Overview | Two layers, one contract | Detection, and the value of refusing to guess | Segmentation | Sentences and n-grams | Sorting | Width, and why your tables are ragged | Normalisation and transliteration | What tidycjk is built on | References

Four conversions, four different levels of trust | Hangul jamo: exact | Kana: safe, with two things to know | Simplified and traditional Han: right on characters, blind to vocabulary | Romanisation: depends entirely on the language | Width forms are a separate concern | Normalisation is the fourth axis | References

The problem | Choosing an engine | "icu": dictionary-based, and already installed | "character": the honest baseline | Why there is no default | locale does not pick the dictionary | In a pipeline | Registering another engine | gibasa, for Japanese | jiebaR, for Chinese | A caution about argument matching | Comparing engines | References

One character is not one column | cjk_pad() | cjk_truncate() | cjk_wrap() | Why this is not "split every N columns" | One rule that is not universal | Indents | The shared contract | Ambiguous width | References

Credentials are never arguments | Iceberg REST catalog | With OAuth2 client credentials | Extra, non-secret properties | Object storage | S3-compatible storage (MinIO, R2, Ceph) | AWS Glue | Local warehouses | Performance notes | Troubleshooting | References

The three levels | File pruning, measured | Projection pushdown | Which predicates push down | Three things that do not push down | Case sensitivity | Reading a plan for cost, not just count | References

Snapshot history | Reading a snapshot by id | Reading as of a time | The schema travels too | Isolation, and why a handle looks stale | What is not here | References

What survives unchanged | The five that need care | NaN comes back as NA | factor comes back as character | long needs bit64 | timestamp_ns loses sub-microsecond precision | Snapshot ids are character | Nested types | Reading the schema rather than guessing it | Why the bridge is an Arrow bridge | References

A namespace, then a table | Appending | Compression | Commits are atomic, and optimistic | Registering a table someone else wrote | What is refused, and why that is the design | Checking before you write | References

Proportional-symbol (bubble) maps | Spike maps | Equal-area tile grids | Flow maps | Small multiples | Labels | Maps that need optional packages | Value-by-alpha: the cartogram's non-distorting cousin | Country adjacency and distance

Abstract | Introduction | Core data assembly | world_data() | country_data() and attach_geometry() | Visualising: the choropleth and beyond | One-line choropleths | Proportional-symbol maps | Equal-area tile grids | Flow maps | The join engine | Diagnostics: never lose a country silently | Reference data and code translation | Analysis helpers | Inequality, correlation and convergence | Performance and offline use | Conclusion | Session information

Every column knows its source | Fetching with the record attached | Pinning a release | Your own source, on the same footing | Two sources, one indicator | What the record holds

A map-ready tibble in one call | Your first choropleth | Choosing indicators | Next steps

1. The classification is doing the talking | 2. Grey is not a value | 3. Small denominators shout | 4. The projection is doing the talking too | 5. Say what the map is | 6. Where the data comes from, and when | 7. Islands are not missing at random | References

Standardise any frame | One call to a map | Reconcile two messy tables | Reconcile many tables at once | Check before you trust | Historical data: dissolved countries | Repair what can be repaired | Custom origins | Point data onto the spine

An equal-area, projected choropleth | The world as a globe | Just the canvas | Recentring and the antimeridian | Region subsetting | Simplifying for the web

A small panel | Joining two panels | Membership as of each row | Classifications as they were | The region that moved | Lags keyed on the year | Aggregates that know what they are missing

Emit a query (no dependencies) | Render in the database | In a Quarto / R Markdown document | Why bother?

