Science

Science is the flowss studio for illustrated scientific figures. You compose a figure on an icon canvas across four domains (Biology, Chemistry, Physics and Mathematics) using a curated shape library, real 2-D molecular structures drawn from SMILES, typeset LaTeX equations, panel letters and your own micrographs. What sets it apart is that the studio reads what you write. Every label, arrow label, title and caption is checked as you work: reactions are balanced exactly, written relations are tested for dimensional consistency, quoted constants are compared against CODATA 2022 values, reported statistics are recomputed, scientific names and identifiers are checked for form, and every correction that can be computed is offered as a one-click fix. Drop the tiles from the Models category and the drawing itself becomes computable: a compartment model simulates, a circuit solves, a Markov chain finds its stationary distribution, and every number you pin back onto the figure carries a record of how it was computed. You open Science at /science.

At a glance

On Windows and Linux, read ⌘ as Ctrl, ⇧ as Shift and ⌥ as Alt throughout this page.

Opening Science

1. Sign in. Science requires a signed-in account; if you open a Science link while signed out, you are sent to sign in and then returned to the same address, including a shared figure in the link.
2. Go to /science, or choose Science under Studios in any studio's menu (a padlock and Starter mark it when your plan does not include it).
3. If your plan does not include Science, you are redirected to /pricing. Starter, Plus, Ultra and Enterprise all open it.

The first time the studio loads you see a short loading card while the canvas starts. The first time you open a science tool you may also see Loading the science kernel… for a moment; the tools load on demand so the studio opens quickly.

Science remembers where you were. When you come back it restores the domain you were in, whether the library was on Shapes or Templates, whether the flowss Science Agent's sheet or pane, the comment list and the science tools were open, which science tool was showing, whether you pinned the library open, and the inspector drawer's width. It also restores your last view of the canvas (pan and zoom).

The workspace

The canvas is the page. Everything else floats on it in fixed places.

The studio menu is the same in every studio: Search & commands (⌘K), Home, Studios (switch to another studio), Library, Flow (⌘J), Keyboard shortcuts (?), Theme, Layout (Modern or Classic) and Help & docs. See A tour of the workspace.

The ⋮ file menu

The ⋮ menu holds the figure's document verbs. Every row is also a ⌘K command.

There is no "New figure" row. The Science canvas holds one working figure at a time; a new figure is a cleared canvas, or a template. To keep several figures, keep them in your Library.

Clear canvas… asks first: the first press shows the toast Clear the whole canvas? It can be undone. with a Clear button. Click Clear (or choose the row again within three seconds) to clear. Clearing removes every shape and connector, every comment thread and the figure's provenance ledger. Undo brings the shapes and connectors back, but not the comment threads or the ledger (see Saving, undo and recovery). The title and caption are left as they are.

The view island

The view island at the bottom right holds Undo and Redo, Zoom out, the zoom percentage, Zoom in, and a help button. On a phone the view controls fold into the menus to save room.

Click the zoom percentage for its menu:

On a phone the menu also carries Zoom in and Zoom out.

The help button opens Keyboard shortcuts (?) and Help centre.

You can zoom from 20% to 400%. Scroll or pinch to zoom, and double-click empty canvas to zoom in.

Choosing a domain

The domain decides which shapes and which templates you see, which worked examples the flowss Science Agent suggests, and which model tiles are offered. It does not change your figure: you can switch domains at any time and mix shapes from all four in one figure.

You switch domain in three places: the four-button Science domain control at the top of the library flyout, the same control on the empty-canvas card, and ⌘K (type the domain's name).

Switching domain also returns the library to its Shapes tab and clears any search that was filtering the old library.

The tool rail

Hold Space with any tool to pan.

On a phone the rail becomes one row in the bottom dock. Select, Shapes and the insert tools come first; Hand and Box-select go last. Whatever does not fit folds into a More button at the end of the row, so every tool stays one tap away.

The shape library

Click Shapes on the rail to open the library. It opens beside the rail and stays open while you add shapes, so you can build a figure without reopening it. Close it with its close button, Esc, a click outside it, or the Shapes button again. Use the pin (Keep open) to keep it open through outside clicks and Esc; Science remembers the pin on your next visit, and a pinned library opens with the studio. On a phone the library opens as a bottom sheet.

The head of the flyout holds the Science domain control and the Shapes and Templates tabs. Below them is a search box, Search shapes….

Adding a shape

1. Find the tile. Every tile is named for what it depicts.
2. Click it, or focus it and press Enter or Space, to add it near the middle of your current view. Repeated clicks scatter new shapes slightly so they do not stack on one spot.
3. Or drag the tile onto the canvas to place it exactly where you drop it.

The new shape is selected, ready for its label. Each tile's tooltip says "Click to add …, or drag it to a spot" and shows its icon id.

Favourites

Every tile except the Models tiles has a star in its corner. Click the star to add the tile to Favourites, which then appears as the first group of the library in every domain. Click it again to remove it. Favourites are remembered in this browser only.

The groups, in order

Click a group's heading to collapse or expand it. Searching expands every matching group, and the search box's clear button empties the search. The search matches a tile's name and its icon id; molecules also match by SMILES or group, and equations by LaTeX or group.

Tip: Icons from the curated library and the bundled packs work offline and export reliably. Icons from the online search are fetched through flowss and then stored inside your figure, so they also export, but you need to be online the first time you add them.

Templates

Open Templates from the library's tab strip, from ⋮ › Templates…, or from the empty-canvas card. The tab's badge shows how many templates the current domain has.

