Why | What | Example | The one-call path | The masking plan: roles and actions | Editing the plan as code | A plan the package refuses | More than one table | Multi-environment structure | Figure: the multi-environment coverage plot | Results | Limits | What to read next | Reproduce

The question | The tools | The plan | Scenario A: the full trial | Scenario B: one site, one replicate, three rates | The plan as a file | Writing a plan | The schema and the repair loop | A mistyped edge | Provenance and a targets pipeline

Why | What | Do | Is the simulator available, and what happens when it is not? | Reading a simulation file | Prediction | Estimation: fitting a reduced-form mechanism | The forward model: one machine for three verbs | Calibration | Sensitivity analysis | Emulation | Validation | Ground truth: APSIM as a causal test-bench | Composing with the orchestra | Read | Limits | What to read next | Reproduce

Why | What | Do | A nitrogen rate from a real yield response | The firewall: unverified evidence abstains | A variety choice from a multi-environment trial | A quality-grade decision | From an upstream manifest | Composing a season plan | Read | Limits | What to read next | Reproduce

Why | What | Do | Emit, verify, consume | Provenance is not optional for a derived result | A verdict is not an ensemble | Refusals and abstentions | Grounding tokens | Comparing two manifests on meaning, not position | Adapting a component | Two class identities, one contract | Read | Limits | What to read next | Reproduce

What can it fit? | A first fit | Switching method, same interface | The small-sample gate | Visualising a fit | Competing risks | Cumulative incidence | Decision-relevant summaries | Triangulating across methods

The problem | Package capabilities | Addressing the problem | A target with both a formula and a sample | Method (i): one normal distribution with the sample's mean and spread | Method (ii): several normal distributions fitted to the sample | Method (iii): several normal distributions fitted to the formula | The three fits side by side | What each method improves in each round | One normal distribution fitted two ways, against the exact answer | Comparison with mclust, mixtools and flexmix | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A posterior with no sampler | Fit the proxy | Check the fit before using it | The normalising constant | Read the answers off the proxy | Check the answers against exact integration | Error bars on the proxy itself | Comparison with Stan, INLA, a Laplace approximation and BayesianTools | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A sample from two groups | The recovered groups | Check the fit | What the compression loses | Use the proxy | The effect of the bandwidth | Comparison with ks, np, KernSmooth, mclust and mixtools | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Which fitting method applies | A target with no sampler | Fit the proxy | Check the fit before using it | Compare the proxy with the target | Use the proxy | Comparison with the Laplace approximation, Stan and BayesianTools | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Fit the three shapes that cover the whole plane | Compare each proxy with its target | Measure each fit in two ways | Give the ring more components and more trial points | A density with hard edges | When the trial points miss the target | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Data with two groups and known missing values | Impute with a mixture and with a single normal distribution | Where the imputed values fall | Pool the mean of x2 | Pool a regression slope with mice | Comparison with mice and Amelia | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | From an objective to a distribution | Two minima in one dimension | Check the fit before using it | Four minima in two dimensions | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Data in which the large values go missing | Impute under two assumptions | Sweep the assumed slope | Censoring at a known limit | Comparison with mice, Amelia, sampleSelection and the Tobit model | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Regression: lm() | Kernel regression: Nadaraya-Watson | Clustering: kmeans() | Principal components: prcomp() | Penalised regression: ridge | Treatment effects: a separate model for each arm | Other uses | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Why some quantities are exact | Entropy of a mixture | Distance between two mixtures | Dependence between variables | Cooling the fit | Maximum-entropy targets | Choosing the number of components | Which variables are related | Comparison with FNN, mclust and pcalg | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A model and a stable series | The same series with a broken final value | What the filter expected | How the subsampling p-value is reached | How often each calibration rejects when the model is known | Comparison with strucchange, changepoint, KFAS and dlm | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A mixture to work with | Pass the mixture through a sensor | Update after a noisy measurement | Check the update against the Kalman formulas | Add variables together | Condition on values known exactly | Two measurements, in turn or together | Track an object over time | Keep the number of components small | The whole filter in one call | Comparison with dlm, KFAS and particle filters | Interpretation | Limitations | Further reading | References | Reproduce

Why | What | Example | Depth controls: hiding design structure | The conditional clone: preserving the treatment effect | Figure: the effect on each clone | What the copula carries: monotone, not non-monotone, association | The leakage audit, and a refused write | Multi-table sets | Geographic coordinates | Results | Limits | What to read next | Reproduce | References

