curve.fit changelog

New features, usability improvements, and fixes. Calculation notes highlight updates that may affect your fitted values, uncertainties, or plots.

Some earlier version numbers were assigned later. Their dates group related changes and may not match the exact date they became available. For reference datasets and details of the numerical checks, see Numerical validation.

2.17.0 — October 16, 2026

Maintenance update. Fitting, reports, numerical methods and instructor workflows are unchanged.

2.16.1 — October 16, 2026

• Browser Report plots keep their view fixed so swiping over a plot scrolls the page. Point hover and PNG/SVG downloads remain available; workspace plot controls are unchanged.

2.16.0 — October 15, 2026

• Browser Report presents fit statistics and uncertainty settings in separate tables, with interactive data/fit and residual plots using the workspace plotter.
• Evaluate Model uses one list for forward and inverse calculations. The PDF-style Fit Image appears at the bottom of the report and loads when approached.
• Browser Report identifies the session's current Lab Space, course, section, instructor and student/group when available.
• New workspace and Lab Spaces video tours explain the main workflows with short, captioned videos and selectable chapters.

Calculations, stored precision, parameter uncertainties and numerical tolerances are unchanged.

2.15.0 — October 13, 2026

• Check inputs opens over a darkened page when Fit finds missing values, invalid numbers, nonpositive uncertainties, zero-valued uncertainty rules or model-domain problems. Go beside each problem moves to the entry. After you close the dialog, the affected cells stay outlined (red to fix, amber for warnings) and explain themselves when you point at them, and a count on Fit shows how many entries still need attention. Outlines and the count clear as you correct each entry, and the plot is never covered.
• Fit failures, temporary failures, import errors and clipped-paste notices appear in the same dialog, and your entries are preserved.
• When a custom equation overflows, divides by zero or leaves its real domain at a value the solver tries (for example 10^A with a large A), the solver now steps back and keeps searching instead of stopping. If no usable step remains, the message explains which problem occurred.

Calculation note: Fits that completed before give identical results. Custom-equation fits that previously stopped with an overflow, division-by-zero or domain message may now complete; such a result is kept only if a second, confirming solve from it meets no invalid value. A negative number raised to a fractional power at the starting values is now reported as a domain error. Numerical tolerances are unchanged. Zero measurements remain valid for models defined at zero; logarithmic-axis guidance is a display warning, not a reason to reject a fit.

2.14.2 — October 11, 2026

• Parameter limits always name their parameter, as in b ≥ 0, b ≤ 5 or 0 ≤ b ≤ 5. This applies to Browser Report, Single and Double PDFs, the combined Parameter Limits text in Data & Fit CSV, and the Limit box tooltip in Results.
• Single and Double PDFs show parameter limits in a compact Limits column of the parameter table instead of extra lines, so reports with limits keep their full-size plot. A parameter held at a limit shows AT LIMIT beside its n/a uncertainty, and a fixed parameter's ignored limits appear in gray.
• The Join Lab Space button appears with the other toolbar buttons as the page loads, so the toolbar no longer shifts once your session finishes loading.

2.14.1 — October 11, 2026

• Browser Report shows each parameter's complete limits, including both bounds when it is at a limit and saved bounds ignored for a fixed parameter.

• Saved fits retain their parameter limits when reopened, so reports, downloads and restored settings can use the original bounds.

• Data & Fit CSV and Excel parameter tables include separate numeric lower and upper limits and at-limit status. CSV also identifies fixed parameters.

• Single and Double PDFs print complete limits in compact annotations. Reports with limits keep the plot and parameter table within the page without clipping; ordinary curve reports without limits keep their layout.

• Browser Report identifies the measurement uncertainty choices and their saved rules. Its model equations match the workspace; Excel uses readable equation text.

• Weighted Mean PDF and PNG reports now use the same layout and styling as other models, with sharp plot artwork in PDFs, consistent residual captions and number-format notes. Its PDFs also show the existing Zero degrees of freedom warning for a single observation.

• Browser Report keeps waiting for a download when the service is briefly busy, as the workspace already did.

• Lab Space and instructor controls recover when a connection stalls, without repeating a change that may already have been made.

Calculation note: Fitted values, uncertainties and solver behavior are unchanged. This corrects the information shown in exports.

2.14.0 — October 10, 2026

• Instructors can create Lab Spaces at Instructor, share a six-character code, link or QR code, and review joined sessions and their saved fits through Browser Report. Students need no account; unfinished entries stay in their session.

