count and arrangement of internal spots, and a clear central window. The same fields supply every microscope view and the optional descriptions. Candidate letters indicate positions, not species or answers. A case is solved only when both the candidate and evidence match. Incorrect pairs receive the same feedback. Repeated attempts are allowed and counted; this is not secure assessment. Hints and descriptions are recorded as assistance. Showing a worked example marks a case reviewed rather than solved. Switching activities preserves the current page session; reloading resets it. Old version-1 links still open the discovery warm-up.

• Nikon MicroscopyU · Specimen contrast — weakly absorbing objects and the difference between amplitude and phase effects.
• Evident · Darkfield illumination — blocking direct illumination and imaging redirected light.
• Nikon MicroscopyU · Introduction to phase contrast — converting phase differences to intensity, halos and shade-off.
• Nikon MicroscopyU · Oblique illumination — directional contrast and apparent relief.

References checked 24 September 2026. Background reading does not imply endorsement. Play Invisible Swimmers.

Down the Column: illumination, image focus and contrast

This is an original, dimensionless bright-field mass-thickness teaching analogy. The specimen contains invented scattering inclusions. It is not an experimental image, a crystal-diffraction calculation, an operating procedure or an assessment of microscope competence.

The illumination envelope responds to a simplified paraxial condenser crossover and two beam shifts. A finite minimum width represents omitted source and wave effects. The side-view envelope above the specimen and representative image-forming rays below it are separate diagrams: animated marks are not counted electrons. The lower diagram uses thin-lens propagation to illustrate focus and angular selection in the objective back focal plane. Several real lenses are omitted. The image model assumes near-parallel illumination; the upper convergence diagram does not feed a complete angular-distribution calculation.

The same invented scattering field supplies each image. Its brightness combines weak attenuation with narrow direct and wider scattered angular distributions. A centred objective aperture admits a fraction of each distribution; it does not crop specimen coordinates. A Gaussian image-blur approximation combines baseline blur, focus error and an inverse-aperture diffraction term. The model deliberately makes an excessively small opening lose fine detail. This is not a calibrated TEM contrast-transfer function, and its control values are relative settings rather than physical units.

Each capture uses the same exposure factor and square-root display mapping. There is no counting noise, accumulated dose, damage or time-based reward. Mission one measures corner illumination, balance and excessive central concentration in the captured image. The specimen-image missions check a visible intensity recovery between the known synthetic pair, sufficient contrast and the player's observation of the arrangement. Several settings can pass. Old captures are invalidated when controls change. Starting-view comparisons preserve current settings and cannot be submitted as current captures. Hints and resets have no score penalty; checks and hint use are reported as learning activity, not skill certification. Reloading starts over; client-side code can reveal answers.

The condenser-stigmator mission is a separate beam-only alignment view over an empty area. Two correction components change a symmetric astigmatism matrix; propagation around crossover produces a Gaussian beam whose ellipse changes orientation. The displayed X/Y values are correction components, not horizontal/vertical beam shifts. The two prescribed condenser settings have opposite offsets of equal magnitude. Their images share a fixed flux, exposure and display scale. A paired capture is accepted only when both displayed beam images are sufficiently round, compact, visible and not clipped. A single least-confusion circle cannot complete the mission. Starting and corrected pairs use matching display sizes. This ideal model omits aperture clipping, higher-order aberrations and instrument-specific alignment coupling; it does not model objective astigmatism or certify alignment skill.

The later specimen-image missions are separate prepared exercises. Their illumination is set for the image model; the stigmator Gaussian is not propagated through the complete TEM column.

• University of Oldenburg · JEOL 2100F TEM guide, page 5 — condenser-stigmator circularity checks on either side of crossover; instrument-specific operating instructions are not transferred into the game.
• JEOL · Condenser-objective lens — illumination versus image formation and TEM versus STEM beam conditions.
• JEOL · Objective lens — the first image-forming lens after the specimen.
• JEOL · Objective aperture — selection at the objective back focal plane.
• Indiana University Electron Microscopy Center · Aperture holders — contrast and resolution trade-offs.