Why | What | Example | What a recipe holds | Printing is redacted | The round-trip | Fail-closed translation | Saving the recipe | Multi-table bundles | Figure: the clone the analyst actually develops against | Results | Limits | What to read next | Reproduce

Why | What | The pass rule | Example | The eight designs | Masking and measuring | The table | Figure: the F statistic per term, clone against original | Results | Limits | What to read next | Reproduce

Why | What | Do | The registered sources | Registering a new source | A request | Collecting (live) | Offline replay (zero network) | A figure: what the replayed series looked like | Sovereign coordinates (coarsening) | The response contract | Graceful refusal | Batch collection | Sources at a glance | Read | Limits | What to read next | Reproduce

The problem | Package capabilities | Addressing the problem | Data in which the large values go missing | Impute under two assumptions | Sweep the assumed slope | Censoring at a known limit | Comparison with mice, Amelia, sampleSelection and the Tobit model | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A model and a stable series | The same series with a broken final value | What the filter expected | How the subsampling p-value is reached | How often each calibration rejects when the model is known | Comparison with strucchange, changepoint, KFAS and dlm | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Which fitting method applies | A target with no sampler | Fit the proxy | Check the fit before using it | Compare the proxy with the target | Use the proxy | Comparison with the Laplace approximation, Stan and BayesianTools | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Why some quantities are exact | Entropy of a mixture | Distance between two mixtures | Dependence between variables | Cooling the fit | Maximum-entropy targets | Choosing the number of components | Which variables are related | Comparison with FNN, mclust and pcalg | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A target with both a formula and a sample | Method (i): one normal distribution with the sample's mean and spread | Method (ii): several normal distributions fitted to the sample | Method (iii): several normal distributions fitted to the formula | The three fits side by side | What each method improves in each round | One normal distribution fitted two ways, against the exact answer | Comparison with mclust, mixtools and flexmix | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A posterior with no sampler | Fit the proxy | Check the fit before using it | The normalising constant | Read the answers off the proxy | Check the answers against exact integration | Error bars on the proxy itself | Comparison with Stan, INLA, a Laplace approximation and BayesianTools | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A sample from two groups | The recovered groups | Check the fit | What the compression loses | Use the proxy | The effect of the bandwidth | Comparison with ks, np, KernSmooth, mclust and mixtools | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Fit the three shapes that cover the whole plane | Compare each proxy with its target | Measure each fit in two ways | Give the ring more components and more trial points | A density with hard edges | When the trial points miss the target | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Data with two groups and known missing values | Impute with a mixture and with a single normal distribution | Where the imputed values fall | Pool the mean of x2 | Pool a regression slope with mice | Comparison with mice and Amelia | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | From an objective to a distribution | Two minima in one dimension | Check the fit before using it | Four minima in two dimensions | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | Regression: lm() | Kernel regression: Nadaraya-Watson | Clustering: kmeans() | Principal components: prcomp() | Penalised regression: ridge | Treatment effects: a separate model for each arm | Other uses | Interpretation | Limitations | Further reading | References | Reproduce

The problem | Package capabilities | Addressing the problem | A mixture to work with | Pass the mixture through a sensor | Update after a noisy measurement | Check the update against the Kalman formulas | Add variables together | Condition on values known exactly | Two measurements, in turn or together | Track an object over time | Keep the number of components small | The whole filter in one call | Comparison with dlm, KFAS and particle filters | Interpretation | Limitations | Further reading | References | Reproduce

Why | What | Do | Causality as an assimilation problem | A known-truth test: a recharge oscillator | The null carries no causal information | A divergence between two Gaussians, cross-checked | Grounding the verdict: abstention under model error | The lead-time without re-smoothing | A real ENSO index | Read | Limits | What to read next | Reproduce

Why | What | Do | A mean shift, which any method would find | A variance effect and nothing else | The effect on the treated, with imperfect overlap | The same question through a formula | The figure: what the mean cannot show | Read | Limits | What to read next | References | Reproduce

Why | What | Do | A sample that follows its target | The same sample, shifted | A target supplied as an unnormalised log density | Three sources that agree | One source departs | The figure: what the pairwise matrix is measuring | Read | Limits | What to read next | Reproduce

Why | What | Do | Which backends this build carries | A confounded design | The four ratios | The figure: the weights behind the diagnostic | The causal test under each backend | Read | Limits | When to reach for which | What to read next | References | Reproduce

Why | What | Do | One calibration | Two forward runs from that one posterior | The scenario test | The null case | Restricting the test to named outputs | The figure: the ensembles the test is comparing | What the comparison refuses to do | Read | Limits | What to read next | References | Reproduce