• Join Lab Space links the current session to a Lab Space until Leave. Saved fits in a joined session are visible to the instructor. Edit changes the group name for that Lab Space or, with another code, moves the session to a different Lab Space.

• Fork and Reset stay in the current Lab Space. Examples start unjoined, and Walkthrough practice cannot join a Lab Space. Leave starts a new, unjoined session; the old session and its links remain available.

• Hand In This Fit, under Export, marks an exact saved fit for attention. It is optional, can be used for multiple fits, and repeated clicks keep one marker. Existing reports and downloads work as before.

• The instructor's Sessions list shows the most recent successful fit first, followed by sessions without one, most recently joined first, and sorts again at every refresh. Lab Spaces with more than 100 sessions continue on Next Page, each page sorted on its own.

• Instructors can hide a session and restore it from Show hidden, which adds hidden sessions to the list, marked Hidden. They can also close or reopen a Lab Space, archive a closed Lab Space, and delete a Lab Space without deleting scientific work. Closing prevents new joins and hand-ins while existing sharing continues. The instructor view refreshes automatically while it is open and pauses after 30 minutes without activity.

• An instructor can keep up to 100 Lab Spaces, archived ones included, and each Lab Space holds up to 1,000 sessions; rejoining, Fork and Reset do not count. A session can join, switch, leave or rename at most 10 times in five minutes.

• Browser Report shows Fit & Residuals directly after the model summary, before Evaluate Model.

• The instructor page's left navigation runs the full height of the page and opens Teaching Resources, Model Details, Help and curve.fit itself in a new tab. On My Lab Spaces, Show archived is a toggle, + Add creates a Lab Space, clicking anywhere on a row opens it, and each row's Archive button closes and archives it after confirmation.

Fitted values, uncertainties, warnings and calculation methods are unchanged.

2.13.1 — October 9, 2026

• Help now lists help@curve.fit for support and info@curve.fit for general questions and model suggestions. The instructor handout uses info@curve.fit.

Calculation note: Fits use the official ODRPACK95 3.2.0 release of the solver. Fitted values, uncertainties and saved results are unchanged.

2.13.0 — October 5, 2026

New models and clearer names

• Five new models: Lorentzian Peak for spectral line shapes, with its center and full width at half maximum; RLC Resonance, the current in a series RLC circuit, which reports the resonant frequency f₀ and quality factor Q; Exponential Decay (Time Constant), which reports τ directly; Logistic with Baseline, an S-shaped step between two nonzero plateaus; and Single Slit Diffraction. Each chooses its own starting values and has a model page.
• Lorentzian is now Driven Oscillator Amplitude, which describes what it fits: the displacement of a driven oscillator, or a capacitor’s voltage. For a circuit’s current, use RLC Resonance; for a line shape, Lorentzian Peak. Hill Sigmoid (4 Parameter) and Logistic (5 Parameter) are now Logistic 4PL (Hill) and Logistic 5PL. Saved fits and links are unchanged; only the names shown are new.
• A custom equation written in one of the new forms, such as A*exp(-x/B)+C, suggests the matching built-in model.
• Every model page opens an example dataset with uncertainties, at /examples/<model>. Five use certified NIST data with the model’s form (Pontius, DanWood, Rat42, MGH17 and Misra1a), Linear and Gaussian use curve.fit’s Pearson and Gaussian data, and the rest use simulated measurements.
• Automatic starting values are more reliable for Exponential with Offset, Power Law, Power Law with Offsets, Logistic, Gaussian, Driven Oscillator Amplitude, Oscillator, Damped Oscillator, Two Slit Interference and Arrhenius. Zero or negative data no longer stop the first three with an error, and a blank Guess now finds the peak, frequency or fringe pattern far more often.
• Damped Oscillator automatic guesses skip unusable trial curves, so an overflowing trial no longer prevents fitting valid data with negative x values.
• Fits with a parameter or x value far below 1 now get their model uncertainty band and Evaluate uncertainties. Before, a custom equation such as Misra1a’s A(1 − exp(−Bx)) was plotted without a band and Evaluate reported larger x values as undefined, and Evaluate showed no uncertainty for the Natural Log example.
• A custom equation whose plot extends past where it is defined, such as A*sqrt(x-B)+C on an axis reaching below B, now draws its model uncertainty band wherever its curve is, as built-in models do, and Browser Report model samples include δf(x) there. In PDFs and PNG images, a band with a gap on the right or in the middle of the plot, such as Natural Log fitted to negative x, is now drawn correctly; before, it gained a false gray stripe across the plot.
• A fit that the solver abandons with a numerical error now reports that error, with advice to try other starting values or another model. Before, it appeared as a result with zero uncertainties and a reduced χ² of 0, which the solver had not computed. Saved fits are unchanged.
• Power Law, Power Law with Vertical Offset and Power Law with Offsets now fit data with a point at x = 0 (x = xₒ with offsets), and Logistic 4PL and 5PL accept a zero-dose control at x = 0, including when choosing starting values. Before, such a fit stayed at its starting values, shown with zero uncertainties, or reported an error.
• Poisson fits handle large counts. Counts above about 170 no longer overflow, values far out in the tail are computed accurately instead of as 0, and the starting λ comes from the mean of the data, so means in the hundreds fit; before, such fits failed or stayed at their starting values. Cauchy's automatic starting values are now its exact least-squares solution, so other units of y no longer add a false "Not full rank" warning.

