A continuous run of typing with no pause longer than 2 seconds. Each burst roughly maps to one "thought unit." Longer bursts = more fluent, ideas flowing. Short bursts = more fragmented, stop-and-think processing.
Chenoweth & Hayes, 2001
IKI (Inter-Key Interval)
Milliseconds between consecutive keystrokes, capped at 5s. The mean captures average hesitation — higher = more deliberation per keystroke. The std dev captures consistency — high means you alternate between rapid and slow typing.
Epp et al., 2011 (CHI)
Commitment Ratio
Final character count divided by total characters typed (including deleted). Low ratio = heavy self-editing. A ratio of 1.0 means you kept everything you typed.
Large vs Small Deletions
Small deletions (<10 chars) are corrections — typos, word swaps. Large deletions (≥10 chars) are revisions — rethinking what you meant. The split between first-half and second-half deletion chars shows whether revision happens early (planning) or late (restructuring).
Faigley & Witte, 1981
Revision Chains
Sequential deletion keystrokes within 500ms of each other count as one chain. More chains = more revision episodes. Longer chains = deeper per-episode rethinking.
Leijten & Van Waes, 2013
Hold Time / Flight Time
Hold time = how long a key is pressed (keydown to keyup). Measures motor execution — tremor, rigidity, finger coordination. Flight time = gap between releasing one key and pressing the next. Measures cognitive planning — word retrieval, sentence planning. These are independent signals collapsed into IKI by most tools.
Kim et al., 2024 (JMIR)
Keystroke Entropy
Shannon entropy of the inter-key interval distribution. Measures how unpredictable your typing rhythm is. High entropy = irregular, switching between fast bursts and long pauses. Low entropy = metronomic, consistent rhythm. Different from IKI std — entropy captures the shape of the distribution, not just its spread. Correlated with executive function at d=-1.28.
Ajilore et al., 2025 (BiAffect/Frontiers in Psychiatry)
Permutation Entropy
Ordinal pattern distribution of consecutive IKI triplets. Instead of looking at timing values, looks at their rank order (is this interval faster, slower, or middle?). High = all patterns equally likely, genuinely novel composition. Low = habitual rhythm, cognitive autopilot. Invariant to device differences and caffeine.
Bandt & Pompe, 2002
DFA Alpha (Fractal Scaling)
Detrended Fluctuation Analysis exponent. Measures whether your keystroke timing has fractal structure. ~0.5 = white noise (no structure, copying). ~0.7-0.9 = pink/1/f noise (long-range correlations, healthy engaged cognition). ~1.0+ = brown noise (over-correlated, perseveration).
Peng et al., 1994; Van Orden et al., 2003
RQA (Recurrence Quantification)
Treats the IKI series as a trajectory through cognitive state space. Determinism = how predictable the trajectory is. Laminarity = how often the system gets stuck in a state (fixation). Trapping time = how long stuck states last. High determinism + high laminarity = deep structured processing.
Webber & Zbilut, 2005
Transfer Entropy
Causal information flow between hold times (motor) and flight times (cognitive). TE hold-to-flight = motor driving cognition (automatic writing, fingers know before mind). TE flight-to-hold = cognition driving motor (deliberate writing, each keystroke decided). The dominance ratio shows which direction controls.
Schreiber, 2000
Mouse/Cursor Trajectory
Cursor movement sampled every 200ms during typing pauses (>2s). Cursor distance = total px moved while not typing. Fidget ratio = distance per ms of active typing (involuntary motor restlessness). Stillness = proportion of pause samples with <5px movement. Drift-to-submit = how many times the cursor entered the submit button during a pause then left (hesitation-to-commit signal).
BioCatch (Unit 8200 cognitive biometrics)
Error Correction
Three-phase model: deletion execution speed = mean IKI within deletion chains (how fast you delete once you start). Postcorrection latency = ms from last delete to next insert (cognitive re-planning after an error). Separates motor automaticity from the thinking required to recover.
Springer, 2021
Revision Distance
How far back from the leading edge (end of text) the cursor is when a deletion happens. Mean revision distance captures typical reach-back depth. Max revision distance captures the deepest restructuring event. Deep revisions indicate rethinking structure, not fixing typos.
Lindgren & Sullivan, 2006 (ScriptLog)
Punctuation Latency
Flight time (key release to next key press) before punctuation keys vs. letter keys. Punctuation requires syntactic decisions (where to end a clause, whether to use a comma). The ratio captures how much harder punctuation placement is relative to word production. Higher ratio = more syntactic planning load.
Plank, 2016 (COLING)
Linguistic Densities
Word category counts divided by total words. NRC emotion categories (anger, fear, joy, sadness, trust, anticipation) come from the NRC Emotion Lexicon. Cognitive, hedging, and first-person densities come from Pennebaker's LIWC categories. Slopes of these over time are the real signal.
Mohammad & Turney, 2013; Pennebaker, 2011
Session Delta
Journal value minus calibration value across 7 dimensions. Isolates what the reflective question provoked beyond your neutral writing behavior. Same person, same keyboard, same day — any shift came from the prompt, not the typist. In production, these get z-normalized against your personal history after 10+ sessions.