Limitations and accuracy#
GraphX has no universal accuracy figure. The uncertainty of a digitised value depends on the source image, the calibration, and the placement or trace decision.
Source and calibration quality#
Resolution, JPEG compression, blur, scans, anti-aliasing, line thickness, marker size, and photographed-page distortion all affect what can be read from a graph. Use clear, widely separated calibration references where possible: a fixed placement error represents a smaller fraction of a longer calibrated span. Choose Perspective calibration for planar perspective, but not as a cure for lens distortion or a warped page. Identify logarithmic axes correctly before calibrating.
Points, tracing, and displayed digits#
Manual placement is limited by the source pixels and the interpretation of a line or marker. The Precision magnifier helps inspect those pixels; it cannot create information that was absent from the image. Curve tracing can be fast and useful, but its colour-guided path can be ambiguous near intersections, similar colours, grid lines, markers, gaps, and low-contrast features. Inspect its preview before accepting it; see Curve-tracing method.
Displayed digits are not measurement accuracy. Dataset Precision changes formatting, not the certainty of the calibration or source reading. Six displayed significant figures do not mean the source graph supports six physically meaningful figures.
Extrapolation and regression#
Calibration transforms and trend models can be mathematically evaluated outside their calibrated or fitted regions. That does not make such values reliable. Calibration extrapolation concerns a point outside the reference plot bounds; regression extrapolation concerns a model drawn outside the data used to fit it. See Fit and display ranges.
A fitted equation models digitised points, not necessarily the physical process. Its accuracy cannot exceed the quality of the digitised data, and a high does not establish scientific validity. See Understanding R².
Practical checks#
Calibrate from clear, well-separated references; inspect calibration geometry; zoom or magnify when placing references and points; compare extracted points with the source; review trace previews; avoid claiming more precision than the source supports; and treat extrapolation cautiously.