Scientific background checked 24 September 2026. Sources provide background, not endorsement or calibration of this model. Play Down the Column.

Grid Hunter: looking for promising places

Cryo-EM screening examines a grid at different scales: an overview of grid squares, holes in the support film, and closer particle images within those holes. Researchers consider ice, particle distribution and other evidence when choosing where to investigate further. Brightness alone is not an ice-thickness measurement or a guarantee of useful particles.

The game uses original, dimensionless synthetic fields. Its overview is schematic. Each close-up has fixed particle positions, outlines, background and patches; revisiting a location preserves that field. Brightness and contrast differences are deliberately exaggerated. No electron scattering, defocus transfer, measured ice thickness, particle motion or beam damage is modelled. Invented rings and other shapes do not identify real proteins or species. No physical scale is given.

The optional scouting mission asks for three fields from different squares with several separated, visible particles and a quiet background. Its review uses the known generated field category. This is a gentle discovery objective, not evidence that a real grid is suitable: other particle distributions, including sparse fields, can be useful in real experiments. Every square contains at least one field that meets the game's goal. Empty bright fields, overlapping shapes, clumps, weak contrast and obscuring patches provide contrasting examples. Descriptions are optional text alternatives, and the review records how many selected images were described. Repeated comparisons are allowed without penalties.

A version-1 seed repeats the same grid. Links share the grid, not collected images or progress. The notebook lasts only while the page stays open. There is no score, certification or secure assessment; client-side code can reveal the categories. Three chosen images do not validate a grid or produce a molecular structure.

• CryoEM 101 · Chapter 3, Grid Screening and Evaluation — evaluating grids, ice and particle distribution.
• CryoEM 101 · Chapter 4, data collection — locating and selecting areas at different magnifications.
• Rice et al. (2018), Routine determination of ice thickness for cryo-EM grids — measurement methods and why an image's brightness is not a calibrated thickness value here.
• Cheng et al. (2023), Fully automated multi-grid cryoEM screening using Smart Leginon — grid screening and target selection in an automated acquisition workflow.

Scientific background checked 24 September 2026. Sources support the teaching context, not calibration or endorsement of the game. Play Grid Hunter.

Element Detective: energies, sampling and overlapping evidence

EDS uses characteristic X-ray energies to support elemental identification. These are original teaching specimens and simulated counts, not experimental micrographs or spectra. NIST experimental transition energies supply the selected line positions. Gameplay rounds them to two decimal places in keV: Si Kα₁ 1.74; Fe Kα₁ 6.40 and Kβ 7.06; Cu Kα₁ 8.05 and Kβ 8.91; Ti Kα₁ 4.51 and Kβ 4.93; V Kα₁ 4.95 and Kβ 5.43. The single Kα teaching peak is positioned at Kα₁; fine structure and many other lines are omitted.

The virtual beam is fixed at 20 kV, above the K-shell excitation thresholds of the selected elements. Gaussian teaching peaks use a constant 0.13 keV full width at half maximum. That is an illustrative approximation inspired by typical EDS resolution at Mn Kα, not a prediction of a detector's energy-dependent response. Background shape, line strengths and count variations are deliberately simple. They are not calibrated detector efficiency, a complete physical counting model, or concentration percentages.

In the boundary case, one invented patch contains two adjacent domains. A one-dimensional Gaussian spatial average blends the iron and copper contributions at the selected points. The displayed halo is schematic, and no physical specimen length or depth is assigned. This illustrates why X-rays can originate around and below the beam position; it is not an electron-transport model. The selected spectra can support a change across those sampled locations, not a reconstructed three-dimensional arrangement. Two elemental signals together do not prove an alloy, compound or core–shell structure.

