# Show2D Storyboard Use with [Storyboard](storyboard). ## Stories ### S2D-01: Open A Large Real Image Quickly **User story**: As a microscopist opening an image, I want a useful preview in about a second for normal working sizes, and still within seconds for heavy stress data, so I can decide whether the file is worth inspecting. **Primary widgets**: Show2D. **Data to use**: one real 4k or larger image; repeat with an 8+ panel real or real-derived gallery for heavy signoff. **Acceptance checks**: - Load the image from a real file path on the backend, from an in-memory array, and from exported HTML when supported. - Measure first visible paint and note display bin/native bytes. - Verify the title/info badge communicates preview/detail state when binning is active. - Verify the widget remains usable while the backend/kernel is idle after first paint. - Verify the notebook does not save the full 4k array unless the user explicitly chooses an export path that embeds data. ### S2D-02: Arrange Panels For Comparison **User story**: As a user comparing several real-space images, I want to choose the number of columns so I can fit panels to my monitor, notebook, or paper figure layout. **Primary widgets**: Show2D. **Data to use**: at least 8 real or real-derived panels. **Acceptance checks**: - Change columns through 1, 2, 3, 4, 6, 8, and 12. - Verify the menu does not offer impractical counts above 12. - Verify labels, scale bars, stats, histograms, and borders remain aligned. - Verify the current zoom center and contrast do not jump during reflow. - Enable Reorder, drag panels into a non-source order, and verify `panel_order`, keyboard navigation, hidden-panel export, and `to_show3d()` follow the displayed order while source-index labels/stars remain correct. ### S2D-03: Hide Unimportant Panels **User story**: As a user screening many images, I want to hide unimportant panels while preserving the scientific state of the remaining panels. **Primary widgets**: Show2D. **Data to use**: a multi-panel gallery with labels and visible scale bars. **Acceptance checks**: - Hide one panel, multiple panels, and all-but-one panel. - Verify layout, labels, stats, histograms, keyboard selection, export, and saved state ignore hidden panels. - Use labels and title spans containing symbols and inline math such as `\lambda=0.03 raw` and `$\\chi^2$/pixel`; verify the panel menu, stats row, and panel titles render symbols/math and never show raw markup. - Restore panels and verify original order and panel state return. ### S2D-04: Inspect Native Pixels From A Fast Preview **User story**: As a user inspecting atomic or lattice detail, I want a fast binned preview to stream native-resolution tiles when I zoom in. **Primary widgets**: Show2D. **Data to use**: a 4k or larger real image with recognizable high-frequency structure. **Acceptance checks**: - Zoom past preview resolution and verify a detail request is issued. - Verify the returned tile uses native row/column coordinates and a tile ``bin`` smaller than ``_display_bin_factor`` when possible. - Pan across detail boundaries and verify stale detail tiles are never drawn after the view changes. - Verify cursor readout reports native ``(row, col)`` and labels value source as preview, detail, or native. ### S2D-05: Adjust Contrast On Noisy Data **User story**: As a user adjusting contrast on noisy microscopy data, I want histogram interactions to be smooth and visually correct. **Primary widgets**: Show2D. **Data to use**: noisy real data where stale tiles or contrast flashes are easy to see. **Acceptance checks**: - Drag histogram min/max handles quickly. - Drag histogram center/range repeatedly while hover readout is visible. - Toggle auto contrast off/on and verify manual range is preserved until the user asks for auto again. - Verify no stale square tile, ghost rectangle, white flash, or delayed color update remains. - Record FPS for histogram drag and slider movement. ### S2D-06: Link And Unlink Comparison State **User story**: As a user comparing related panels, I want linked zoom, pan, contrast, denoise, frequency-filter, and padding edits to be optional and reversible. **Primary widgets**: Show2D. **Data to use**: multi-panel real data with shared features. **Acceptance checks**: - Enable linked zoom, pan, and contrast; verify panels move together. - Disable each link mode; verify independent panel state works. - Pass ``cmap=["gray", "inferno", "RdBu", ...]