draft-mode

turn polished ai text into realistic, human-generated drafts.

this tool introduces "biomechanical" typing errors into clean text. unlike other tools that add random noise (e.g., swapping 'z' with 'm'), draft-mode simulates the physical constraints of typing on a qwerty keyboard

it is perfect for making ai-generated feedback, engineering notes, or student look authentic.

features

• biomechanical error logic:

  • fat finger (45%): hits keys physically adjacent on a qwerty keyboard (e.g., 'r' instead of 'e').
  • transposition (25%): swaps sequential letters due to fast typing (e.g., "locig" vs "logic").
  • omission (20%): simulates weak key presses (skips letters).
  • double tap (10%): simulates key bounce.

• web interface:

  • dark/light mode (auto-detects system pref, remembers choice).
  • adjustable error percentage (0% to 100%).
  • "force lowercase" toggle for that casual chat aesthetic.
  • one-click copy to clipboard.

• cli support: includes a python script for command-line processing.

live demo

[https://antonl-dev.github.io/draft-mode]

---

usage (web)

1. clone the repo:

   git clone https://github.com/antonl-dev/draft-mode.git

2. open index.html in any web browser.
3. paste your text, adjust the percentage (3% is recommended for realistic typos), and click humanize.

---

usage (python cli)

you can use the python script for automation or batch processing.

# basic usage (defaults to 5% error rate)
python3 type-error-adder.py --text "the simulation environment is unstable"

# specify error rate (0.0 to 1.0)
python3 type-error-adder.py --text "Debugging the controller logic" --rate 0.03

# process a text file
python3 type-error-adder.py --file my_draft.txt

how it works

the algorithm iterates through the input string and rolls a die based on your error rate. if an error occurs, it selects a type based on realistic human probabilities:

1. neighbor map: the script contains a full adjacency map of the us qwerty layout. it knows that 'g' is surrounded by 'r, t, y, f, h, v, b'.

2. case handling: by default, it forces the text to lowercase to simulate casual, rapid typing, but this can be toggled off in the web interface.

license

MIT License