The Ti Kβ and V Kα reference positions are about 0.020 keV apart. They are not separately resolved by this teaching response. Case 3 first shows a cropped energy window of the captured spectrum; the wider view reveals an additional energy range from the same data. It does not improve energy resolution or collect additional counts. A visible companion feature near 5.43 keV supports vanadium within the game's restricted candidate set. Without that feature the conclusion is “no separate clue resolved,” not that vanadium is certainly absent. Real interpretation must consider additional lines, interferences, background, sensitivity and instrument conditions.

Answers are checked against captured spectrum features. Cases vary the assignment of elements, the direction of the boundary and whether a vanadium companion is present. Repeated scans of one spot recall the same teaching spectrum. Captures, solved cases and optional descriptions are preserved when switching cases within the page. Reloading or opening new cases starts over. Descriptions and repeated checks have no penalty; their counts are reported honestly. Client-side code reveals the generated model, and repeated guesses are possible. This is an exploratory learning game, not a secure assessment or a real-material identification service.

• NIST · X-ray Transition Energies Database — experimental energies, converted from eV to keV. KL₃ corresponds to the selected Kα₁ line; the selected Kβ values use KM transitions.
• Oxford Instruments · WDS applications — the Ti Kβ / V Kα overlap at approximately 4.932 / 4.952 keV.
• JEOL · Energy dispersive X-ray spectroscopy — characteristic X-rays and illustrative energy-resolution context.
• Gatan / EDAX · Composition metrology in compound semiconductors — why lower beam voltage confines the X-ray interaction volume and can introduce EDS peak overlaps.
• Gatan / EDAX · Improving EDS quantitative results — why measured intensities require quantitative corrections.
• JEOL · Elemental mapping — the difference between X-ray intensity maps and corrected concentration maps.

Scientific background checked 24 September 2026. Sources explain the principles; they do not calibrate or endorse this game. Play Element Detective.

Relative family strengths use approximate values from LBNL X-ray Data Booklet, Table 1-3: Kβ₁,₃ divided by summed Kα₁ + Kα₂ is 15/150 for Ti and V, 17/150 for Fe, and 17/151 for Cu. Each selected family is collapsed into a single equal-width Gaussian, so its area ratio equals its peak-height ratio. These source-specific approximations are not universal measured EDS ratios or concentration estimates.

The shared Ti/V peak’s relative strength can suggest vanadium. This game then asks for an independent companion line. Real spectrum fitting can use intensities, shapes and residuals to analyse overlapping lines: NIST, overcoming peak overlaps in titanium and vanadium analysis.

Fringe Climber: finding height through contrast

Coherence scanning interferometry records interference signals while scanning through a range of positions. The position of a signal’s contrast envelope supports locating the surface. Real processing can also use phase information. A repeating bright fringe alone does not establish an unrestricted height.

This original schematic uses invented flat terraces, a Gaussian contrast envelope and a repeating phase term. A six-cycle spatial carrier makes bands visible across each patch; it is a teaching display, not a complete tilted-reference interferometer. Patch boundaries stay sharp. No focus blur, calibrated length, specimen colour or experimental measurements are implied. The scan coordinate increases in the same sense as the invented height; it does not prescribe an instrument’s motor direction.

All lengths use relative game units. The envelope width is 7 and the phase wavelength parameter is 4. The phase varies as 4π times the scan-height difference divided by that wavelength, illustrating round-trip path change. The grayscale display uses 0.48 plus 0.37 times the envelope and cosine term, with no per-frame brightness normalisation. Spatial-band contrast is calculated from these rendered pixels. This simplified spatial display is not the temporal envelope-extraction process used by a real instrument.

Players save views explicitly. A weaker saved view never replaces a stronger saved view of the same terrace. Checks require sufficiently clear recorded bands and the correct height order; there is a range of acceptable scan settings, not one exact slider value. Contrast thresholds of 0.42, 0.46 and 0.60 are game-design choices, not metrology standards. The last climb’s gate keeps the accepted scan ranges separate for its closer terraces. The final profile reveals known invented heights; it is not a reconstruction from the saved images.