Why | What | Do | A linear causal effect | No causal effect, confounding only | A curved causal effect | The formula interface | The figure: two nulls, two verdicts | When the test declines to answer | Read | Limits | What to read next | References | Reproduce

Why | What | Do | Paired data: coarse climate to paddock yield | The figure: the downscaled surface against the field it came from | Bags of points: distribution regression | Choosing the outer kernel | No paired data: inverting a known aggregator | The three primitives side by side | Read | Limits | Notes on practice | What to read next | References | Reproduce

Why | What | Do | The kernel specification | The Gram matrix | Two samples, same distribution | Two samples, one shifted | Dependence a correlation cannot see | The figure: where the observed statistic sits | The three verdicts side by side | Read | Limits | What to read next | Reproduce

Why | What | Do | A held-out sample | A calibrated ensemble | An over-confident ensemble | A displaced ensemble | The package's own diagnostic | The figure: what each ensemble got wrong | The same check on a PESTO manifest | Read | Limits | Notes on practice | What to read next | References | Reproduce

Why | What | Do | A synthetic eight-site panel | The same data, two permutation schemes | The same panel with the effect removed | The stratified breakdown | The figure: how the two nulls differ | Read | Limits | Choosing the scheme | Notes on practice | What to read next | References | Reproduce

Why | What | Do | Does the factorisation reproduce the kernel? | What it costs and what it saves | The figure: where the two curves separate | Read | Limits | Notes on practice | What to read next | References | Reproduce

Why | What | Do | A synthetic twenty-farm trial | The decomposition | The three weighting conventions on the same data | The same trial with no treatment effect | The figure: where the signal sits | Read | Limits | What to read next | Reproduce

Why | What | Do | A stand-in simulator | The design | One pass through the simulator, then the screen | The package's ranked diagnostic | The figure: the dependence the index is measuring | Handing the survivors to the calibration | Read | Limits | Notes on practice | What to read next | References | Reproduce

Why | What | Do | A simulator built to separate the two questions | The first-order scan | The figure: the distributions behind the index | Comparing against Sobol | Total-order indices, and a pure interaction | A near-additive design for contrast | An interval on the total-order index | The properly calibrated significance test | Read | Limits | Notes on practice | What to read next | References | Reproduce

Why | What | Do | The calibration target | The synthetic observations | The calibration call | Convergence | Parameter recovery | Identifiability: what the data can constrain | Out-of-sample validation, and the under-dispersion caveat | The record the run emits | Part 2: calibration to real observed data | Read | Limits | What to read next | References | Reproduce

Why | What | Do | A run worth handing over | Wrapping it | Writing, reading, verifying | Three formats, three integrity contracts | What tampering looks like | A run that declined to happen | What a conformant consumer does | Read | Limits | What to read next | Reproduce

Why | What | Do | The cross-tool benchmark | Does PESTO drive the binary faithfully? | Where the cost model comes from | A worked example that runs here | Read | Limits | What to read next | References | Reproduce

Why | What | Do | A problem whose posterior is known | The bare smoother under-disperses | Is this finite-ensemble collapse, or something else? | Inflation re-expands the spread | Localisation suppresses spurious correlations | The collapse trajectory | Read | Limits | What to read next | References | Reproduce

Why | What | Regime of applicability | Do | The test problem | The prior ensemble | Training a process on it | The check that matters: the surrogate against the simulator | The blended smoother update | Does the blended update change the answer? | A favourable regime | When the regime is unfavourable | Effect of nonlinearity | Random Fourier features for large ensembles | Read | Limits | What to read next | References | Reproduce

Why | What | Do | A synthetic recovery example | The forward-model contract | Parallel, fault-tolerant ensembles | The number that decides whether to believe the answer | Where the boundary is | What the spread-ESS diagnostic can and cannot see | Driving APSIM through apsim_callback() | Multi-fidelity calibration | Simulators that are systems of equations | Assimilating a season as it happens | When to prefer the .pst path instead | Read | Limits | What to read next | References | Reproduce

Why | What | Do | Installation, and what needs an external binary | A synthetic season | A prior, and the inversion | Did the misfit fall? | What the data did to the prior | The run as a portable record | Under the bonnet | Read | Limits | What to read next | References | Reproduce

Why | What | Do | Change-point detection | The penalty controls over-segmentation | Hidden Markov models | Read | Limits | What to read next | Reproduce

Why | What | Do | A linear-Gaussian state-space model | Filtering: recovering the state online | Smoothing: using the whole series | A vector state | Read | Limits | What to read next | Reproduce