Each template is a card with a small preview and its short name (the part of its title before the dash). Hover a card to read its full title and description.

Loading a template

1. Click a card.
2. The template replaces the whole canvas, fits it in view at no more than 100% zoom, and clears any comment pins.
3. If you had a figure on the canvas, a toast says Diagram replaced with an Undo button. ⌘Z also brings your shapes back (comment threads removed by the template do not come back).

A template does not change the figure's title or caption. Update them in the Caption tab.

Warning: If the figure on the canvas was opened from your Library, loading a template first detaches it from that Library document, so the template never overwrites the saved document. Your previous figure is still one undo away, but it is no longer linked to the Library copy.

What each domain offers

Every domain's list starts with the computable templates, the ones built from model tiles, whose captions state numbers the studio reproduces from the model itself. Then come templates built from real structures and equations, then (in Biology, Chemistry and Physics) the essentials set, then the curated figures. The lists below are in the order the cards appear; the empty-canvas card offers the first four.

The essentials set holds finished teaching and publication figures. Every label in it passes the studio's own checker, and where a figure needs measurements to show a calculation they are labelled as example values. Two of them compute: the Kirchhoff network carries a circuit model and the pulled-box free-body diagram carries a free-body model, so the numbers on them are reproduced by the Model tool.

The template lists grow over time; the badge on the Templates tab always shows the current count.

High-fidelity modules in Studio

Below the cards, a list headed High-fidelity modules · open in Studio links to templates for the same domain that render in the main Studio with a specialised engine: 3-D molecules, phylogenetic trees, pedigrees, phase diagrams, KaTeX equation sheets and similar. Each row shows the module's title and engine, and opens in Studio in a new tab. See Starting from a template, Choosing an engine and Maths, science and academic engines.

Building a figure

The empty canvas

On an empty canvas with the library closed, a card titled Compose a Biology figure (or the current domain) reads "Add shapes, start from a template, or describe it below." and offers:

• the Science domain control
• Open shapes, which opens the library with the cursor in the search box
• the domain's first four templates as one-click chips
• All N templates, when the domain has more than four, which opens the Templates tab

The prompt at the bottom of the screen is the third way to start: describe the figure and let the flowss Science Agent assemble it (see AI in Science).

What a shape can be

Selecting and moving

• Click a shape to select it. ⌘-click adds a shape to the selection, or removes it.
• With Select, hold Shift and drag empty canvas to draw a selection box; anything the box touches is selected. With Box-select, just drag.
• ⌘A selects everything.
• Drag to move. Shapes snap to an 8-pixel grid.
• While you drag a single shape, alignment guides appear when its edges or centre line up with another shape, and the shape snaps to them.
• Delete or Backspace deletes the selection. Deleting a shape also deletes its connections.

Connecting shapes

Every shape has four connection handles (top, right, bottom, left) that appear on hover. Drag from a handle to another shape to connect them; the connector starts from whichever side you dragged.

On a touch screen, or whenever a handle is hard to hit, use the Connector tool: click the first shape, then the second. A toast reminds you: "Click the shape to link from, then the shape to link to".

New connectors are curved and end in an arrowhead. Change their shape and style from the selection toolbar. The line shape you last chose there (Straight, Curved, Elbow) is also used for the connectors you draw next, until you choose another.

Labels and captions

• Double-click a shape to rename it in place. Enter commits, Shift+Enter adds a line, Esc cancels.
• Double-click a connector to edit its label in a small box at its midpoint. Enter commits, Esc cancels.
• A connector's label sits beside the line, wrapped to the space between the shapes, on an opaque white pill, so it does not cover the arrow or the cards. When the middle of the line is taken, the label is moved along it. Clicking, double-clicking or right-clicking the label acts on its connector.
• Double-clicking a molecule or an equation opens it in the Properties tab instead (see Editing a structure or equation). Double-clicking a model tile or a modelled arrow opens the model inspector.

Captions understand a small markup, the same as the figure caption: _{ } for subscripts, ^{ } for superscripts and *…* for italics. H_{2}O prints as H₂O, Ca^{2+} as Ca²⁺ and *E. coli* in italics. For a single character you can drop the braces (CO_2, 10^9). Put a backslash before a marker to print it literally (run\_2). An unclosed * is just an asterisk. The markup changes only how a label looks: the checkers read the plain text, so italics can never change whether a claim checks out. A shape's caption gets as many lines as it needs, up to four, and the card makes room for them, so captions are normally shown whole. Only a caption longer than four lines is clamped on the card; hover it to read the whole text.

Resizing, rotating and restacking

• A selected shape shows resize handles on its frame. Drag them to resize; the glyph grows with the box.
• A selected shape shows a rotation knob. Drag it to rotate; hold Shift to snap to 15°.
• Resize handles are hidden while a shape is rotated. Use Smaller and Larger in the selection toolbar's ⋯ menu instead, which work at any angle, or reset the rotation.
• Bring to front and Send to back (in ⋯) change which overlapping shape is on top.

The selection toolbar

When something is selected, a toolbar appears just above it. What it holds depends on the selection.

The toolbar hides while you are renaming, while the Connector or Hand tool is active, and while you drag.

Reaction arrows

A connector between chemical species can be a reaction arrow. Select the connector, click Reaction arrow in the toolbar and choose a form:

Click the lit form again to turn the arrow back into a plain connector.