`` for a multi-panel gallery and verify each panel keeps its requested colormap in the live UI, saved state, static notebook preview, and standalone HTML export. By default, verify the Color dropdown is linked across panels. Then open More, turn off ``Link Color``, select each panel, and change Color from the UI; verify only that panel changes and no extra Link Color text appears in the main toolbar. Turn ``Link Color`` back on, then off again; verify the previous individual panel colormaps return instead of being replaced by the shared colormap. - In the View menu, set Padding to 10%, choose Median fill, and verify the canvas and histogram update while the stored raw arrays remain unchanged. - In a drift-correction gallery, toggle Padding from All to selected-panel scope, set one panel to 20% Mean fill, and verify only that panel's ``pad_ratios`` / ``pad_fill_modes`` entry changes while every panel shares the larger comparison canvas. - Save and restore the state; verify ``pad_ratio``, ``pad_ratios``, ``pad_fill_mode``, ``pad_fill_modes``, and ``pad_scope`` round-trip and ``reset_view_ops()`` clears the border bit-identically. - [x] **S2D-PANEL-DN-1**: With denoise edits unlinked, select panels configured as raw, Gaussian σ2, and Gaussian σ8. Change only the σ2 panel to σ20, then revisit all three panels. The editor and canvas must retain `None/4`, `Gaussian/20`, and `Gaussian/8`; the two neighboring canvases must be pixel-identical before and after the edit. - Resize the grid and verify view anchors do not jump. ### S2D-07: Use FFT To Inspect Periodicity **User story**: As a user looking for periodicity, I want FFT views for every visible panel and I want them to remain fast during layout changes. **Primary widgets**: Show2D. **Data to use**: real data with lattice peaks; include one suspicious panel for reference comparison. **Acceptance checks**: - Toggle FFT and verify every visible panel gets the expected FFT view. - Change columns and resize panels; verify FFT alignment and spacing remain correct. - Zoom and pan in FFT mode; verify events target FFT, not stale real-space layers. - Verify every interactive FFT panel always shows its own accurate multiplier, including the default `2.0×`; independent and linked wheel/pinch zoom must update the correct labels, and reset must restore `2.0×` without changing real-space zoom. - While the viewer is paused, switch to another browser tab and return. Every real-space and FFT canvas must repaint automatically with the same zoom, pan, multiplier, and selected panel. FFT cache-hit state may change, but FFT miss and compute counters must not increase merely because the tab returned. - Compare one suspicious FFT against NumPy or a known reference. ### S2D-08: Measure Features With Overlays **User story**: As a user measuring image features, I want profile and ROI overlays to remain stable while I draw and drag them. **Primary widgets**: Show2D. **Data to use**: real image or gallery where line profiles and ROIs are useful. **Acceptance checks**: - Toggle Profile; draw, move, and delete a line profile. - Toggle ROI tools; draw, drag, resize, save, restore, and delete ROIs. - After drawing circle, rectangle, square, and annular ROIs, call ``get_roi_geometries()`` and verify the returned ``(row, col)`` center, radius, corners, bounds, visibility, and colors match the visible overlays and survive ``state_dict()`` / ``load_state_dict()``. - Verify high-frequency pointer labels do not pop or lag. - Use keyboard navigation for previous/next panel, reset zoom, and delete ROI. ### S2D-09: Export And Share A Static Result **User story**: As a user preparing a shareable result, I want export choices to say exactly what they will save and produce files that reopen correctly. **Primary widgets**: Show2D. **Data to use**: single image and multi-panel gallery. **Acceptance checks**: - Open Export in live Jupyter and standalone HTML. - Verify the single-HTML choices distinguish `encoding="full"` from `encoding="uint8"` and show approximate sizes when known. - Collapse controls for presentation-style viewing and verify Export remains reachable from the title chrome in live and standalone HTML. - Export single HTML with both encodings where supported. - [x] **EX-2**: Export a denoised view and a frequency-filtered view, then open each standalone page without touching a control. The first canvas must already show the filtered pixels; the histogram and canvas must not disagree. - Open both files