Calculation note: Fits with every Guess typed give the same results as before, as do models not named above. For the ten models above, a blank Guess starts from a better estimate: fits that failed now run, and fits that settled on a worse solution can reach the best one. Logistic 4PL (Hill) curves have two equivalent forms; a blank Guess now gives the one with positive B, so A is the plateau at low x. The new models’ results were checked against independent least-squares fits. Fitted values are unaffected by the band changes. Bands and Evaluate results that were already shown are unchanged, except that a PDF or PNG band with a gap on the right or in the middle loses its false stripe; otherwise the changes only supply values that could not be computed. Fits without a point at x = 0 are unaffected by the x = 0 corrections. Poisson and Cauchy fits reached the same results as before in every trial; only their starting values changed. Poisson values are unchanged except where the old formula overflowed, showed 0 or lost digits (counts above about 170, or far out in the tail); these are now accurate.

2.12.1 — October 5, 2026

• Fit tabs show Model Info under the fitted model's equation, as Setup does. It opens that model's page in a new tab, including Custom Function's.

2.12.0 — October 5, 2026

Custom functions, teaching labs and model pages

• Custom equations accept eleven more functions: ln, the inverse hyperbolic functions asinh, acosh and atanh, the hyperbolic cotangent coth and its inverse acoth, fact for factorials (computed as Γ(x + 1)), degrees and radians to convert angles, and the Bessel functions besselj0 and besselj1 (J₀ and J₁) for circular apertures and membranes. Saved fits, Browser Reports and PDFs typeset them.
• A new Custom Function page lists every function and constant, with tips and examples. Help and the custom-equation guide carry the same function table.
• The teaching labs page and its downloadable PDF describe the features added since 2.10: Fit Statistics, 1σ or 95% intervals, parameter limits, Evaluate's y → x direction and lists, AIC comparisons in Previous, XLSX workbooks, PNG and SVG plot images and typeset custom equations. Its model library links to every model page, and its screenshots and example Single and Double PDFs show the current version, with the ±1σ band.
• A four-page instructor handout, linked from the teaching labs page, summarizes the model library and custom equations, fits the Pearson–York example with three uncertainty settings, compares the results with the published values and annotates an example report.
• In Setup, Model Info under the model's equation opens that model's page in a new tab, including Custom Function's.
• Model pages suggest Evaluate's y → x for calibration curves and AIC for choosing between models. Help uses the current names for Evaluate and Rule.

Calculation note: Built-in models and previously accepted custom equations fit, evaluate and report exactly as before. Only equations that use the new functions are newly accepted.

2.11.1 — October 5, 2026

• Custom equations in saved fit tabs, Browser Reports and PDFs use mathematical notation, including fractions, powers and function names. Equations fit within the existing display area; Setup keeps the editable equation as typed.
• Custom-equation formatting preserves the stored equation and fitted results. If the browser cannot typeset an equation, it shows the original text.