The model omits high-NA imaging, slopes, reflectivity variations, noise, dispersion, multiple reflections and instrument errors. Automatic sweeping is optional. Text descriptions and repeated checks are allowed; text-help use is recorded honestly. Progress lasts while the page is open. This is an exploratory puzzle, not a secure assessment, measurement service or instrument procedure.

• Research on interference fringe visibility and surface positioning — signal-envelope localisation and visibility-based processing.
• University of California, Santa Barbara · Michelson interferometer — round-trip optical path, phase and fringe visibility.
• Fourier optics modelling of coherence scanning interferometers — more complete modelling of the optical imaging process.

Scientific background reviewed 24 September 2026. References support the underlying principles, not calibration or endorsement of this game. Play Fringe Climber.

Before It Fades: signal, grain and specimen change

Ideal independent Poisson counts have a variance equal to their expected count. Increasing counts can therefore reduce relative counting noise. Real images also depend on detector behavior, motion, contrast transfer and other effects. Electron exposure can change a sensitive specimen; fine structural information may weaken before large features look obviously different. Damage can occur within one exposure.

This game uses original, dimensionless invented particles with two, three or four small bright features. A broad Gaussian particle stays unchanged, while the expected intensity of its small features decays exponentially with cumulative exposure. Each exposure step adds independent Poisson counts sampled from the current model specimen. Recorded counts are accumulated and never decayed or removed. The equally weighted cumulative count image is divided by the total exposure and a fixed count scale, then clipped and rounded to eight-bit grayscale. Earlier saved images remain frozen.

The changing specimen has a fine-signal term proportional to exp(−D/k). In the ideal continuous limit, its exposure-normalized accumulated contribution is proportional to k(1−exp(−D/k))/D. It is not the instantaneous fine signal at the end of the exposure. Later low-detail frames can dilute fine contrast in this simple equally weighted image. Real frame selection and dose weighting can preserve useful early information; this game does not find a universal optimal exposure.

The expected count scale is 5 per game step at unit intensity. The three invented specimens have decay constants in the ranges 22–24, 12–14 and 16–18 relative units. There is no physical dose, wavelength, specimen scale or material calibration. The 128-step limit is an interface limit. Bright dots do not identify atoms, molecules or a real material. The model omits contrast transfer, motion, detector imperfections, dose-rate effects and detailed damage mechanisms.

Evidence checks use exactly the displayed eight-bit pixels. They compare the weakest known model feature’s central-versus-surrounding contrast with background standard deviation. Accepted views have background deviation below 0.17, local contrast at least 0.085 and contrast/noise at least 1.65. These are game choices, not metrology or resolution standards. The background region has a small underlying intensity gradient. No exact exposure value is required. The third challenge requires the selected kept view and another view of that same copy to differ by at least eight steps.

Pausing also blanks the simulated beam and stops exposure. It does not undo damage. Stopping recording on a real instrument does not necessarily blank the beam. A fresh copy means a new identical fictional specimen with fresh counting noise; it retains the notebook, not the old specimen. Actual low-dose practice can include focusing or tracking away from the recorded area; that workflow is not represented here.

Text descriptions provide an assisted way to identify sufficiently visible model features, and assistance is recorded at each check. Repeated attempts are allowed. Known model answers are revealed after success. The puzzle is not a secure assessment, instrument procedure or recommendation for a physical dose. Progress lasts while the page is open.

• Postek and Vladár (NIST, 2016) · Dealing with Noise — independent Poisson noise and count statistics.
• Grant and Grigorieff (2015), Measuring the optimal exposure for single particle cryo-EM using a 2.6 Å reconstruction of rotavirus VP6 — exposure-dependent information loss and dose weighting; its quantitative results are not used as this game’s calibration.
• SerialEM · Low Dose mode — separating view, focus, tracking and record areas.
• SerialEM · Program properties and shutter configuration — instrument-dependent beam and shutter control.

Scientific background reviewed 24 September 2026. Sources support the principles, not calibration or endorsement of the game. Play Before It Fades.