Why animate? | How geom_taichi() composes with gganimate | Animating over a third variable | Keeping glyphs round | Spin animation | Export helpers

Abstract | Introduction | The changepoint problem | Penalised cost minimisation | Search-based and multiscale methods | Beyond the mean | Design of the package | The ggcpt result contract | Tidy methods and the plotting layer | Design principles | The unified dispatcher | A tour of the method families | Penalised and optimal partitioning | Multiscale and search methods | Changes in slope | Bayesian detection | Nonparametric and sequential detection | Robustness to drift, autocorrelation, and model ambiguity | Multivariate and high-dimensional detection | Structural breaks in regression | Beyond detection | Batch detection over many series | Stability diagnostics | Evaluation, interactivity, and citations | Discussion | Acknowledgements | References

Abstract | Introduction | Problem setup | Visual comparison | Accuracy metrics | Multi-annotator evaluation | Visual evaluation | Uncertainty beyond point sets | Engine-native confidence intervals | Bootstrap stability | Penalty-path sensitivity: CROPS | A benchmarking workflow | Discussion | References

as_ggcpt(): changepoints in, result out | cpt_register_method(): dispatch by name | Returning a finished result | Recipe: a Python detector through reticulate | Reducing the install | What the contract is | References

1. Where could the changepoint be? | Four provenances, one contract | A changepoint that is an interval | 2. Is the change real? | 3. How many changepoints? | 4. What is driving the answer? | Which observation | Which setting | What the detector computed | Putting it together | References

1. A monitor, one batch at a time | 2. Scoring: delay, not location | 3. Replaying a finished series | 4. The three methods | edetector: the default, governed by alpha | cpm: governed by arl0 | alpha and arl0 are alternative parameterisations, not two knobs | ocd: multivariate only | 5. Delay is a function of shift size | 6. Assumptions: the thresholds are for independent observations | 7. What to report | References

Labels | Scoring a segmentation | The label error curve | Learning the penalty | Where this fits | References

Abstract | The result object and its methods | Unified detection | Engine wrappers | The classical core (0.1.0) | The search and pruning wave (0.2.0) | The engine wave (0.4.0) | The penalty path | The visualisation layer | Comparing methods | Batch detection and stability | Evaluation against ground truth | Simulation and canonical signals | Citing the methodology | The 0.5.0 surface | Time indices | Inference | Selecting the number of changepoints | Diagnostics | Supervised detection | Choosing and combining methods | Communication | Benchmarking | Streaming | Power and study design | Extending | Closing note | References

Why taichi? | Reading a single symbol | The example data | A first taichi grid | Fewer cells, bigger glyphs | Which source should be yin? | Customizing the color scales | Removing the panel padding | Comparing places with facets | Theming | The glyph's other channels | Rotation | Data-driven eyes | Categorical fills | Missing values | Geom parameter routing | Shared limits and a single legend | The fish geoms are exported | Faster rendering | New in v0.3.0 | How much bigger? The explicit channel | The same numbers, as a table and as a heatmap | Palette pairing is a correctness problem | Binned fills, for grids too dense to read | Hovering for the exact values | Following the theme | When (not) to use taichi | Acknowledgement

Overview | Example data | Detecting and summarising emoji | emoji_summary() | emoji_filter() | Extracting emoji | emoji_extract_nest() | emoji_extract_unnest() | emoji_tokens() | A note on grapheme-aware detection | Counting emoji across the corpus | emoji_frequency() | top_n_emojis() | Categorising emoji | emoji_type(): faces versus everything else | Scoring emoji sentiment | emoji_sentiment() | Sentiment distribution | Sentiment by category | The sentiment lexicon | Interpretation risk: how much do readers disagree? | Scoring emoji emotions | Bringing your own lexicon | Relating emoji: co-occurrence and sequences | The words around an emoji | Measuring how emoji are used | Emoji over time | Emoji as model features | Text-emoji mismatch | Translating emoji to and from text | Searching the emoji catalogue | Emoji in language-model pipelines | Recording provenance | Bundled datasets | References

The decision nobody writes down | A corpus built to break things | The ladder, run rather than described | It holds on the awkward cases | wrap is part of the contract, not a style choice | Do not throw away the signal you just removed | What it saves, and how to measure it honestly | Which policy, by intent | Put the versions in the methods section | See also

What this package is for | A table to work with | Inspecting a table | Reading | Pushdown | Verifying that pushdown happened | One caveat about limit | Time travel | Writing | Creating a table | Type fidelity | Knowing what is not supported | Where next | References

The problem inherited from Hive | What Iceberg keeps instead | Why the column statistics matter so much | Concurrency, without a lock server | Where R sits in this | What this package does not do | References