2.11.0 — October 3, 2026

Models, statistics and exports

• Fit Statistics show point and parameter counts, degrees of freedom, χ² p-values, AIC/BIC and adjusted R² where applicable. Choose 1σ or 95% parameter intervals; the plot's band, PDFs, Browser Report, Excel and CSV use the selected level.
• Model Evaluator solves y → x within the fitted range, chosen with an x → y | y → x switch, and evaluates lists of up to 100 points, with CSV download and clear notices for multiple roots or flat regions. A blank δx or δy is 0; its results are always 1σ.
• Previous groups saved fits by data set and compares their recorded AIC. Load older fits in bounded batches to extend the comparison.
• Excel workbooks are available from Results and Browser Report, including full-precision tables and the plot. Plot toolbars now offer separate PNG and SVG downloads.
• Weighted Mean uses the same plot styling and tools as other models, with observation numbers on x. Switching between model types preserves the correct axis ranges; Home restores the data bounds after a zoom and tab change. Browser Report typesets LaTeX math in its heading and labels.
• The Residuals view shows error bars as the PDF does: ±δy or ±δx, or ±1 on the effective residual when both uncertainties are used.
• Every built-in model has a public page with its equation, parameters, tips and a link to start fitting with that model selected.
• Four new models: Logistic (5 Parameter), Exponential Decay (Half-life), Double Exponential, and Gaussian on Linear Background.
• Lower and upper parameter limits can constrain a fit, set from a Limit box beside Fix. Parameters that reach a limit are marked "at limit" with an unavailable uncertainty; the remaining uncertainties are conditional on that boundary.
• Report titles and axis labels can use accented and Greek letters and symbols such as °, µ and ±. A label that is not valid LaTeX, such as Speed 50% of c, now prints as typed instead of stopping the export; an axis label no longer loses the text after a bare %, and <, > and | outside math print as themselves rather than as ¡, ¿ and a dash. A character a report cannot print, such as an emoji, is named before export.
• The workspace no longer loads advertising, and its scripts and styles download about 30% smaller.
• The Expression option for measurement uncertainties is now called Rule, and every toggle shares one design. Model Evaluator is now Evaluate and Results is Metrics; Parameters and Evaluate gain ⓘ help like Absolute | Nominal, and warnings share one label style.

Calculation note: Limits change the fitted solution when active. Wide intervals preserve interior solutions, and width/phase conventions respect the requested bounds. Existing unbounded saved solutions retain their published values and uncertainties. The plotted model band is now ±1 standard uncertainty at 1σ (previously a 68.27% Student-t interval, slightly wider with few degrees of freedom) and the Student t or normal 95% interval at 95%. Fits saved earlier keep the 68.27% band they were published with, in the page and in new exports.

2.10.0 — October 3, 2026

Fitting library

• curve.fit now fits with ODRPACK95, the maintained successor to the ODRPACK solver it used before. The weighted orthogonal distance regression and least-squares methods, starting values, stopping rules and uncertainty conventions are unchanged.
• Reports and Model Evaluator results for fits saved with 2.9.0 (September 23, 2026) or later keep their published numbers. Older fits, and any saved without their solution, are recalculated with the new solver when a report or evaluation needs them; the calculation note describes the differences.

Calculation note: Built-in models give the same fitted values and uncertainties as before, typically to nine or more significant digits. Custom equations, which use finite differences, can now reach a slightly better fit: in the Gaussian Energy example the weighted sum of squares falls from 112.369 to 112.195 and the parameters move by at most a third of their uncertainties. Some custom-equation fits that previously stopped at the iteration limit now converge, for example a Lorentzian whose linewidth tends to zero. The solver includes a correction to its handling of nearly dependent parameter directions: a difficult exponential starting point now reaches the good fit instead of stopping prematurely with a rank warning. Poor starting values can still lead to different endpoints; convergence alone does not establish a good fit. Fits whose parameters cannot be determined separately still show the rank warning; for some of them the undetermined parameter's uncertainty is now a very large number instead of zero.

2.9.4 — October 2, 2026

Teaching laboratories and guides

• A new page at curve.fit/labs introduces curve.fit for teaching labs. It covers the fitting methods, numerical validation, reports, student workflow, model library and classroom use, and the whole page can be downloaded as a PDF. It also offers example Single and Double PDF reports for the Pearson–York example.
• Help links to the new page from Getting started.
• New guides explain measurement uncertainties, weighted least squares, weighted means, custom equations and Gaussian peaks, with worked examples.
• Public pages show descriptions in search results and when shared; individual reports stay out of search results.

Fit calculations and parameter uncertainties are unchanged.

2.9.3 — September 26, 2026

Plot and table interactions, PDF minus signs

• Negative numbers on PDF plot axes show their minus signs again, for example on residual plots. They had been missing from PDF exports since July 2026. A capital Γ in a title or axis label now appears too.

• Data tables and interactive plots load faster and keep editing, undo, error bars, fit curves and residuals.

• Clicking a plotted observation selects its table row. The toolbar fades in at the top right when you hover over the plot in Setup, Data and Residuals, with Home, Pan, Zoom +, Zoom −, Rectangular Select Zoom and Save (PNG), separated into groups. Point-selection tools and the separate Autoscale button have been removed.

• Row menus support mouse and keyboard navigation, with focus restored after resize or scrolling.

