What Alice measures and how.

Complexity of the IKI series at multiple temporal scales (orders 3-7). High PE = unpredictable rhythm. The spectrum separates local complexity from global structure.

Amplitude-weighted PE that captures both ordinal pattern frequency and the magnitude of fluctuations. More sensitive to large IKI excursions than standard PE.

Jensen-Shannon complexity: the product of entropy and disequilibrium. Distinguishes structured complexity from randomness. Maximum at the edge between order and disorder.

Proportion of ordinal patterns that never occur. Deterministic systems forbid patterns; stochastic systems explore them all. Sensitive to low-dimensional structure in the IKI series.

Algorithmic complexity of the symbolized IKI series via LZ76 parsing. Measures compressibility of the temporal sequence. Complements PE by capturing non-ordinal structure.

Detrended fluctuation analysis. Self-affine scaling exponent of the IKI series. Alpha near 0.5 = white noise (no temporal memory). Alpha > 1 = long-range correlations (persistent rhythm).

Width of the multifractal singularity spectrum. Measures richness of scaling behavior. Wide spectrum = multiple coexisting temporal dynamics. The most ghost-resistant behavioral dimension by 4x: the reconstruction adversary cannot reproduce multifractal structure from a single stochastic process.

Asymmetry of the singularity spectrum (left- vs right-skewed multifractal structure) and the Holder exponent at the spectrum peak. Together they characterize whether large or small fluctuations dominate the multifractal signature.

Statistical asymmetry under time reversal. Nonzero irreversibility implies the IKI series is generated by a dissipative, non-equilibrium process rather than a passive noise source.

Five spectral features from Welch PSD of the IKI series: spectral slope (1/f exponent), respiratory-band peak frequency, peak typing frequency, low/high frequency ratio, and fast/slow variance ratio. Spectral slope distinguishes random (flat) from self-organized (steep) timing. The respiratory peak detects physiological entrainment of typing rhythm.

Entropy of the transition matrix between successive ordinal patterns, and the count of transitions that never occur. Captures temporal dependencies that standard PE discards by treating patterns as independent.

Six metrics from reconstructed phase-space attractor: determinism (repeatable patterns), laminarity (laminar states), trapping time (average diagonal length), recurrence rate (overall self-similarity), recurrence time entropy, and mean recurrence time.

Graph-theoretic analysis of the recurrence matrix: transitivity (clustering of recurrent states), average path length (separation between attractor regions), clustering coefficient, and assortativity (degree correlation). Reveals geometric structure of the phase-space attractor that RQA diagonal/vertical line statistics miss.

Effective information (EI) quantifies determinism and degeneracy of temporal transitions. The causal emergence index (CEI) measures whether coarse-graining the IKI series increases its causal structure. Optimal causal scale identifies the temporal resolution at which the system is most informative about itself.

Partial information decomposition of hold and flight time contributions to IKI prediction. Synergy = information available only when both channels are observed together. Redundancy = information available from either channel alone. High synergy suggests integrated motor-cognitive control.

Branching ratio measures propagation of activity (IKI exceedances). Ratio near 1.0 = critical dynamics (self-organized). Avalanche size exponent characterizes the power-law distribution of burst cascades. Together they test whether the keystroke process operates near a critical point.

Data-driven decomposition of the IKI series into spatial-temporal modes: dominant oscillatory frequency, decay rate (damping vs growth), active mode count, and modal spectral entropy. Extracts the dominant rhythmic components without assuming stationarity.

IKI distribution decomposed into Gaussian mixture components via BIC-selected model order. Component count reflects the number of distinct timing regimes. Motor proportion identifies the fast (automatic) component. Cognitive load index measures the relative weight of slow (deliberative) components.

Information flow between hold times and flight times using the KSG estimator. Measures causal coupling between motor execution (hold) and cognitive planning (flight). Dominance indicates which channel drives the other.

IKI distribution decomposed into Gaussian (mu, sigma) and exponential tail (tau) via MLE/EM. Tau captures attentional lapses. Tau proportion = tau / (mu + tau).

Regularity measure of the IKI series. Low = rhythmic, metronomic. High = chaotic, irregular. More robust than approximate entropy for short series.

Third derivative of the IKI position series. Measures smoothness of the temporal trajectory. High jerk = abrupt rhythm changes.

Proportion of IKIs exceeding mu + 3 sigma. Captures attentional lapses as statistical outliers in the timing distribution.

Linear trend in IKI over session duration. Negative = speeding up (warming up). Positive = slowing down (fatigue).