Below the forms:

• Reagents (placeholder "above the arrow") is written above the arrow, for example a catalyst. Reagents are carried, not balanced.
• Conditions (placeholder "below the arrow") is written below it, for example a temperature or solvent.
• The rate law picker sets how the arrow behaves in a reaction model: k (mass action, v = k·Π[reactant]), MM (Michaelis–Menten), Hill, or f(…) (an expression over the figure's symbols). The fields that follow are k and, for an equilibrium arrow, k reverse; Vmax and Km; Vmax, Km and n; or a free-text rate law such as k1·A·B. Values take units, for example 0.1 /s or 1 /µM/s, or a parameter's symbol.

Changing the form keeps the reagents, conditions and rate law you typed.

Right-click menus

Right-clicking a shape that is not selected selects just that shape first.

Copy, cut, paste and duplicate

• ⌘C copies the selected shapes and the connectors between them. ⌘X cuts. ⌘V pastes. ⌘D duplicates in place, offset slightly.
• Pasting the same clipboard repeatedly cascades the copies so each lands clear of the last.
• The clipboard format is shared with Weave. Shapes copied from Science paste into Weave, and Weave shapes paste into Science. A shape that has no equivalent in the other studio arrives simplified, and the paste toast says so.
• If the browser blocks the system clipboard, the toast says the copy "only pastes inside Science".
• Pasting a block of spreadsheet cells (at least three rows, mostly numbers, and either two columns or five rows) creates a Data node instead. See Data nodes and pasted data.

Arranging a figure

Micrographs and scale bars

1. Add an image with Image on the rail, ⋮ › Import › Add an image…, or by dropping image files onto the canvas. You can add up to eight at once.
2. Images larger than 1,600 pixels on their longest edge are resampled to fit, and a toast tells you the new size ("Image resampled to … so the figure still saves").
3. Select the image. The Properties tab shows Field width (µm): type how wide the field of view is in micrometres.
4. A scale bar appears in the image's bottom-right corner. Its length and label are computed every time from the field width and the image's current drawn width, as the longest round length that fits (1, 2, 5, 10, 20, 25, 50, 100, 200, 250 or 500, in Å, nm, µm or mm), so resizing the image can never make the bar wrong.
5. Clear the field to remove the calibration.

If an image cannot be added you see one of: "That image could not be read", "That file is not an image" or "That image could not be decoded". Try saving the image as PNG or JPEG first.

The scale bar is recomputed for the exported SVG at the exported size.

Editing a structure or an equation

Double-click a molecule or an equation, or select it and click Inspect. The drawer's Properties tab opens with:

• Structure (SMILES) or Equation (LaTeX): the source, in an editable box.
• For a molecule, the molecular formula, molar mass in g/mol and any net charge, recomputed on every keystroke. A SMILES that cannot be read shows the parser's error in red instead.
• Apply (⌘↵), which commits the change. It is one undo step. Esc closes without applying.

A molecule or equation that cannot be drawn shows Invalid SMILES, Invalid LaTeX, No SMILES or No equation in red on the canvas; hover it to see the source.

The Properties tab also shows a model tile's quantities (see The model inspector) and a micrograph's field width. With nothing suitable selected, it reads "Select a modelled shape, a structure, an equation or a micrograph."

The figure title and caption

The figure's title is in the identity island at the top left; click it to rename. The title and the caption are both in the drawer's Caption tab (⋮ › Title and caption…).

The caption matters to the checker. Most of a real figure's statistics live in its caption ("t(28) = 2.20, p = .04"), so the studio checks the caption and title like every label. Any problem found in them is listed directly under the caption, as well as in the Verify tool. The title and caption travel with the figure into the JSON export, the share link, the SVG export and handoffs to other studios.

Checking the science

How checking works

Every label, connector label, title and caption is read by a set of deterministic checkers about half a second after you stop editing. Positions, colours and sizes are never part of the check, so dragging shapes costs nothing.

Four principles shape what you see:

• Silence is a valid answer. Most labels ("Mitochondrion") contain nothing checkable and produce nothing. The physics checker stays silent on mathematics such as y = 2x + 3 rather than give a wrong verdict.
• A finding carries its fix. Where the correct text can be computed, the finding offers it: "not balanced" comes with the balanced equation.
• The working is shown. Findings can be expanded to their derivation, so you can check the checker.
• Refusal beats guessing. An equation whose species admit more than one independent reaction is reported as underdetermined, not balanced one arbitrary way.

No AI model is involved in any check. The same figure always gets the same verdict.

The status glyph

The glyph beside the figure's title summarises the check:

• Empty on an empty canvas.
• Checked once the checker has read a figure and found nothing to report.
• A number when there are problems: errors and warnings. A finding that repeats for the same text on the same shape is counted once; the same mistake written on three different shapes counts three times.
• A loud Not saved pill if the browser refused the last autosave.

Click the glyph to open the status popover. Its Science check section lists up to six problems, errors first. Each row has Show, which selects and centres the shape the problem came from, and Fix, when there is a one-click correction. With no problems it reads "Every claim on the canvas checks out." The link at the bottom reads Open the Verify tool, or N more in the Verify tool when more than six problems exist. The popover also holds the figure's readiness chip, which can hand the figure to the flowss Science Agent.

What is checked

To avoid false alarms, the scientific-name checker only speaks when a label is written like a name: it resolves to a known organism, or you italicised it. "Cell membrane" and an italicised Latin phrase such as in vitro are never flagged.

The science tools

The nine science tools live in the inspector drawer's Tools tab. Open them in any of these ways:

• Science tools on the rail, then choose a tool.
• ⌘K, then the tool's name: Verify science, Chemistry tools, Units and constants, Statistics, Experiment design, Molecular biology tools, Publish check, Periodic table, Model and analyse.
• ⇧⌘X toggles the drawer on the tool you used last.
• ⌘. toggles the drawer. If one molecule, equation, image or model tile is selected it opens that shape's properties; otherwise it opens the science tools.