and drive columns, hide panels, FFT, histogram, zoom, and reset. - Use Copy and verify output corresponds to the current visible state. ### S2D-10: Save And Reopen A Notebook **User story**: As a notebook user, I want ``Cmd+S`` and reopen to preserve a visible compact output without embedding huge pixel buffers. **Primary widgets**: Show2D. **Data to use**: a live Jupyter notebook with real or real-derived data. **Acceptance checks**: - Press ``Cmd+S`` in JupyterLab and reload/reopen the notebook. - Verify the saved static output is visible and compact. - Draw one circular ROI and then multiple ROIs on a single image, then save and reopen the notebook without rerunning the cell. The saved PNG fallback must show the full image with all ROI overlays plus one right-side ROI zoom/crop panel per visible ROI. Multiple ROI zoom panels should use comparable crop width/height where image boundaries allow it, so a report reader can compare marked features without mistaking crop scale for scientific size. - [x] **S2D-ROI-SAVE-1**: Execute and save a notebook containing one circular ROI and one rectangular ROI. Reopen the saved output without rerunning the cell and verify the PNG fallback shows the full field plus separate ROI crop panels, with crop-panel scale bars recalculated for the smaller field of view. - Check ``metadata.widgets`` or ``get_state()`` for heavy-buffer leaks: ``frame_bytes``, ``_detail_bytes``, offline stacks, and export payloads must not be present when ``save_state=False``. ### S2D-10B: Save Microscope-Stage Inspection States **User story**: As a scientist driving Show2D interactively in a notebook, I want to save named view states like microscope stage positions, so I can return to an ROI, FFT, contrast, padding, or panel-selection view without writing programmatic export code. **Primary widgets**: Show2D first; later extend the same pattern to Show3D. **Data to use**: a single real or real-derived image with one defect ROI, plus a 3-panel comparison gallery with raw/corrected/residual-style labels. **Acceptance checks**: - From the More menu, save a state named `defect A ROI` after drawing an ROI, zooming, enabling FFT, and adding padding. Verify `saved_view_states` gains exactly one lightweight entry with a readable summary. - Save a second state with different hidden panels, selected panel, contrast, and padding. Load the first and second states repeatedly; verify ROI, selected panel, hidden panels, FFT toggle, padding, contrast, and view box restore without changing the source arrays. - Use Update on an existing state after moving the ROI or changing padding; verify the state keeps one entry and later loads the updated view. - Delete one state and verify the other remains loadable. Then Delete All and verify the list is empty and the widget stays usable. - Run the same actions from Python with `save_view_state()`, `load_view_state()`, `delete_view_state()`, and `clear_view_states()`. - Save and reopen a notebook or round-trip `state_dict()` / `load_state_dict()`; verify the named states survive but heavy buffers (`frame_bytes`, `_detail_bytes`, `panel_stack_bytes`, standalone HTML payloads) are not duplicated inside each saved state. ### S2D-11: Use The Widget On A Phone Or Narrow View **User story**: As a user checking results on a phone or narrow screen, I want controls to wrap and remain usable without covering the scientific image. **Primary widgets**: Show2D. **Data to use**: single image and multi-panel gallery. **Acceptance checks**: - Test a narrow mobile viewport. - Verify controls wrap, labels do not overlap, panels remain usable, and any horizontal scrolling is intentional. - Test touch-style zoom, pan, menu open, column selection, and panel visibility. - For iPhone-specific claims, serve the page to a physical iPhone Safari test. ### S2D-12: Review A High-Throughput Denoising Batch **User story**: As a user reviewing denoising or drift-correction results, I want to open dozens of 4k images as a gallery, arrange them quickly, hide weak outputs, and keep interaction fast enough to screen the batch without exporting manual contact sheets. **Primary widgets**: Show2D. **Data to use**: real or real-derived 4k x 4k files from a denoising, drift, or ptychography workflow. Test at least 30 panels for routine signoff; use 45 and 85 panels when backend storage and memory allow. **Acceptance checks**: - Load the gallery from file paths on the backend without copying files to the