• Pasting beyond the table limits warns when nonblank data is omitted. The pasted portion can be undone in one step.

• PNG downloads capture the view at the moment Download is clicked, even if the plot changes while the image is prepared. Cancelled pans restore the previous view.

• Confidence bands crossing zero remain visible up to the boundary of a logarithmic axis. Very large constant values retain usable axis ranges.

• The Feedback button stays available. It was planned to close after October 1, 2026.

Fit calculations and parameter uncertainties are unchanged.

2.9.2 — September 25, 2026

Walkthrough and Help

• The walkthrough highlights the Absolute/Nominal control and reflects the current layout and behavior.
• Model Evaluator messages such as "Enter an x value." now appear beside its heading instead of moving the panel.
• Plot Labels & Axes and Model Evaluator use matching, slightly tighter heading spacing.
• The Absolute/Nominal information icon now sits before the buttons.
• Help now matches recent changes. The Absolute/Nominal choice stays when you change models, the − and + buttons sit beside the model selector, and Calculate sits beside the Model Evaluator heading. Help also lists more fitting software defaults.

Fit calculations and exports are unchanged.

2.9.1 — September 24, 2026

Compact fitting controls

• Model selection and custom parameter controls share the Model heading row, leaving a full-width equation area below.
• Absolute and Nominal sit beside the Uncertainties heading, with their explanation available from the information icon.
• Calculate sits beside the Model Evaluator heading, with wider input and output fields below.
• Setup and saved-fit panels use the same compact layout, with a little more vertical spacing.
• Custom Function shows only − and + beside the model selector, and switching to or from it no longer shifts the controls.
• Absolute or Nominal stays selected when you change models.
• The Absolute/Nominal explanation covers what each choice means, what it changes and other software's defaults, and fits without scrolling.
• The workspace is ready sooner on slow connections.

Fit calculations, uncertainty conventions and exports are unchanged.

2.9.0 — September 23, 2026

Weighted Mean and Parameter Uncertainties

• Weighted Mean combines measurements in y using the supplied δy uncertainties (y = μ), with a plot, residuals, Browser Report and PDF exports. It supports a single measurement.
• The Uncertainties panel offers Nominal and Absolute. Every model defaults to Nominal, including Weighted Mean. Select Absolute for the usual propagated uncertainty of a weighted average. Saved results and PDF footnotes record the convention.
• Expanded Help explains weighting, parameter uncertainties, confidence bands and the Model Evaluator.

Calculation notes: Absolute retains the unscaled parameter covariance; Nominal retains the previous residual-variance scaling. This choice changes uncertainties and confidence bands, not fitted values or goodness of fit. Model uncertainty bands use Student-t coverage for residual-estimated scale and normal coverage for supplied scale. When there are no residual degrees of freedom, residual-scaled fits now report unavailable parameter and propagated uncertainties instead of zero; a one-point mean can still report its supplied uncertainty.

Custom-equation fits using only x uncertainties now use a uniform positive artificial y uncertainty, as built-in models already did. This avoids failed or degenerate fits when a y value is zero and can change x-only results. Previously saved compatible solutions are reused without refitting.

2.8.4 — September 19, 2026

Current labels in Browser Report

• Browser Report uses the current saved-fit title, labels and display scales for its image and PDF downloads. Previously opened reports keep the settings captured when opened.
• Fit Revision links open the requested Results tab instead of Setup.
• Export notices stay compact without shifting the Results panel. Details provides the full explanation and next steps.
• Unavailable exports and label errors give more specific guidance.

Fit calculations, parameter uncertainties and saved scientific inputs are unchanged.

2.8.3 — September 19, 2026

Updated guided practice

• The walkthrough follows the current Results, Export and Browser Report controls, including where to find fit counts and CSV/JSON downloads.
• Custom-equation lessons explain automatic parameters, equation feedback and choosing Keep Custom when a built-in model is suggested.

Fit calculations are unchanged.

2.8.2 — September 2026

Stylesheet loading correction

• Fixed a stylesheet-loading error that could disrupt the layout in Chrome and other browsers without a cached copy.

Fitted values, uncertainties and PDF contents are unchanged.

2.8.1 — September 2026

Clearer export controls

• Export groups larger Single PDF, Double PDF and Browser Report buttons at the bottom of each saved fit.
• Tabs and buttons share uppercase lettering; fit tabs include the date and time. Numerical values retain their formatting.
• Setup and saved-fit action panels align. Panels use white backgrounds, unavailable evaluator values stay blank, and warning details appear on hover or focus.
• Larger loading indicators and disabled export controls make pending work clearer and prevent repeated clicks.
• The Python template is now a minimal runnable example with sample data and a plot. Help explains the available exports.