Hold and flight times per bigram (key pair). Biometrically stable across sessions. Used for motor fingerprint stability tracking.

Lempel-Ziv complexity of the quantized IKI series. Low = repetitive rhythm. High = complex temporal structure.

Correlation between consecutive hold times. Captures finger coordination and motor planning continuity.

Sample entropy computed at multiple coarse-graining scales. The complexity index (area under the MSE curve) quantifies overall multi-scale regularity. Healthy systems show high complexity across scales; degraded systems lose complexity at coarse scales first.

Trace of the Fisher information matrix at the MLE/EM parameter estimates. Measures estimation precision: high trace = tight confidence region around (mu, sigma, tau). Low trace = flat likelihood, unreliable parameter estimates. Provides a per-session quality gate for ex-Gaussian parameters.

Spearman rank correlation between hold duration and subsequent flight time. Measures the monotonic coupling between motor execution and transition planning, robust to distributional outliers.

Pauses categorized by location: within-word (lexical retrieval), between-word (local planning), between-sentence (high-level planning). Distribution reveals cognitive load allocation.

Revision bursts (ending in deletion) vs inscription bursts (ending in new text). High R-burst ratio = revision-heavy process. Captures writing strategy independent of output quality.

Deletions of entire thoughts (heuristic: large deletions at paragraph boundaries). Genuine cognitive revision, not typo correction.

Normalized session time when deletions overtake insertions. Early transition = early commitment. Late = extended exploration before settling.

Detected changes in typing style (IKI distribution shift) within a session. Multiple shifts suggest exploration across approaches.

Propositions per word. The original Nun Study biomarker: low idea density in early writing predicted Alzheimer's decades later.

Inverse mean word frequency rank. Higher = rarer vocabulary. Proxy for active vocabulary breadth.

Booster / (booster + hedge) ratio. High = assertive, certain. Low = tentative, hedging. Captures confidence independent of content.

Contrastive + integrative connective density. Higher = more differentiation and integration of competing perspectives.

Causal, temporal, and intentional connective density. Measures how explicitly the text links ideas across sentences.

Gzip compression ratio as Kolmogorov complexity proxy. High = more information per character. Low = repetitive or formulaic.

Average pairwise semantic similarity across all sentence pairs in the response. Measures overall thematic unity: high global coherence = a tightly focused response. Low = scattered or multi-topic.

Average semantic similarity between adjacent sentence pairs. Measures thought-to-thought continuity: high local coherence = smooth transitions. Low = abrupt topic jumps.

Ratio of global to local coherence. High = thematically tight but with some local jumps (structured argument). Low = locally smooth but globally diffuse (stream of consciousness).

Rate at which pairwise semantic similarity decays with sentence distance. Steep decay = ideas are locally clustered. Flat decay = thematic connections span the full response.

Character trigram model trained on all prior entries, scored against today's text. Higher = more novel language relative to personal history.

Normalized compression distance between today's text and entries at various time lags. Measures semantic distance over time. High NCD = divergent content.

Cosine similarity of today's digraph latency profile to rolling baseline. Motor fingerprint stability. Low = something changed physically.

Co-occurrence graph density from sliding window analysis. Higher density = more interconnected concept network in the response.

Exponential decay rate of Jaccard vocabulary similarity across time lags. Fast decay = rapidly diversifying vocabulary. Slow = stable lexicon.

Character trigram model trained on quantized IKI history, scored against today's motor timing. The motor analogue of textual self-perplexity. Higher = today's typing rhythm is more novel relative to motor history.

Number of distinct communities detected in the word co-occurrence graph via modularity-based clustering. More communities = more distinct conceptual clusters in the response.

Proportion of edges in the co-occurrence graph that connect different communities. High bridging = integrative writing that links distinct conceptual clusters. Low = compartmentalized thinking.

DFA alpha (long-range correlation) + P-burst length (sustained production) + tempo drift. Captures whether the writing flowed or fragmented.

Permutation entropy + first keystroke latency + between-sentence pause density. High = effortful, exploratory thinking before and during writing.

Abandoned thought count + R-burst/I-burst ratio + inverse commitment. How much the writer restructured after committing text.

Final/typed ratio + phase transition point + inverse strategy shift count. Whether the writer stuck with a direction or explored multiple approaches.

Euclidean distance from previous entry in 4D subspace. How behaviorally different this session was from the last.

Ex-Gaussian tau (attentional lapses) + lapse rate + correction rate. Cognitive-motor disruption, editing heat.

RQA determinism + transfer entropy dominance + inverse tab-away rate. Sustained coherent engagement with the writing task.