laptop. - Record file count, native shape, dtype, total native bytes, first-paint time, display bin, and browser memory if available. - Change columns through 2, 4, 6, 8, and 12; verify panel labels, scale bars, stats, histograms, and hover readouts remain aligned. - Hide poor panels, restore them, and verify selection order and export state stay correct. - Pan, zoom, histogram-drag, and resize repeatedly; record the interaction FPS method and result. - Verify zooming into one panel streams or displays the highest-resolution available tile for that panel, while the rest of the gallery remains responsive. ### S2D-13: Keep Loading And Storage Lightweight **User story**: As a notebook user working with large files, I want loading to show a useful view quickly and saving to keep the notebook small, so I can come back later without embedding gigabytes of image data. **Primary widgets**: Show2D. **Data to use**: one 4k or larger image and one 30+ panel 4k gallery from real backend files. **Acceptance checks**: - Compare live Jupyter loading, saved-notebook reopen, and standalone HTML export paths. - Verify live Jupyter uses backend file/array access for detail streaming rather than serializing every native pixel into widget state. - Press ``Cmd+S``, reload the notebook, and verify the saved output is visible, compact, and labeled as preview/detail/offline as appropriate. - Open Export and verify the single-HTML `full` and `uint8` encoding choices follow the Show3D wording and show approximate file sizes when known. - Confirm saved notebook state and exported HTML payload sizes are recorded in the signoff report. ### S2D-14: Stress Interactive Controls On Many 4k Panels **User story**: As a scientist screening high-throughput image results, I want all high-frequency controls to remain smooth even when many large panels are on screen, because slow hover, histogram, or zoom feedback makes the viewer unusable for triage. **Primary widgets**: Show2D. **Data to use**: 30, 45, and 85 real or real-derived 4k x 4k panels when available; otherwise record the largest real batch tested and why the larger case was skipped. **Acceptance checks**: - Measure first paint, column reflow, histogram drag, mousewheel zoom, pan, hover readout, FFT toggle, and reset on the heavy gallery. - Verify target interaction remains near 30 FPS for the controls under test, or record the limiting hardware/browser/data condition. - Confirm stale preview/detail tiles are not drawn after rapid zoom, pan, resize, or contrast changes. - Verify controls remain keyboard and pointer reachable when the gallery is taller than the viewport. - Add failures or near misses to the performance log with the data path and exact shape so the case can be replayed. ### S2D-15: Inspect Images Full Screen On A Large Monitor **User story**: As a microscopist using a workstation backend and a laptop or desktop browser as the frontend, I want Show2D to use the available screen cleanly so I can inspect the scientific image without fighting notebook chrome, oversized controls, or wasted whitespace. **Primary widgets**: Show2D. **Data to use**: one real 4k or larger image, plus a 4+ panel real or real-derived gallery. **Acceptance checks**: - Launch from a remote Jupyter backend path when possible: the HPC/workstation backend owns the data and Python kernel; the browser drives the widget from the local laptop. - Open the notebook or exported HTML in a wide browser viewport and use browser full-screen mode. - Verify the scientific image or gallery grows with the viewport while controls remain compact, content-sized, and aligned to the same design language as Show3D and Show4DSTEM. - Verify top-right actions such as Export, Reset, Copy, and panel controls sit on the right edge of the widget header when there is available width. - Drive zoom, pan, histogram center drag, FFT, profile, ROI, and panel reflow in the large view; record whether any interaction loses visible FPS compared with the notebook-sized view. - Return to a normal notebook viewport and verify the layout contracts without clipped controls, wrapped labels, or stale full-screen sizing. ### S2D-16: Compare Static EDS Maps With Local HAADF Frames **User story**: As a microscopist reviewing an EDS acquisition, I want static elemental maps beside the acquisition's HAADF frame stack so I can inspect drift or damage without forcing every comparison panel onto one global frame. **Primary widgets**: Show2D. **Data to