Fitted values, uncertainties and PDF contents are unchanged.

2.8.0 — September 2026

Help with custom equations

• Catch common equation errors as you type, with suggested corrections you choose whether to apply.
• Parentheses are paired as you type to help you close them correctly.
• Use pi or π in your equation. When you enter a rounded approximation, you can choose to replace it with the constant.
• Switch to a matching built-in model with your parameters mapped to the corresponding controls, or keep your custom equation. Undo lets you reverse the switch.

Calculation note: Existing equation calculations are unchanged. Applying an equation correction changes the input; switching to a built-in model can change the starting estimates and fitting behavior.

2.7.1 — September 2026

More reliable session copying and report links

These changes reached users in 2.8.0; 2.7.1 was not released separately.

• Fork works again, so you can make a separate working copy of a session. Your unsubmitted data and settings survive a reload until the new session saves its first fit.
• The address bar shows the correct session link after Reset, Fork, or opening an example dataset. The session identifier remains readable when highlighted.
• Open a saved fit's browser report with the clearly labeled Detailed Report action.
• Blank or whitespace-only session names are no longer accepted.

Fitted values and PDF plots are unchanged.

2.7.0 — September 2026

A complete report in your browser

• Open a saved fit in a browser report with the fit and residual plots together, full-precision data tables, and the Model Evaluator.
• Review the report without downloading a file, or download its data and results as CSV or JSON.

Fitted values and PDF plots are unchanged.

2.6.1 — September 2026

Open the exact saved fit from a PDF

• Newly generated PDF reports include a direct link to the fit that produced them, as well as the session link.

Fitted values and plots are unchanged.

2.6.0 — September 2026

Six additional fitting models

• Fit with Shifted Reciprocal, Saturation (Michaelis–Menten), Power Law with Vertical Offset, X Log X, Hill Sigmoid (4 Parameter), and Malus’s Law, without entering a custom equation. The new models include automatic starting estimates.
• Exponential is now named Exponential with Offset to make its added constant clearer.
• Malus’s Law expects angles in radians.

Existing models keep the same calculations and starting-value rules.

2.5.0 — September 2026

Clearer plots and equation feedback

• See more detail in model curves and confidence bands. The Setup preview shows all entered observations.
• Paste common mathematical symbols into custom equations and get clearer feedback about missing parameters and unsupported names.
• Saved-fit numbering, undoing a file import, and recovering an export work more reliably. Keyboard navigation and access to the equation field are improved.
• The Setup tab stays blue even when another tab is selected.

Fitted parameters, existing valid custom-equation calculations, and PDF curve sampling are unchanged.

2.4.1 — September 2026

Consistent layout and corrected PDF exports

• The desktop panel arrangement stays consistent on narrower screens. Below 1440 pixels, scroll horizontally to reach the remaining controls.
• PDF export works after changing a completed fit's title, axis labels, or display scales, including in guided practice.

2.4.0 — September 2026

Edit reports without rerunning a fit

• Change a completed fit's title, axis labels, and linear or logarithmic display scales without fitting again. PDF exports use the settings you select.
• Report edits stay with each fit while the page is open. Edits to the newest fit also carry into Setup unless you have already changed that field there. Reloading restores the original saved report settings.
• Try the optional Walkthrough with sample data, fitting, residuals, model evaluation, and exports. Practice opens in a separate tab and leaves your own work untouched.
• Read data on the left and work with plots and fitting controls on the right. Larger text, clearer table shading, and more consistent tabs and controls improve readability. Switching between saved fits is faster.
• Each saved fit keeps its original data even when you edit, cut, or undo changes in Setup. Residual plots keep a linear vertical scale and a clear zero line.
• Import files are checked before replacing your data. Limits are 1 MiB per file, 10,000 data rows, and 256 characters per cell. Slow exports can be rechecked, and export problems have clearer messages.
• Invalid report labels no longer stop an in-page fit; you are asked to correct them when exporting. Repeated PDF exports are faster, and the Model Evaluator no longer fails in certain otherwise valid cases.
• Custom-equation messages distinguish writing errors from calculation problems such as division by zero.

Export change: The Plot PNG button is removed. Existing PNG links still work; Single PDF, Double PDF, Data & Fit CSV, and Fit JSON remain available.