Ghost in the Image: test the measurement

AFM images depend on both the specimen and the measurement. A probe with multiple contact points can produce repeated apparent features. Limited feedback tracking can create direction-dependent lag. An unusual repeated arrangement can also be genuine surface geometry. Reliable interpretation uses complementary checks; a single changed or unchanged image does not establish the cause.

This game isolates those three possibilities on stable, invented dimensionless height fields. Geometric acquisition uses i(x,y) = max[s(x+u,y+v) − t(u,v)], with a finite parabolic primary apex. The two-apex case takes the maximum over both contact geometries; it does not add duplicate height signals. A specimen rotation is applied to the surface before the probe and feedback model. Scan axes remain fixed. Artificial companion offsets therefore remain in instrument coordinates while real internal spacing rotates with the specimen.

The replacement probe has the same primary apex shape but only one contact point. This controlled comparison leaves the non-probe cases exactly unchanged. Real replacement probes can differ in radius and shape and may change widths or fine detail even on genuine geometry. Ordinary finite-tip broadening is present in every model case.

Feedback tracking follows a first-order recurrence in scan traversal order: measured height approaches geometric height by a fixed fraction at each pixel. The lag case uses 0.19 for the original scan and 0.67 for the slower scan. Reversal changes traversal order without mirroring the displayed field. Other cases use ideal tracking. This illustrates directional lag and smoothing, not oscillatory feedback ringing. A slower scan is not a universal cure for all real streaks; debris, tip changes and other mechanisms can produce artifacts.

Images use a fixed monotonic height-colour mapping from eight-bit synthetic heights. Brighter indicates higher model height; there is no physical calibration. Each test changes one condition from the original setup. The game omits force dynamics, adhesion, noise, drift, damage, wear and mixed causes. The final surface view is the known model, not a reconstruction.

Observation checks use acquired image differences, connected high regions, height-weighted spatial moments and the difference from a rotated original image. Probe-shape and genuine-surface explanations require probe replacement and specimen rotation. Tracking-lag explanations require reversed and slower scans. These are authored evidence rules within the restricted model, not general diagnostic proof. Every recorded observation is checked, including optional experiments. A new seed shuffles case order and varies feature geometry. Repeated comparisons preserve the same field. Descriptions provide assisted image observations and are noted honestly; client-side answers and repeated guesses remain possible. This is not a secure test or an instrument procedure.

• Villarrubia, NIST · Algorithms for scanned probe microscope image simulation, surface reconstruction, and tip estimation — geometric imaging and morphology; equation 2, printed page 429.
• ThermoMicroscopes · SPM Guide, hosted by KTH — probe/feedback artifacts and rotation, scan-direction and scan-speed checks; printed pages 72–75.
• Bruker · Recognizing Probe Artifacts, hosted by KTH — multiple contact points and repeated features.
• Nanosurf · Roughness analysis and tracking artifacts — scan speed, feedback and lost small-scale information.

Scientific background reviewed 24 September 2026. Sources support the principles, not calibration or endorsement of the game. Play Ghost in the Image.

Pattern Forge: arrangements and diffraction

This game evaluates coherent scattering from identical point scatterers in a plane. Its dimensionless intensity is I(q) = |Σ exp(−2πi q·r)|² / N² for N > 0. Here q is in cycles per model unit, not the angular wavevector convention used in some references. An empty arrangement is explicitly displayed as zero. A single point has uniform normalised intensity. The drawn size of a point is not included as a form factor.

The diffraction view is sampled on 201 × 201 positions spanning −2.25 to +2.25 cycles per model unit on both axes. A fixed log display mapping log(1 + 24I) / log(25) makes weaker features visible. The central forward intensity is 1 for every nonempty arrangement; this normalisation does not compare absolute scattered power between different point counts. Visible colours encode intensity, not wavelength.

A finite rectangular point array has maxima at reciprocal spacings. Reducing its horizontal point spacing expands the horizontal diffraction spacing, and with fixed point count also broadens its peaks. Rotating the structure in the plane rotates the ideal pattern in the same direction. Translation of the entire arrangement leaves its intensity unchanged. For these identical real point weights, opposite scattering coordinates have equal intensity, so a half-turn also leaves the pattern unchanged. Intensity does not uniquely determine the structure.