use**: one real Velox/EDS acquisition containing multiple HAADF frames plus at least three static elemental maps. Keep private microscope data outside git and record only shape, dtype, frame count, and timings. **Acceptance checks**: - Pass a list containing 2D maps and one or more `(frames, rows, cols)` items; verify only stack panels get an in-panel slider and play button. - Scrub and play each stack independently. Verify pixels, histogram, stats, profile, diff, and FFT follow the current frame without changing static maps. - Sustain a 30 Hz rAF-paced slider input stream on a normal-size EDS frame, or record the browser/data limit. Confirm the final canvas, not only the slider label, reaches the requested frame. - Hide, restore, and reorder a stack panel; verify its frame index is unchanged. - Select a stack panel and verify left/right arrows scrub it. Select a static panel and verify arrows retain ordinary gallery navigation. - Save a notebook with `save_state=True`, close it, and reopen without running the cell. Verify the slider, current frame, and pixels restore. - Export single HTML with `encoding="full"` and `encoding="uint8"`, open both without a kernel, and repeat scrub, play/pause, stats, FFT, hide/restore, and narrow-viewport checks with no browser errors. ### S2D-17: Scrub A Paged Comparison Gallery **User story**: As a user comparing a parameter sweep organized as pages (for example lambda values, each page holding the same panel set), I want the page slider and page playback to swap the whole panel grid at once while every per-panel control - histograms, labels, hide state - follows the page I am looking at, not the union of all pages. **Primary widgets**: Show2D. **Data to use**: a real or real-derived `(pages, panels, rows, cols)` stack with at least 3 pages x 4 panels, per-page distinct intensity ranges so histogram mismatches are visible at a glance. **Acceptance checks**: - Construct with 4D pages plus `page_labels`; verify the page slider shows `1/N`, the label text, and play/pause advances pages at the configured fps. - With `link_contrast=False`, verify exactly `panels_per_page` histograms are shown - never one per panel across ALL pages - and that their ranges change when the page changes. Repeat with the FFT histogram grid. - Hide a panel on one page; verify the histogram grid drops that panel's histogram, and other pages' hide state is independent. - Drag one page's histogram thumbs; verify only the current page's matching panel changes and the setting is still applied when scrubbing back to that page. - Verify stats, hover readout, and selection follow the current page's panels after several slider scrubs and one full playback loop. - Export offline HTML and repeat the histogram-count and page-scrub checks without a kernel. ### S2D-18: Watch A Live EMD Folder In Place **User story**: As a microscopist collecting survey images, I want one already displayed Show2D gallery to add each completed EMD as a full-resolution panel, without rerunning the cell, creating another widget, or losing the curation and measurement state of images already present. **Primary widgets**: ``Show2D.from_folder(...)`` in live JupyterLab. ShowFolder is a separate cached-preview and selection workflow; a ShowFolder refresh or a standalone HTML snapshot does not prove this direct full-resolution watcher. **Data to use**: Genuine compatible Velox image EMD files from one acquisition session, copied through atomic rename into a temporary watched folder. Use at least three files for routine verification and ten for release signoff. Include an incomplete file, an incompatible-shape file, and a later compatible file. Record source path, shape, dtype, sampling, native bytes, backend host, browser, widget commit, and watch interval. Keep the source data and generated report outside git. **Acceptance checks**: - In live JupyterLab, display exactly one ``Show2D.from_folder(folder, pattern="*.emd", watch=True, watch_interval=1.0, debug=True)`` object and capture its Python identity, widget model ID, and browser container before any later file arrives. - Start once with a valid EMD and separately start before the first acquisition. The empty-folder case must show a visible waiting state and accept the first stable EMD without rerunning the cell; until that works, mark this check ``Fail`` rather than substituting a prepopulated folder. - Keep one compact accessible watch badge near the folder/title area in stable