Why | What | Do | A cointegration spread | Fitting the mean reversion | Tracking the volatility regime | A causal read on an event | Read | Limits | What to read next | Reproduce

Why | What | Do | The 2005 model | Why least squares on the observed spread misprices the trade | Estimating the model through the filter | The filter and the trading rule | The single-regime base case of the regime filter | Read | Limits | What to read next | Reproduce

Why | What | Do | Orderings as a design space | A worked example | The delta hook | Same trajectory, fewer evaluations | Counting true evaluations | A figure: the sparse footprint of true evaluations | Read | Limits | What to read next | Reproduce

Why | What | Do | The easy path | Which optimisers are available | An optim()-compatible result | The power path | A figure: where the optimum sits in the landscape | Read | Limits | What to read next | Reproduce

Why | What | Do | A synthetic surface with four equally good answers | Two paths, one function | Asking auto for a map | When the map cannot be honoured | The orchestra manifest | Read | Limits | What to read next | Reproduce

Why | What | Do | The objective | Engines available on this build | Tier 1: the rules | Tier 3: racing | Asking for a race directly | Budgets are respected | Why the race commits to a leader | The evidence | Read | Limits | What to read next | Reproduce

Why | What | Do | The registry | Running one method directly | The recommender explains itself | When unsure, ask several: consensus | The benchmark evidence behind the routing | Scoring and benchmarking your own runs | Beyond a sample: an evaluable density | Read | Limits | What to read next | Reproduce

Why | What | Do | From counts to a correlation matrix | Discovery directly from the statistic | Two ways to prune, at a sample size large enough to matter | Letting the recommender route it | Passing the result on | Read | Limits | What to read next | Reproduce

Why | What | Do | Engines available on this build | A real row-column trial | Fitting the Gaussian process | Two paddock blocks over the trial | The smoothed surface, with the block footprints marked | Block-average predictions: mean and standard deviation | Why this is not a cell-wise average | Abstention: when the data will not bear the answer | Emitting the orchestra contract | Scaling to large fields | Read | Limits | What to read next | Reproduce | References

Why | What | Do | The request: a pass | The battery | One lens: delayed-rejection adaptive Metropolis | Diversity: a heavy-tailed structure variant | Engines available on this build | Figure: the stacking weights | Figure: the pooled predictive envelope | Mixed models: a random intercept | Multi-environment trials: variety and trial together | When koine abstains | Read | Limits | What to read next | Reproduce

Why | What | Do | A nitrogen-rate decision | The grounding firewall, applied to a nitrogen rate | The same engine, a different domain: sizing a trading position | The pipeline tail: deciding from an upstream manifest | A threshold-economics loss: grade-band payoffs | Read | Limits | What to read next | Reproduce

Why Pearson is not enough | Quick start | Non-linear detection | Asymmetry -- a heteroscedastic example | Partial smooths (controlling for covariates) | Changing the palette | Handling missing data | Scaling up -- order = "hclust" | Programmatic access -- janusplot_data() | Shape metrics explained | Derivative views: theoretical justification and applied use | What derivatives reveal that the fit hides | Estimation -- the LP matrix | Noise amplification and why we cap at $k = 2$ | Applied use: gain estimation and dose--response | References cited in this section | Limitations | Citation

Design | Pre-registered hypotheses | Precomputed demo | Recovery curves (headline figure) | Archetype confusion | Archetype-level accuracy grid | Numerical summary | Running your own sweep | Custom shape subsets + cutoffs | References

Why this package exists | 1. Estimand and estimators | 2. Simulate a staggered-treatment BACI panel | 3. Naive paired-DiD | 4. Time-varying synthetic-control comparator | 5. Proponent-level cluster bootstrap | 6. Goodman-Bacon decomposition (diagnostic) | 7. Oracle-recovery log | 8. Drop-in targets skeleton | Reproducibility

What this vignette demonstrates | Stage 1 --- load the toy cohort | Stage 2 --- build a small illustrative feature matrix | Stage 3 --- stratified train / test split | Stage 4 --- fit the three models | Stage 5 --- predict on the held-out fold | Stage 6 --- evaluation (the manuscript headline) | Calibration headline | Decision-curve net benefit at clinically relevant thresholds | Discrimination (supporting) | How this maps to the manuscript

Status | Data source | Unit of analysis | Inclusion criteria | Exclusion criteria | Outcome | Prediction window | Variable contract | Splits and resampling | Invariants (executable contract) | Reproducibility | Out of scope | Reporting | Pedagogy (D5 default --- balanced) | Changelog