Calculation note: Numeric input now retains its original precision, so high-precision datasets may give different fitted values or uncertainties. Decimal-comma uncertainty input is also handled more consistently. The updated confidence-band calculation for built-in models can slightly change the band, but does not itself change fitted parameters or their uncertainties. Model Evaluator rounding may differ.

2.3.2 — September 2026

Improved fitting accuracy and clearer uncertainty warnings

These changes reached users in 2.4.0; 2.3.2 was not released separately.

• Fits are checked more closely before they are considered finished. The included NIST reference cases show improved agreement in fitted parameters and uncertainties, including when using their published starting values.
• Fits can identify additional cases where the data cannot determine reliable parameter uncertainties, including a difficult narrow-peak example.
• Long-running fits report timeouts more reliably, without extending the existing calculation time limit.

Calculation note: Fitted values, uncertainties, and warnings can change for any model or uncertainty setting. A small residual score alone does not establish that every parameter is accurately determined.

2.3.1 — September 2026

Browse your full fit history

These changes reached users in 2.4.0; 2.3.1 was not released separately.

• Find every earlier fit in Previous, using Newer and Older to move through long histories. Fit numbers stay consistent, and already-open fits are labeled.
• Fits with warnings are easier to spot. Reduced chi-squared uses the same highlighting in the workspace and reports.
• Reset and Fork work reliably again. Reset clears the workspace; Fork starts a separate session from the selected fit.

Removed: The old Raw Data output and its previously saved links no longer open. Other saved fit links and the existing PDF, PNG, CSV, and JSON outputs remain available. The Model Evaluator is available in the fit panel.

Fit calculations are unchanged from 2.2.8. The documented comparison found identical results for all fourteen reference cases.

2.3.0 — September 2026

Fit results directly in the workspace

These changes were included in the intermediate 2.3.1 update and reached users in 2.4.0; 2.3.0 was not released separately.

• Press Fit! to see your result in a tab beside Setup, with the plot, equation, parameter values and uncertainties, fit statistics, and warnings. The three most recent fits stay open; earlier fits are available through Previous.
• Each fit shows the data it used, its fitted curve and 68.27% confidence band, and a switch between the fit and residual plots. Saved-fit data are read-only; later edits in Setup do not change earlier results.
• Use the Model Evaluator to calculate a model value and its uncertainty at an x value you choose.
• Choose exports after fitting rather than before: Single PDF, Double PDF, Plot PNG, Data & Fit CSV, and Fit JSON. Files are prepared when requested.
• Reopen a session in another browser with its most recent data and three latest fits, keeping their original fit numbers.
• Custom parameter buttons work again, blank rows before your data are accepted, and Reset clears logarithmic display settings correctly. Fit requests start faster, with a countdown when you need to wait before fitting again.
• Plot size and positioning stay consistent between tabs. Axis labels remain in Setup and downloaded reports rather than on the in-page fit plot.

Removed: Raw Data is no longer offered as an output choice. Its saved links were also retired in 2.3.1.

Fit calculations are unchanged from 2.2.8. The documented comparison found identical results for all fourteen reference cases.

2.2.8 — September 2026

Help and template buttons work again

• TEMPLATE and HELP respond to clicks again; the feedback control no longer blocks them.

Fit calculations are unchanged.

2.2.7 — September 2026

Compare a fully specified model with your data

• Hold every parameter fixed to plot the model exactly as entered. Results include residuals and fit statistics, with a clear “All parameters fixed: model plotted, not fitted” notice. Parameter uncertainties are shown as zero and no confidence band is displayed.
• Corrected overstated uncertainties in some fits with fixed parameters. Fits with no fixed parameters are unaffected.
• Press Enter while Fit! is focused to start a fit. Help includes screen-size guidance, and the feedback button is easier to use.
• Expanded checks of fitting behavior and uncertainty estimates; see Numerical validation.

Calculation note: Fitted parameter values are unchanged. Uncertainties can change only in the affected fixed-parameter cases; fully fixed models now produce a defined result.

Which fixed-parameter results are affected?

The correction applies when at least one parameter is fixed and a data point has zero sensitivity to every free parameter. Examples include x = 0 with a fixed intercept, or points far from a Gaussian peak with a fixed offset. Uncertainty scaling now uses the number of observations minus the number of free parameters.

2.2.6 — September 2026

Fewer misleading fit warnings

• Corrected false “Not full rank” warnings for exact fits and for fits with parameters on very different scales, including the NIST Pontius example.
• Data with constant y values can be fitted. R² is shown as n/a when it is undefined rather than causing the fit to fail.
• Corrected oscillator confidence bands when the displayed phase is adjusted across ±π, and fixed the feedback button's appearance.