DOM. Require green-dot ``Watching`` only while the actual background worker is alive. Enter ``Updating`` while a discovery poll is active and keep it through real candidate validation, append, and authoritative full-resolution paint. An idle poll may briefly show ``Updating`` but must return to ``Watching`` without decode, transfer, or repaint. Use amber ``Waiting for file completion`` for an incomplete EMD, red ``Watch error`` with corrective detail for a bad shape or worker failure, and gray ``Stopped`` or ``Not watching`` after stop/close or when liveness cannot be established. A ``watch=False`` snapshot has no badge. A restored notebook model, static fallback, or standalone snapshot must never restore a green ``Watching`` state without a live worker. Capture browser assertions and screenshots for every state; color alone is not the status signal. - Atomically complete a genuine EMD after the widget is visible. Verify one new labeled panel paints automatically without calling ``poll_folder()``. Compare its shape, calibration, representative pixels, and canvas checksum against ``read_image(path).array`` at source resolution. - Verify the Python object, widget model, and browser container remain the same. Preserve selected source path, stars, hidden panels, panel order, rotations, contrast, FFT, ROI/profile state, view box, and zoom when a naturally sorted filename is inserted before existing files. - Leave an incomplete EMD in place and then complete it. Verify it remains retryable, appends exactly once when stable, and never stops the watcher. - Add an incompatible-shape EMD followed by a compatible EMD. Report the mismatch without resizing scientific data or blocking the later valid append. - Measure stable-file-to-Python-append and stable-file-to-first-canvas-paint separately for every arrival; report median and p95 with the poll interval. A trait/count change without visible paint is not a pass. - Verify idle polls perform no reread, transfer, canvas repaint, or state reset. Exercise idempotent stop/resume, then call ``close()`` and prove no watcher thread can mutate the widget afterward. ### S2D-19: Compare Independent Slice Stacks Across Reconstructions **User story**: As a tomography or multislice reconstruction scientist, I want each Show2D gallery panel to hold a different reconstruction volume with its own slice slider and play/pause control, so I can browse each result at its own depth while keeping all reconstruction alternatives side by side in one viewer. **Why this matters scientifically**: One study may produce four reconstructions with different slice counts, slice thicknesses, regularization settings, alignment choices, or reconstruction methods. A feature or failure can be clearest at a different depth in each result. Independent local slice controls let the scientist stop each panel at the informative depth while linked zoom, pan, and contrast preserve the spatial comparison. This avoids opening several widgets or forcing every reconstruction onto one global slice index. **Primary widgets**: Show2D. Use Show3D instead when every panel should advance on one shared frame or slice axis. **Data to use**: four real or real-derived tomography or multislice reconstructions, each shaped `(slices, rows, cols)`. Include different slice counts or slice thicknesses when scientifically meaningful, plus an optional static 2-D reference. Record source, shape, dtype, slice count, slice thickness, display binning, and timings. Put the reconstruction method and slice thickness in each panel label because the compact slider readout reports only the current slice and total count. **Acceptance checks**: - Pass a Python list of at least four 3-D reconstruction arrays; verify every stack panel gets its own slider, play/pause button, and `current/total` readout. Verify an optional 2-D panel gets no slice controls. - Start panels at different `panel_frame_indices`, including a negative index, and verify each index resolves against that panel's own slice count. - Scrub one panel and verify only that panel's index and canvas change. Its histogram, stats, FFT, profile, ROI, and diff inputs must follow the selected slice while neighboring panels remain unchanged. - Play two stacks simultaneously, pause one, and verify the other continues; verify each stack wraps against its own slice count. - Construct with `panel_playback_fps=4`, measure several slice transitions, and verify the shared local-stack cadence is