The first challenge fixes 25 points in a 5 × 5 array, with vertical spacing 1.2 model units. Horizontal spacing varies from 0.65 to 1.50. Checks find the nearest strong positive horizontal side peak in the calculated intensity and accept a position from 1.25 to 1.80 reciprocal model units. The two nearest strong side peaks are circled; other bright orders may already lie inside the amber boxes. The second challenge starts afresh with 15 points in a 5 × 3 array, spacings 0.8 and 1.2. Only its in-plane rotation changes. Four target disks of radius 0.115 lie around the first axial reciprocal points of the 30° reference. Each must contain an intensity above 0.72. These authored tolerances accept a range of orientations, including the equivalent half-turn, rather than one hidden angle.

The studio allows up to 36 model points and does not grade a unique diffraction solution. Keeping a creation requires an edit and at least three points; that is a creative prompt, not a physical rule. Edits use half-unit positions within a central −3 to +3 grid, with optional rotation. Undo, saved chapter outcomes and one frozen kept creation last for the page visit. Exported pictures state the fixed view windows and ideal-model qualification. Hints and text descriptions are optional learning aids, not secure assessment safeguards.

This is an ideal 2D Fourier interference model. It omits chemical bonding, element-dependent scattering factors, multiple scattering, specimen thickness, Ewald-sphere selection, noise and instrument transfer effects. It is not a calibrated TEM diffraction image. Amber target guides and the structure cursor are interface annotations, not measured signal.

• Li & Zhang (2021), Unifying the concepts of scattering and structure factor in ordered and disordered samples — coherent amplitudes, point scatterers and intensity; sections 2.2–2.3.
• IUCr · The reciprocal lattice — reciprocal geometry and diffraction.
• Diamond Light Source · Reciprocal space viewer — inverse relationships between real and reciprocal spacing.
• Proffen & Neder · Interactive tutorial about diffraction — in-plane rotation and the pattern.

Scientific background reviewed 25 September 2026. Sources support the principles, not calibration or endorsement of this game. Play Pattern Forge.

Tiny Repair Shop: The Wobbly Robot

This fictional repair puzzle distinguishes reflectance, measured topography and missing data. Each case has three 10 µm-wide fields: a level marking, a level patch with a weak return, and a raised ridge. Site order, ridge height (0.55–0.75 µm), width and marking position vary. The robot and repair are a story device; their scale, movement and the effect of the ridge are exaggerated. The optical measurement does not remove material. A separate fictional workshop tool performs the repair.

The numerical model samples 49 × 49 surface points. A bounded Gaussian coherence envelope A exp[−(z − h)²/(2σ²)], with σ = 0.10 µm, is sampled every 0.025/N µm across a 1, 2 or 4 µm height window centred on the level reference. It uses an envelope surrogate, not a raw interferogram or a full CSI reconstruction. The selected height is the sampled envelope maximum. A return is rejected when its peak-to-model-noise ratio is below 4, or the peak lies within 0.20 µm of the window edge. Missing heights stay null and are hatched, never converted to zero or interpolated.

Signal sampling N = 1, 4 or 9 increases the number of axial light samples and reduces the assumed independent signal-noise scale as 0.05/√N. This represents acquiring more signal information before height reconstruction, not averaging completed height maps to fill missing pixels. A small deterministic sinusoidal perturbation, bounded by 0.01/√N µm, makes profile uncertainty visible; it is illustrative, not a random shot-noise model. Accepted heights use a fixed −0.20 to 1.00 µm colour scale. The displayed profile is the centre row. Classification uses its heights relative to the mean of its first and last seven points. A complete profile is “raised” above +0.15 µm, “depressed” below −0.15 µm, otherwise “level”. These authored tolerances fit these specific synthetic specimens and are not uncertainty estimates, calibrated limits or universal metrology rules.