Calculation note: Fitted parameter values and their uncertainties are unchanged. Warning messages and R² displays are corrected in the cases above; affected oscillator confidence bands can change.

2.2.5 — September 2026

Easier file import and feedback

• Import comma-, tab-, or semicolon-separated data, skip blank lines, and import the same file again when needed. A file with no numeric rows now produces a clear message.
• Find FEEDBACK beside TEMPLATE and HELP during the feedback period.

Fit calculations and file-column mapping are unchanged.

2.2.4 — September 2026

Clearer warnings and corrected uncertainty bands

• Results show all relevant fitting warnings, rather than only the first. A parameter uncertainty that cannot be determined is shown as n/a, not zero. Warnings are also included in JSON downloads.
• Corrected confidence bands and Model Evaluator uncertainties for Gaussian and Lorentzian fits whose width was adjusted from negative to positive for display.

Calculation note: Fitted parameter values and their individual uncertainties are unchanged. Confidence bands and Model Evaluator uncertainties can change in the affected width-correction cases.

2.2.3 — September 2026

More reliable results for very small coefficients

• Corrected fits with extremely small automatically estimated coefficients, including the NIST Pontius example. These cases could previously show an incorrect warning or an uncertainty that was too small.
• Failed calculations now show an explanation instead of returning results containing nan. Input uncertainties must be finite and positive; percentage and proportional uncertainties use the magnitude of the data value.
• Added this changelog, the footer version link, and Help's Numerical validation section.

Calculation note: Fits affected by the very-small-coefficient issue can change; the certified NIST Pontius values are reproduced. Fits that previously returned non-numeric results now fail with an explanation. Other fits are unchanged.

2.2.2 — September 5, 2026

Clearer uncertainty controls

• Shaded δx and δy columns show which uncertainty values are currently unused. You can still edit them; they become active when you select Table for that uncertainty.
• Column shading updates correctly when you switch settings.

Fit calculations are unchanged.

2.2.1 — September 4, 2026

Corrected unexpected zero uncertainties

• Corrected some fits that reported zero uncertainty for a free parameter when its automatic starting estimate was extremely close to zero.
• The fix applies across all four uncertainty settings: none, x only, y only, and both x and y. Starting values are unchanged.

Calculation note: Affected fitted values and uncertainties may change. Recheck earlier fits with unexpectedly zero uncertainty for a free parameter. This does not resolve models that the data cannot adequately constrain. The correction was refined in 2.2.3.

2.2.0 — September 4, 2026

Updated built-in fitting calculations

• Updated calculations for all 19 built-in models, using derivatives obtained directly from their equations.
• Corrected x-only uncertainty handling that could incorrectly show a free parameter as undetermined with zero uncertainty.

Calculation note: Rerunning built-in models can change fitted values, uncertainties, and whether a fit converges. The custom-equation fitting method is unchanged.

Validation results and limitations

The derivatives were checked independently, and built-in models were tested with synthetic data using x-only and combined x/y uncertainties.

For the Gaussian Energy reference case, fitted parameters changed by less than 0.5% and reported uncertainties by less than 4% compared with the previous implementation. Those observations apply to that test case, not to every dataset.

Some Lorentzian, exponential, and other poorly constrained fits can still give unreliable parameter uncertainties. A successful fit or smaller residual does not establish that every parameter is well determined. This release did not yet provide all the warnings for those cases.

2.1.0 — September 1, 2026

More reliable custom-equation fitting

• Some valid custom equations that were previously rejected can now be fitted successfully.
• Plots skip regions where a custom equation is undefined instead of failing the entire result.
• Added a temporary feedback form for reporting issues.

Calculation note: Previously rejected custom fits may now succeed, but results must still be finite and real at the observed data points. Built-in calculations are unchanged.

2.0.0 — August 2026

An updated curve.fit

• Introduced the updated application while retaining the familiar model choices, uncertainty controls, starting and fixed parameters, and the PDF and Raw Data outputs available at the time.
• Pages and plots load more quickly. Reset clears old points and error bars, and Raw Data tables stay within the report page.
• Added NIST Norris, Pontius, Misra1a, Eckerle4, and DanWood datasets, alongside the Pearson–York benchmark and Gaussian examples.

Calculation note: All nine reference cases passed the documented comparison with the previous application within the specified tolerances. This does not guarantee identical results for every dataset.

1.0.0 — Legacy version

The original curve.fit application, before the version 2 transition. Earlier changes are not reconstructed here.