approximately 4 fps without a new toolbar control. Verify the configured value survives state restore and standalone HTML export. - With FFT visible, traverse every slice twice. After the initial FFT is visible, verify first-time slice computations retain the previous valid FFT instead of showing a full dark loading veil; on the second traversal, cache hits increase while misses and computes stay unchanged. - Pause every local stack on a different slice, switch browser tabs, and return. Verify all real-space and FFT canvases restore without pressing Play, every local slice index is unchanged, and no cached FFT magnitude is recomputed. - Verify linked zoom, pan, and contrast preserve spatial comparison without linking the local slice indices. - Select each stack and scrub with left/right arrows. From Python, use `set_panel_frame(panel, slice_index)` with either a source index or unique panel label. - Hide, restore, reorder, and responsively wrap panels; verify each local slice index remains keyed to its source reconstruction and controls do not overlap. - Save and reopen with `save_state=True`. Export single HTML with `encoding="full"` and `encoding="uint8"`, then repeat independent scrub, play/pause, FFT, and narrow-layout checks without a kernel. - On real reconstruction data, record first paint, memory and payload size, slider latency, and canvas repaint rate. Confirm the final canvas, not only the slider label, reaches every requested slice. ### S2D-20: Mark Panels, Preset Contrast, And Flip Orientation During Review **User story**: As a scientist comparing several related 2-D maps, I want to tag panels with simple colors, choose a shared percentile contrast preset, and temporarily flip one panel from the UI so I can tell collaborators and agents which map to inspect without rewriting the notebook. **Primary widgets**: Show2D. **Acceptance checks**: - Start from a minimal multi-panel `Show2D([...])` when possible, then use the UI to open More and change the Contrast preset. - Verify `marker_colors` / `identity_colors` paint durable panel strips and survive saved state plus standalone HTML export. - Verify `1-99`, `2-98`, and `3-97` contrast presets update all visible panels while the histogram stays visible below the image. - Flip one panel horizontally and vertically from More; verify only the display changes, stored data and ROI coordinates remain in the original `(row, col)` convention, and neighboring panels stay unchanged. - Use More → Rotate with Scope = All, then Scope = Panel. Verify 90°, 180°, and 270° rotations survive saved state and standalone HTML export, and that scale bars, FFT labels, ROI overlays, and right-side ROI crops remain legible. - Verify the More menu opens compactly: Rotate should appear as a simple switch first, and Angle/Scope controls should appear only after the user turns Rotate on or when an orientation is already active. - Verify rotation state never edits the scientific title or panel label. The live viewer may show only a compact direction glyph such as `↺90°` or `↻90°`, preferably over the image chrome, not as another text label above or below the panel. - Drive FFT, ROI, pan/zoom, denoise/filter, saved states, and export after the flip/contrast changes so the feature is tested as a real review session. ### S2D-20B: Embed Calibration Inset Plots In Image Panels **User story**: As a scientist tuning denoise or reconstruction parameters, I want each Show2D panel to carry a compact scientific inset plot, such as ACF versus `R`, so the image and the calibration evidence stay together in the same review figure without another notebook cell. **Primary widgets**: Show2D. **Data to use**: a 3 × 3 panel set with different calibration curves per panel. Use a real denoise/reconstruction calibration sweep for release signoff; a synthetic lattice is acceptable for fast API and hover regression checks. **Acceptance checks**: - Construct from `Show2D(..., inset_plots=[...])