The drawer opens at 416 pixels wide; drag its left edge to resize it, and use Keep open to pin it. Its tabs are Tools, Properties and Caption. Within Tools, a strip of tabs switches between the nine tools.

Everything you type into a tool is kept while the drawer is closed, so you can close it, edit the figure and come back to the same numbers. Every tool shows the line "Every result here is computed from cited constants and your inputs — nothing is generated."

Verify

The full report behind the status glyph.

• A summary badge: All checks pass, or the number of issues, beside a one-line verdict.
• Issues only hides the passing claims; Show all shows them again.
• Findings grouped by kind (the groups in What is checked), each showing the checked text and a message.

Each finding row can carry these controls:

With nothing checkable, the tool says "Nothing checkable on the canvas yet." and suggests examples to try: a reaction such as CH4 + 2O2 → CO2 + 2H2O, a relation such as E = m c^2, a measurement such as 9.81 ± 0.02 m/s², or a citation such as doi:10.1038/nature12373.

Chem

In the "leave one blank" cards, the blank field is the unknown the card solves for. The thermodynamics cards work in SI: energies in J/mol (not kJ/mol), entropy in J/(mol·K) and temperatures in K. The ideal gas law takes p in Pa, V in m³, n in mol and T in K; the combined gas law accepts any consistent units for p and V. The Nernst row takes E° in volts, the number of electrons n and the reaction quotient Q.

Physics

Stats

A χ² test on a 2×2 table with an expected count below 5 is refused by name and points you to Fisher's exact test, and a 2×2 table with a zero cell is refused rather than silently corrected.

Design

Bio

Publish

With nothing drawn, Publish says it has nothing to measure and checks nothing. It never invents a size.

Table

The periodic table, all 118 elements, coloured by category. Search by name, symbol, atomic number or category (iron, Fe, 26, noble gas); non-matches dim. The tool opens on carbon. Click an element for its card: atomic number, atomic weight, category, group, period and block, electron configuration, electronegativity, oxidation states (common ones starred), ionisation energy, melting and boiling points, density, phase at 298 K and year of discovery.

Model

Runs the models drawn on the canvas and analyses data. See the next section.

Models: making a drawing compute

A drawing becomes a model when its shapes carry a model role. The easiest way is the Models group at the top of the library: each tile drops a shape with its role and sensible default quantities already set. You then write the numbers, with units, as you would in a lab notebook.

The model types

Where two model types share a tile name (such as Parameter), the tile shows the model type underneath.

Quantum gates and measurements snap to the nearest qubit lane when you drop or drag them, and their order along the lane is read from left to right.

The model inspector

Double-click a model tile or a modelled arrow (or select it and click Inspect) to open the model inspector in the Properties tab. Its badge names the model type and role. It holds:

• Caption, the shape's label.
• One field per quantity the role declares, for example Initial value, Rate, Resistance, Probability or Angle θ. Type values with units (990, 0.3 /d, 10 kΩ) or an expression over the figure's symbols (β·S·I/N). Under each field a badge shows the dimension it read. A field whose dimension does not fit turns red.
• For an arrow or a parameter, new parameter symbol (γ) and Add, to add your own named parameters. Remove one with its ×.
• Structured attributes for some roles: tick boxes, numbers, comma-separated lists, key: value maps and, for a Bayesian variable, a probability table editor with + row, + column, Paste rows (CSV, TSV or spaces), a × to remove a row, and a running sum per row. A row that does not sum to 1 shows its sum in red with a ÷Σ button ("Divide this row by its sum so it sums to 1").
• Any problems with this shape, each with Apply "…" when a fix exists.
• Apply to commit (one undo step) and Cancel.

The Model tool

Open Model in the science tools. From top to bottom it holds:

• Quantities on this figure and Paste data (see Data nodes and pasted data), always.
• The models on the canvas, described below.
• The Ledger fold (see Provenance).
• The Regress and Π groups folds, always.

With nothing modelled on the canvas, the models area reads "No model on the canvas yet." and tells you to drop a Stock, a State or a Data tile. A plain drawing of three or more shapes with no model roles at all is still read as a network, under the title Drawn graph.

When there are models, it shows:

• The number of models found, Run all, and a Show model overlays tick box (draw the last run's numbers on the figure).
• One card per model, saying how it was read ("Read as a compartment model: …") and naming any shapes it could not read (click one to open it in the inspector).
• A parameters table: symbol, value and dimension for every quantity, with buttons to edit in the inspector or show on the canvas.
• Run (or Analyse for data and networks). The results list every computed value; each has Pin, which places it on the figure as a text annotation that carries its provenance. Refusals are listed by name with their reason.
• Plots where the analysis produces them, the method line (the ledger entry, the method name and version, and the input hash), and Show working.
• Ask the flowss Science Agent about this, which sends the result and its ledger entry to the flowss Science Agent and asks whether the numbers agree with what the figure's labels and caption state. Its answer must cite the ledger entry.

Extra cards appear under particular models:

On the canvas, a model that has been run can show:

• Overlays: numbers, fills, halos and arrow widths drawn on the shapes. They are derived from the last run, disappear as soon as the model changes, and never appear in exports. Toggle them from the Model tool or ⌘K Toggle model overlays.
• A time scrubber at the bottom of the canvas for a time course: drag it to see the state at each time point without re-running.
• A Time course chart you can show or hide.
• For a network, a picker for the centrality the node halos scale by: Degree, Betweenness, Closeness, Eigenvector or PageRank.

