Add Fuzziness to Text

Text fuzziness is the deliberate introduction of controlled noise into a string. Applications range from adversarial testing of NLP pipelines and spam filter evasion analysis to data augmentation for training robust OCR and text-classification models. A misapplied fuzziness level can render test data useless or produce artifacts that don't reflect real-world corruption patterns. This tool applies four distinct distortion algorithms - homoglyph substitution using Unicode confusables from Latin, Cyrillic, and Greek blocks, stochastic typo injection based on a QWERTY adjacency matrix, Zalgo diacritical stacking via combining characters in the range U+0300 - 036F, and weighted leetspeak mapping - each governed by a probability parameter p ∈ [0, 1]. Note: homoglyph output may appear identical to the original on certain fonts but differs at the codepoint level. Results depend on the rendering engine and typeface.

Each character in the input string is independently subjected to a distortion decision. The probability of any character being altered is governed by the fuzziness parameter:

where f ∈ [0, 1] is the fuzziness intensity from the slider, and w_mode is the mode-specific weight (homoglyph: 0.9, typo: 0.4, Zalgo: 0.7, leet: 0.8). For each character c_i, a uniform random value r ∈ [0, 1) is drawn:

For Zalgo mode, the number of combining marks n stacked on each affected character scales linearly with intensity:

where f at maximum yields up to 16 combining characters per base glyph. For typo mode, the error type is selected uniformly from the set {swap, duplicate, omit, neighbor}, with neighbor-key substitution using a precomputed QWERTY adjacency lookup of 26 × ~4.5 average neighbors per key.