The PDF gives linked page references; digital editions link directly to the named discussion.
Absolute value: Cosine similarity.
Accumulator: Sums over blocks.
Anisotropy: Embeddings.
Assignment: Kuhn–Munkres.
Attention energy: Dominant axis, spectrum, profile.
Basis: Transpose, symmetry, orthogonality.
Block: Sums over blocks.
Canon: Episodes: threading across days.
Centring: Centering and covariance.
Centroid: T² and Q.
Connected component: Stories: single linkage on a day’s articles.
Coordinate: A matrix times a vector.
Corpus: Counting words.
Correlation: Kuhn–Munkres.
Cosine similarity: Cosine similarity.
Covariance: Centering and covariance.
Covariance matrix: Centering and covariance.
Damping: TF‑IDF.
Dense storage: TF‑IDF.
Derivative: Truncation.
Diagonal matrix: What the SVD is.
Dimension: Transpose, symmetry, orthogonality.
Document frequency: TF‑IDF.
Dominant axis: Dominant axis, spectrum, profile.
Dot product: Cosine similarity.
Eigengap: Why nightly refinement is stable.
Eigenspace: Two ambiguities.
Eigenvalue: Eigenvectors: the axes of the cloud.
Eigenvector: Eigenvectors: the axes of the cloud.
Embedding: Embeddings.
Energy: Energy × (1 + ν) for ranking.
Episode: Episodes: threading across days.
Euclidean norm: Cosine similarity.
Explained variance: How many axes to keep.
Fit weight: Weights that balance the days.
Frobenius norm: Truncation.
Function: Truncation.
Gaussian distribution: ±2σ on the spectrum.
Gram matrix: A matrix times a matrix.
Gram–Schmidt: Latent semantic analysis.
Graph: Stories: single linkage on a day’s articles.
Greedy matching: Kuhn–Munkres.
Identity matrix: Transpose, symmetry, orthogonality.
Inverse: T² and Q.
Latent semantic analysis: Latent semantic analysis.
Least squares: Truncation.
Linear combination: A matrix times a vector.
Loading: Varimax.
Logarithm: TF‑IDF.
Marginal scaling: T² and Q.
Matrix: A matrix times a vector.
Mean: Centering and covariance.
Moment: Orienting an axis from power sums.
Naming rotation: Varimax.
Norm: Cosine similarity.
Novelty ratio: T² and Q.
Operator norm: Why nightly refinement is stable.
Orthogonal: Transpose, symmetry, orthogonality.
Orthonormal: Transpose, symmetry, orthogonality.
Orthonormalization: Latent semantic analysis.
Outer product: Centering and covariance.
Overnight matching: 0.8 for matching axes across nights.
Oversampling: Latent semantic analysis.
Percentile: Stories: single linkage on a day’s articles.
Perturbation: Why nightly refinement is stable.
Positive semidefinite: Eigenvectors: the axes of the cloud.
Power iteration: Latent semantic analysis.
Power sum: Orienting an axis from power sums.
Precedent: Precedents.
Precision: Episodes: threading across days.
Principal component: Eigenvectors: the axes of the cloud.
Procrustes alignment: Why nightly refinement is stable.
Profile: Dominant axis, spectrum, profile.
Projection: Projection, reconstruction, residual.
Quadratic form: T² and Q.
Randomized SVD: Latent semantic analysis.
Range: Latent semantic analysis.
Rank: Truncation.
Recall: Episodes: threading across days.
Reconstruction: Projection, reconstruction, residual.
Relative error: Truncation.
Residual: Projection, reconstruction, residual.
Rotation invariance: T² and Q.
Score: Projection, reconstruction, residual.
Shape: A matrix times a matrix.
Single linkage: Stories: single linkage on a day’s articles.
Singular value: What the SVD is.
Singular value decomposition: What the SVD is.
Sketch: Latent semantic analysis.
Skewness: Two ambiguities.
Sparse storage: TF‑IDF.
Spectrum: Dominant axis, spectrum, profile.
Standard deviation: ±2σ on the spectrum.
Standardization: ±2σ on the spectrum.
Story: Stories: single linkage on a day’s articles.
Streaming: Sums over blocks.
Subspace: Why a newspaper needs linear algebra.
SVD: What the SVD is.
Term frequency: TF‑IDF.
TF–IDF: TF‑IDF.
Threshold: The clustering thresholds.
Trace: How many axes to keep.
Transpose: Transpose, symmetry, orthogonality.
Truncation: Truncation.
T² statistic: T² and Q.
Uncorrelated: Centering and covariance.
Unit vector: Cosine similarity.
Variance: Centering and covariance.
Varimax: Varimax.
Vector: Why a newspaper needs linear algebra.
Weighted average: Weights that balance the days.
Welford update: ±2σ on the spectrum.
Whitening: Varimax.
Z-score: ±2σ on the spectrum.