Long computations (sweeps, heatmaps, Monte Carlo and fits through a drawn model) run in the background so the canvas stays responsive. The sensitivity card shows a progress percentage and a Stop button; the Fit button reads Fitting… while it works.

Data nodes and pasted data

A Data node is a table on the canvas, drawn as a table. Create one by:

• pasting spreadsheet cells onto the canvas (at least three rows, at least 60% of the cells numbers, and either two columns or five rows);
• dropping a .csv, .tsv or .txt table file onto the canvas (only the first file of a multi-file drop is read, and the node is named after the file);
• the Data tile in Models.

The toast reports what was read ("Read 120 rows × 3 columns (t, A, B), time column 't'"), and how many rows were left out because a cell was missing or not a number. A column headed time, tau, τ or a lower-case t (with or without a unit, such as t (s)) is taken as the time column; an upper-case T is read as temperature, not time. A Data node stores at most 5,000 rows and 64 columns; a larger table is placed with the rows that will be saved, and the toast says so. A file that cannot be read gives "Could not read" and the file name.

The top of the Model tool has two panels that work on the figure's numbers and on data you paste:

• Quantities on this figure (a fold, open by default): every number with a unit written on a label, with its symbol, value, dimension and SI value. Mark two as a and b, choose a ÷ b, a × b, a + b or a − b, and click Derive to compute a new quantity with its uncertainty propagated. If none is written yet, it explains how ("γ = 0.1 /d", "40.0 mA").
• Paste data: paste CSV, TSV, semicolon-separated (decimal comma) or space-separated data, with or without a header. Unreadable cells are listed by row. Four cards then work on its columns, each with its own column pickers and Compute (then Again): Describe (n, mean, SD, median, IQR, D'Agostino–Pearson normality, histogram), Mean CI (95% t interval), Regression (ordinary least squares y = a + b·x with its 95% confidence band) and Welch t (two-sided, with Cohen's d and Hedges' g). Pin a value; Describe and Regression also offer Pin chart. A paste is read up to 50,000 cells and 512 columns.

At the bottom of the Model tool are two more folds:

• Regress: a generalised linear model (gaussian, binomial or poisson) over columns you paste into its own box. Pick the outcome and predictors; text columns are dummy-coded against their first level and marked factor; the family is suggested from the outcome and can be overridden; add interactions as a:b (separate several with commas).
• Π groups: Buckingham's dimensionless groups for a list of variables ("F depends on ρ, v, L, μ"), named where known (Re, Nu, Pr, Fr …), with the reduced hypothesis. Use quantities on this figure takes the symbols from your labels. Write a unit in brackets, such as μ [Pa·s], to fix a symbol's meaning.

Fitting a model to data

1. Have a Data node on the canvas, with a time column for a time-course fit.
2. In the Model tool, find the Fit card under the Data node's result. With several Data nodes, choose the table first.
3. Choose a model: one of the drawn models on the figure (a compartment model or a reaction network), or a curve from the catalogue: Straight line, Exponential growth or decay, Zero-, First- and Second-order integrated rate laws, Michaelis–Menten, Hill equation, Logistic growth, Arrhenius, Four-parameter logistic (dose–response) and Five-parameter logistic (asymmetric dose–response). Each curve is listed with its formula.
4. For a curve, choose the x and y columns. For a drawn model, check the table of which column observes which shape, and untick under fit: any parameter you want held at its written value.
5. Optionally give σ, a known measurement standard deviation. It weights the residuals and adds χ²; leave it empty to estimate the noise from the data.
6. Click Fit (it reads Fit again afterwards). You get each parameter's estimate ± SE and its 95% CI, then n, degrees of freedom, RSS, s², χ² (when you gave σ), AIC, BIC and R² where they apply, the rank and condition number, the correlation of the estimates (correlations above 0.95 are highlighted), and the fit and residual plots, each with Pin chart.
7. For a drawn model, Apply fitted parameters writes the estimates back onto their own arrows as one undoable step, and records the fit in the ledger. A value you have changed since the fit is skipped, never overwritten.

A fit is refused as not-identifiable when the data cannot pin down every parameter, and the refusal names the direction that cannot be resolved. A drawn-model fit with no compartment model or reaction network on the figure says "No compartment model or reaction network reads on this figure — draw one, or fit a closed form instead."

Model exports

Provenance: pinned numbers and the ledger

Every computation you pin, apply or record (from the Model tool, an analysis card, the Fit card or the flowss Science Agent) becomes an entry in the figure's ledger, numbered L-01, L-02 and so on. An entry records the method and its version, exactly which shapes and connectors it read, any literal inputs (such as a pasted table's digest or a seed), the results and the working, and a hash of the inputs.

A pinned number is a text annotation stamped with its ledger entry. The checker compares the stamp with the figure on every check, without recomputing:

Each stale or untraced finding offers two fixes:

• Recompute runs the models again and rewrites the pinned number and its stamp.
• Detach drops the stamp. The number becomes an ordinary label, still checked like any other.