All three proposed actions are reviewed together against the acquired profiles, not the selected scan settings. Saving an action neither grades it nor changes the surface. An inconsistent plan receives observational feedback without identifying which individual action is wrong. Incomplete profiles cannot justify a repair. A case completes only after every patch has supporting evidence and the repaired patch has a fresh level measurement. Changing settings does not alter a captured scan. Before-repair evidence is retained with its acquisition settings; the repaired profile is compared on the same scale. The displayed level bound is rounded upwards, not to a false zero, and is not an uncertainty estimate. The scan indicator represents axial acquisition across the whole field; the reconstructed map appears on completion. New cases clear the record. All decisions and model data are client-side; this is a learning game, not a secure skills assessment.

Real coherence scanning interferometers localise interference signals during a vertical scan. Reflectance and signal quality influence whether height can be measured. Extra raw signal samples may improve signal-to-noise, but cannot recover every invalid point. The game omits steep-slope collection limits, transparent films, multiple reflections, correlated noise, drift, material changes and repair mechanics.

• Noise Reduction in Coherence Scanning Interferometry for Surface Topography Measurement — signal oversampling versus averaging completed height maps, weak signals and missing points.

Scientific background checked 26 September 2026. These sources support the principles, not validation or endorsement of the game. No independent specialist or learner validation is claimed. Play Tiny Repair Shop.

Light Leash: optical trapping

The game uses an overdamped two-dimensional Langevin model with a finite-depth Gaussian potential, viscous drag and thermal noise. It is an effective teaching model, not a calibrated optical calculation. A higher-index bead is attracted to the bright focus. The bubble comparison reverses the gradient-force direction but uses schematic bead-like drag; it omits buoyancy, heating and axial forces. Other optical configurations can trap bubbles. The collection pocket is fictional retaining apparatus. The deliberately soft trap makes lag and thermal motion easy to see. Many laboratory traps hold comparable beads much more tightly; stiffness and carrying speed depend on the beam, particle and liquid. Close-up magnifies the same positions without adding motion.

• Experimental optical-trap stiffness measurements.
• Volpe & Volpe — Simulation of a Brownian particle in an optical trap.
• Bui and colleagues — Escape forces and trajectories in optical tweezers.
• Microbubble trapping in inverted optical tweezers — bubbles can be trapped with another optical configuration.

These references support the principles, not endorsement or independent validation of this game. Play Light Leash.

Noise Detective: a room leaves clues

One fictional environmental-vibration case. An unchanged synthetic pattern is sampled at individual raster-pixel times with relative beam–specimen displacement. Changing scan speed changes its appearance, not the disturbance amplitude. An engineered support change reduces transmitted motion while both devices keep running.

The 512 nm field contains 256 × 256 pixels. Line periods are 1, 2 and 4 ms, with 10% flyback. Source contributions are chosen values: 18 Hz at 14 nm, 27 Hz at 1.6 nm, and a 3 Hz residual at 0.8 nm. Engineered supports transmit 12% and 15% of the respective source terms. This simplified lateral model omits full structural dynamics, resonance, drift, electromagnetic interference, dose and shot noise. Real isolation is frequency-dependent and can amplify vibration near resonance.

Before, Test and Return preserve their acquired pixels and settings. Edge waviness is measured from the displayed left edge; the motion trace is model displacement over one second. The displayed image, description and trace refer to the same completed record; a separate progress bar shows the next acquisition. Saving keeps exactly that shown record. Optional predictions are recorded before a control change and preserved with the Test, without grading. Saving waits for the entire raster and trace. A supported finding needs matched settings, one changed support, and more than 50% reduction in both RMS readings. This is an activity rule, not an instrument acceptance standard. A later matched return test can support a causal interpretation in this model. Reset changes the phases, not the case. No timer, score, account or competence claim.

• NIST · SEM vibration artifacts
• Thermo Fisher · Environmental interference
• JEOL · Isolation and resonance

These sources support principles, not validation or endorsement of the game. Play Noise Detective.