` with one dictionary per panel. Verify each panel gets a different curve, point marker, legend, annotation, and color. - Start from the human API first: `position`, `margin`, `size`, and `height`. Test `bottom-right`, `top-right`, `top-left`, `bottom-center`, and `center`. Use exact `box=[left, top, width, height]` only for the publication-layout control case. - Test visual style controls: `line_width`, `background_alpha`, `border_width=0`, nonzero `border_width`, `border_color`, `text_color`, and `tick_color`. Verify no-border and subtle-border cases remain readable on noisy images. - Turn on `show_ticks` with explicit `xticks` and `yticks`. Verify tick labels, axis labels, legend, and annotation remain inside the inset and do not cover the selected image feature. - Hover over the live inset plot. Verify the tooltip reports the nearest plotted coordinate using axis labels, for example `R 0.465 · ACF 0.48`, and does not clip at panel edges. - Move the scale bar with `scale_bar_position="bottom-left"` and hide `show_zoom_indicator` when the inset uses the lower-right corner. Verify the scale bar, inset, panel title, star, hide button, and resize handle do not collide. - Verify `state_dict()`, saved notebook static PNG fallback, `export_html()`, panel hide/reorder, responsive wrapping, and linked zoom/contrast preserve the inset plots. - On touch/mobile, verify the hover-only readout has an acceptable fallback before claiming iPhone signoff. If not implemented, record it as missing rather than relying on desktop hover evidence. ### S2D-21: Page A Growing Folder Gallery Automatically **User story**: As a microscopist whose acquisition or reconstruction folder may grow to hundreds of independent images, I want Show2D to show a bounded number of panels at once and create additional pages automatically, so the notebook remains readable while every full-resolution source image stays available. I want a useful default and an explicit override rather than having to reorganize files manually. **Primary widgets**: ``Show2D.from_folder(...)``. This is sequential item paging: each file is one independently curated panel. It is distinct from S2D-17 comparison pages, where the same logical slots repeat across parameter settings. ``Show3D.from_folder(...)`` deliberately does not use this behavior; its files extend one frame axis. **Data to use**: at least 45 naturally named, same-shape real or real-derived images so the default layout contains 20, 20, and 5 panels. Also start with 20 files and append the 21st through an atomic rename while the widget is mounted. Include a small deterministic NPY fixture for CI and genuine EMD images for browser signoff. **Acceptance checks**: - ``Show2D.from_folder(folder)`` defaults to ``page_size=20``. Exactly 20 ready images remain one ordinary gallery; the 21st creates two pages without replacing the widget. ``page_size=N`` changes the limit and ``None`` disables automatic folder paging. ``set_folder_page_size(...)`` regroups the mounted widget in place. Reject booleans, zero, negatives, and non-integers with a corrective message. - Show compact labels such as ``Images 1–20`` and ``Images 41–45`` plus a bounded, tabular ``current/total`` page readout. The final page contains only real source panels: never pad it with fake scientific images. - Keep folder pages as independent items. Hiding, starring, selecting, or rotating one file must remain keyed to that file and must not affect the same numerical slot on another page. Preserve a valid global panel order when paging activates after a live arrival. - Append across 20 -> 21 and later page boundaries while the user is browsing an older page. Preserve the Python object, widget model, browser container, selected source path, active page, ROI/profile, FFT, contrast, zoom, and curation state. Do not jump to the newest page unless the user follows it. - Drive page 1, page 2, the last partial page, and page 1 again. Verify the DOM contains no more than ``page_size`` scientific canvases, current-page histograms and FFT panels match the visible files, number shortcuts select a slot on the active page, and rapid page changes cannot leave stale canvases. - Verify page controls stay grouped on a narrow viewport, use the shared MUI control language, and remain keyboard accessible. Capture light/dark browser screenshots and console logs. - Keep the folder watcher append-only: only the new source file is read from disk and existing scientific files are not reread. ``Updating`` must remain truthful until the newly visible panel paints, then return to ``Watching``. - Record initial scan, append, page-switch, histogram/FFT, browser memory, comm payload, and paint timings for 20, 45, and 100+ images. The first implementation limits visible React panels but still retains and transports the full gallery; do not claim bounded Python/browser memory until an active-page transport, cache, and generation-safe prefetch path is separately implemented and verified.