The Model tool's Ledger fold (shown once the ledger has an entry or the figure has a pinned number) lists every entry with its method, input hash, status and the pins that cite it (click a pin to select it on the canvas), counts pins as current, stale and untraced, and has Recompute all, which re-runs every model a stale or untraced pin came from, in dependency order. It is greyed out when every pin is current. Each entry also has a calculator button, How this was calculated, which opens a dialog with the result, uncertainty, inputs, exclusions, method and version, parameters, transformations, assumptions, units, randomness, diagnostics and limitations, and a Download reproducible JSON button. ⌘K Recompute stale pinned numbers does the same. ⌘K Export ledger (every computed number with its method, inputs and hashes) downloads science-<domain>-ledger.json.

The ledger travels with the figure in the JSON export, the share link and the SVG export's metadata. It is not sent to other people in a live session: they recompute, and until they do, a pin reads as untraced on their screen. A ledger holds up to 500 entries; when it is full, entries no pin refers to are removed to make room. Clearing the canvas clears the ledger.

AI in Science

AI in Science draws and edits figures. It never decides whether a figure is correct: every check described above is arithmetic, and runs without AI.

AI requires your own AI key (from Starter) or hosted AI points (from Plus). See AI points and limits and Bring your own AI key.

The prompt: Quick and Agent

The prompt at the bottom of the screen reads Describe a figure or a change… and has two modes.

⌘E switches the prompt to Quick and toggles a sheet of worked examples for the current domain, titled with the agent and the domain (for example flowss Science Agent — Chemistry); ⌘K Describe with AI opens the same sheet. The sheet reads "Type a figure below, or start from one of these." (or, with a figure on the canvas, "Type a change below, or start a new figure from one of these."). Click an example to generate it. When you click into the prompt, the domain's first three examples also appear as short chips above it (in Biology, Glycolysis, Antibody binding to antigen and Malaria life cycle); Hide suggestions dismisses that row, and Science remembers the choice.

Note: Clicking a worked example generates a new figure even when one is on the canvas. Your previous shapes are one undo away, but the canvas's comment threads are removed and a figure opened from your Library is detached from it first, so the Library copy is not overwritten.

If the request fails, the reason appears in the prompt with a dismiss button. On a plan without AI, add your own key or upgrade (see AI points and limits).

The prompt's attach button opens the import dialog, so you can start from a photo of a whiteboard, a screenshot or a pasted diagram.

The flowss Science Agent

In Agent mode the flowss Science Agent works as the Science Studio specialist, a multi-step scientific illustrator: it plans, draws and checks the figure step by step, and it runs the science checker before it finishes. Open it with the prompt's Agent mode or ⌘K Open the flowss Science Agent. Its sheet is titled flowss Science Agent — Science Studio specialist. Its examples include drawing glycolysis as a left-to-right pathway, building a food web, and diagramming the nitrogen cycle.

Behaviours to know:

• The flowss Science Agent works on the figure you are looking at. When it finishes, its changes are applied and a toast says The flowss Science Agent applied changes, with Undo. If it only read the figure, the toast says The flowss Science Agent made no changes to the figure.
• Revert this run undoes everything the run changed.
• If you edit the canvas while the flowss Science Agent is working, your version is kept and a card titled You changed it too offers Use its version or Keep mine.
• Some changes arrive as proposals under Waiting for your approval, with Approve and Reject (or Re-run and Dismiss if the canvas has changed since).
• You can attach an image for the flowss Science Agent to read, or paste one.
• When the flowss Science Agent quotes a computed number, it cites a ledger chip such as L-04. Click the chip to open that entry in the Model tool.
• The flowss Science Agent can run the models, fit data and run sensitivity analyses itself, and pin results onto the figure with their provenance. It may only cite numbers its own tools produced in that run.
• When you describe a study or protocol, the flowss Science Agent can critique the experimental design with the same deterministic checker as the Design tool.
• A run takes at most 12 steps.

⌘K requests aimed at the platform's lifecycle specialists (such as review, critique, refine, publish and verify) also open in this pane. For more on agents, see The flowss Studio Agent and its modes.

Importing

Exporting and sharing

Export formats

Exports capture the whole figure, wherever you have panned, on a white background with 40 units of padding. Comment pins, collaborators' cursors, selection frames and model overlays are never included. File names say which domain you were in: science-biology, science-chemistry, science-physics or science-math, for example science-chemistry.svg. A journal-width PNG or TIFF adds its width, for example science-biology-1200px.png.

If a journal-width export would exceed what the browser can allocate, the toast says how many pixels short of the request it is, with the dpi actually achieved, and suggests simplifying the figure or exporting SVG. See Exporting your work.

Share link

Share › Copy link (or ⌘K Copy share link) copies a link that carries the whole figure inside it, including its title, caption and ledger. Whoever opens it sees your figure replace their canvas, and their figure's title, caption and ledger are replaced too. Their shapes are one undo away, but the comment threads on their canvas are removed. They need a signed-in account whose plan opens Science. Anyone with the link can read the figure, so treat it as public.

• An empty canvas cannot be shared: the toast says "Nothing to share — the canvas is empty".
• A share link can be at most 1,000,000 characters. Embedded images make figures large; if yours is too large, the studio says so (with the size in MB) and suggests exporting JSON or sharing live instead.
• If the browser blocks the clipboard, the link is placed in the address bar for you to copy by hand.
• A share link that cannot be read, that holds an empty figure, or whose figure cannot be built never replaces your own saved figure; you are told your own figure was opened instead.

Send to another studio

⋮ › Send to another studio… (or Share › Send) converts the figure and opens it in another studio. See Moving work between studios.

Collaboration

Comments

1. Choose Comment on the rail (or press C, or ⌘K Add a comment pin).
2. Click empty canvas where the comment belongs. The tool switches itself off after one pin.
3. The New comment box opens. Type in Write a comment… and press Enter (or the send button) to post it; Shift+Enter adds a line and Esc closes the box.

Pins are amber, blue while open, and green when resolved. Click a pin to open its thread, where you can reply (Reply…), Mark resolved or Mark unresolved, delete a message, or Delete this whole thread. Deleting a thread's first message deletes the thread.

Comments in the top-right island (or ⌘K Show all comments) lists every thread, newest first; click one to fly to its pin and open it. With none, it reads "No comments yet. Pick Comment (C) on the rail, then click the canvas."

Comments belong to the figure document, not to the drawing: they are saved with the figure but never exported, never selected with shapes and never part of undo. A figure holds up to 200 threads, a thread up to 50 messages, and a message up to 2,000 characters; past those limits the oldest thread or replies are removed or the text is trimmed, and a toast says so. See Comments and review.

Live sessions

Share › Start live session (or ⌘K Share live (collaborate)) copies a link that lets others edit the figure with you in real time, with live cursors and shared comment threads. Live sessions start on Plus (up to 3 people, including you) and are uncapped on Ultra; voice and screen sharing are Ultra. On a plan without live sessions the row is locked and shows what opens it. Because Science itself opens on Starter and above, everyone who joins needs a plan that opens Science.

The Share popover holds:

Your undo only undoes your own work, never a collaborator's. When you join someone else's session from a link with a figure already on your canvas, your figure is first set aside in the Vault (restore it from Flow Dock › Vault), and Undo also brings it back once you leave. See Live collaboration.

Saving, undo and recovery

• The figure autosaves in your browser a moment after each change, and again when you leave the page or switch tabs.
• A figure opened from your Library syncs its changes back to that Library document. Loading a template, clearing the canvas, importing, generating with AI on a non-empty canvas or opening a share link detaches the canvas from the Library document first, so those actions never overwrite it. See Cloud sync, offline work and devices.
• Undo keeps the last 50 steps. A drag, a resize, a rotation or a run of stepper clicks is one step.
• Undo covers the shapes and connectors (including their labels, styles and model values). It does not cover the figure title, the caption, comment threads or the provenance ledger. So after you undo Clear canvas…, the shapes come back but the comment threads do not, and pinned numbers read as untraced until you recompute them.
• If the browser refuses to save (storage full or blocked), a toast says "This figure is too large to autosave in this browser. Export it to keep a copy — changes from here on are not being saved." and the status glyph shows Not saved until a save succeeds.
• If your saved figure cannot be read, the studio opens empty, sets the unreadable copy aside instead of overwriting it, and offers Download copy.

See Version history and undo.

Keyboard shortcuts

The zoom keys are bare keys so that ⌘+ and ⌘− stay the browser's own page zoom. Single-key shortcuts are ignored while you are typing in a field. See Keyboard shortcuts.

Using the checkers from code

The checkers and model tools in Science are also available through the flowss REST API with an API key: verify a whole figure or a list of labels, balance an equation, compute a molar mass or oxidation states, convert units, check a written relation, check a quoted constant, validate identifiers, check names, critique a study design, plan for a journal's figure requirements, run and fit models, and more. Every answer is deterministic, with no AI involved. See The REST API for the operations and how to call them.

Tips

• Write formulae, not names, on the shapes of a reaction you want balanced: "CH4" can be checked, "methane" cannot.
• Put your statistics in the caption. The caption is checked like every label, and reported p-values are recomputed from their statistic.
• Italicise organism names with *…* around the whole name. That is what tells the checker the label is a scientific name.
• Use the Connector tool on a touch screen or a dense figure instead of aiming at the small connection handles.
• Pin the library open (Keep open) when you are assembling a large figure from many shapes.
• Plan for the journal before you export: the Publish tool knows the column widths and gives you buttons to export at exactly that size.
• When a pinned number goes stale and you mean to keep it as a fixed claim, choose Detach on its finding, so it stops being tracked against the model.
• A Monte Carlo run only draws parameters whose values state a ±. Write 0.3 ± 0.02 /d, not 0.3 /d, for anything you want varied.

Limits and known constraints

• Science opens on Starter and above. Free accounts cannot open it.
• The canvas holds one working figure per browser. Keep several figures in your Library.
• Curved and Rounded in Line shape currently draw the same curve.
• A connector set to No arrowhead keeps that setting when you reopen the figure in the same browser, but gets its arrowhead back when the figure arrives through an import: a share link, a pasted or imported JSON file, or an edit by the flowss Science Agent. Set it again after such an import, before you export.
• Science exports do not add credits for Font Awesome or other Creative Commons Attribution icons placed on the canvas. Add the credit yourself; see Copyright, licences and attributions.
• Some templates share a short card name, such as Physics, Earth Science, Mechanism or Free-body diagram. Hover a card to see its full title.
• Typeset equations are embedded in SVG exports as 3× images, not vector paths. PNG, TIFF and PDF exports are raster images.
• Undo does not cover the title, caption, comment threads or ledger. Clearing the canvas, loading a template, opening a share link or generating from a worked example over an existing figure removes the comment threads for good.
• Icons added from the online Iconify search need a network connection the first time they are fetched.
• Identifiers are checked for form and check digit only. Nothing is looked up online, so a well-formed DOI that does not exist passes.
• The hazard card matches whole labels only. "Beaker containing sodium hydroxide" is not recognised; "sodium hydroxide" is. It is never a substitute for a safety data sheet.
• Images are resampled to at most 1,600 pixels on their longest edge. You can add up to eight images at once.
• A share link is limited to 1,000,000 characters.
• A Data node stores at most 5,000 rows and 64 columns. The Paste data box reads up to 50,000 cells and 512 columns.
• Model computation budgets: up to 12 qubits (8 for recognising a Bell, GHZ or QFT circuit, and 6 on the smaller side of an entanglement cut), Bayesian variables with at most 6 states and 4 parents, eigenvector centrality and PageRank up to 400 nodes (the other centralities are still computed above that), Boolean attractors up to 20 nodes, and Monte Carlo runs below 20,000 (19,999 at most). A model that hits a budget is refused by name with the limit.
• The ledger holds up to 500 entries (entries no pin refers to are removed to make room) and is not shared live; collaborators recompute.
• A figure holds up to 200 comment threads of up to 50 messages each, and a message holds up to 2,000 characters.
• Undo keeps 50 steps.
• A flowss Science Agent run takes at most 12 steps.

Troubleshooting

"Nothing to export — the canvas is empty". There are no shapes on the canvas. If you expected a figure, check that you are in the right browser and that your figure was not detached by loading a template; press ⌘Z to bring back a replaced figure.

"Couldn't export PNG." (or SVG, PDF, TIFF), followed by a reason. The browser could not produce the file. Try again after the figure has fully drawn; for very large figures, export SVG instead.

"Exported SVG — … structures could not be captured". A molecule had not finished drawing when you exported. Wait for every molecule to appear and export again.

"Exported SVG — … icons could not be embedded". Those icons load from the network. Reconnect, or reopen the figure, and export again. The rest of the figure is in the file.

"Exported at … px — … px short of the … px asked for, which is the canvas limit." The figure is too large for the browser to raster at that width. Simplify the figure or export SVG.

"This figure is too large for a share link (… MB — imported images are embedded)". Embedded images make the link too long. Export JSON or share it live instead.

"Nothing to share — the canvas is empty". Add something to the figure before copying a share link.

"The browser blocked the clipboard — the link is in the address bar of this tab instead." Copy the link from the address bar.

"That share link could not be read, so your own saved figure was opened instead." The link was truncated or altered in transit. Ask the sender to copy it again. The variants "That share link holds an empty figure…" and "That share link could not be opened as a figure…" mean the sender shared an empty canvas or a damaged figure.

"This figure is too large to autosave in this browser." Your browser's storage is full or blocked. Export JSON straight away to keep a copy, then remove large images or free browser storage.

"Your saved figure could not be read, so the studio opened empty." Click Download copy in the toast to save the unreadable copy, then import it or send it to support. See Reliability, status and support.

"No model on the canvas to run — drop a Stock, a State or a Data tile from the Models category". Nothing on the figure has a model role. Add tiles from Models, or load a computable template.

"No reaction network on the canvas — drop Species from the Models category and draw reaction arrows between them." SBML export needs Species tiles connected by reaction arrows.

"No quantum circuit on the canvas — drop Qubit and Gate tiles from the Models category and wire the gates along their lanes." OpenQASM export needs a circuit. If it says the circuit "has errors to fix first", open Verify and fix the Model findings.

"OpenQASM 2.0's qelib1.inc has no …". That gate is not in the standard gate library. Decompose it into gates it has and export again.

"This result was computed on an earlier version of the figure — run the model again, then pin it." You edited the figure after the run. Run the model again, then pin.

"Nothing to apply — the parameters have changed since this fit; fit again". Fit again against the current figure.

"No model on the figure computes … any more". A pinned number's model is gone. Use Detach, or pin the number again from a new result.

"Run a model first — there is no result to pin" (or "… to ask about"). The ⌘K pin and ask commands work on the last run. Run the model first.

"Nothing in the ledger yet — run a model first". The ledger export has nothing to write until something has been computed and recorded.

"No model on the canvas to recompute from — the pinned numbers were left as they are". The pins' models are no longer on the figure. Detach them, or rebuild the model.

"This result could not be recorded in the ledger, so it was not pinned: …". The reason follows the colon. A result that cannot be traced is never pinned.

"No feeding link on the figure". For Arrange by trophic level, label the arrows "eats" (prey → predator) or set their relation in the model inspector.

A finding's Apply says "Nothing to change". The text the fix targets is no longer on that shape, usually because you already edited it.

"Couldn't reach the online icon catalogue — check your connection. The bundled icons above are still searchable." Online search is unavailable; rewording the search will not help. Use the curated and bundled icons, or try again when you are back online.

"Comment limit reached — the oldest thread was removed to make room" or "Thread is full — its oldest replies were removed". The figure is at 200 threads, or the thread at 50 messages. Resolve and delete old threads you no longer need.

"Voice rides a live session — hit "Share live" first". Start a live session before joining voice. Voice and screen sharing require Ultra. "Screensharing rides the voice call — join Voice first" means the same for screen sharing: join voice, then share your screen.

"Auto-layout failed". The layout could not be computed for this figure. Undo is not needed, because nothing moved; arrange the shapes by hand or with Align and distribute.

The status glyph shows a number but I cannot see a problem. Click the glyph and use Show on each row to select the shape it came from. Problems in the title or caption are listed in the Caption tab.

Related pages

• Science: verified science, not plausible science (the design of the checking layer)
• Core concepts and glossary
• A tour of the workspace
• Weave
• Figure
• Lab
• The icon library
• Importing and converting
• Your library
• Moving work between studios
• Comments and review
• Live collaboration
• AI in flowss
• Exporting your work
• The REST API
• Plans and what they include
• Keyboard shortcuts
• Troubleshooting and FAQ