Ai Time Series Analytics

Time Series Foundations

Doctoral work examines data recorded in sequence and the dependence within it. Ordering carries information that independent sample methods entirely lose.

Stochastic Process Research

Research examines mathematical descriptions of randomness evolving through time. Process theory underpins every formal model of sequential data.

Stationarity Research

Doctoral study examines series whose statistical properties remain constant. Stationarity is assumed by many methods and rarely holds in practice.

Nonstationarity Research

Research examines series whose behaviour shifts across the observation period. Most real world series are nonstationary in one way or another.

Unit Root Research

Doctoral work examines testing whether a series wanders without any fixed level. Testing outcomes determine which modelling approaches remain valid.

Differencing Research

Research examines transforming series by taking successive observed differences. Differencing removes trend and can obscure genuine level information.

Trend Research

Doctoral study examines persistent long run movement within observed series. Distinguishing trend from slow oscillation is genuinely difficult.

Seasonality Research

Research examines patterns repeating across fixed and known periodic intervals. Seasonal structure is strong and frequently modelled rather crudely.

Cyclical Component Research

Doctoral work examines recurring movement without any fixed period length. Cycles differ from seasonality and are far harder to identify reliably.

Decomposition Research

Research examines separating a series into interpretable component parts. Decomposition choice strongly affects what analysts eventually conclude.

Autocorrelation Research

Doctoral study examines correlation between observations separated by intervals. Autocorrelation structure is the primary diagnostic in this field.

Partial Autocorrelation Research

Research examines dependence remaining after intervening observations are considered. This diagnostic guides selection of autoregressive model order.

Spectral Density Research

Doctoral work examines how variation is distributed across differing frequencies. Spectral views reveal periodicity that time plots entirely hide.

Frequency Domain Research

Research examines analysing series in terms of frequency rather than time. Frequency methods suit oscillatory signals particularly well indeed.

Fourier Analysis Research

Doctoral study examines representing series as sums of periodic functions. This representation assumes behaviour that stays constant throughout.

Wavelet Analysis Research

Research examines representations localised in both time and in frequency. Wavelets suit signals whose frequency content changes across time.

Time Frequency Research

Doctoral work examines methods describing how spectral content evolves over time. These methods reveal transient behaviour that averaging conceals.

Autoregressive Model Research

Research examines models predicting values from their own recent past history. Autoregression is the most fundamental sequential modelling idea.

Moving Average Model

Doctoral study examines models built upon recent unexplained random shocks. This formulation captures short lived effects of past disturbances.

Mixed Autoregressive Model

Research examines models combining own history with past random shocks together. Combined formulations describe many series parsimoniously and well.

Integrated Model Research

Doctoral work examines models applied after differencing removes wandering behaviour. Integration order is a modelling choice with real consequences.

Seasonal Model Research

Research examines formulations explicitly representing repeating periodic structure. Seasonal terms substantially improve many practical forecasts.

Exogenous Variable Model

Doctoral study examines including external drivers within sequential models. External information frequently improves forecasts considerably.

Model Order Selection

Research examines choosing how much history a model should actually use. Selection criteria trade goodness of fit against needless complexity.

Parameter Estimation Research

Doctoral work examines fitting model coefficients to observed sequential data. Estimation properties differ from those of independent sample settings.

Maximum Likelihood Research

Research examines estimation maximising probability of the observed sequence. Likelihood computation is demanding for many sequential models.

Bayesian Estimation Research

Doctoral study examines estimation combining prior belief with observed data. Bayesian treatment naturally expresses uncertainty about parameters.

State Space Model Research

Research examines models with hidden states evolving beneath observations. State formulations unify a very wide family of sequential models.

Kalman Filtering Research

Doctoral work examines recursive estimation of hidden states from noisy measurements. This recursion remains foundational across engineering practice.

Particle Filtering Research

Research examines sample based estimation for nonlinear state space models. Particle methods handle problems that linear recursion simply cannot.

Hidden Markov Model Research

Doctoral study examines models where observations arise from unseen discrete states. These models suit series that switch between distinct behaviours.

Regime Switching Research

Research examines series moving between several distinct behavioural states. Switching models capture behaviour that a single regime cannot.

Structural Break Research

Doctoral work examines permanent shifts in the behaviour generating a series. Unrecognised breaks badly degrade forecasts and inference alike.

Change Point Detection

Research examines identifying moments when series behaviour changed sharply. Detection supports both monitoring and retrospective explanation.

Exponential Smoothing Research

Doctoral study examines forecasts weighting recent observations more heavily. Smoothing methods remain remarkably competitive despite their simplicity.

Local Level Model Research

Research examines models where an underlying level moves slowly across time. Local level formulations underpin many established smoothing approaches.

Dynamic Linear Model

Doctoral work examines linear models whose coefficients evolve through time. Evolving coefficients accommodate gradually changing relationships.

Volatility Model Research

Research examines modelling variation whose magnitude changes across time. Changing variability matters enormously for risk and uncertainty.

Conditional Heteroskedasticity

Doctoral study examines variance depending upon recently observed behaviour. This model family dominates financial volatility modelling practice.

Stochastic Volatility Research

Research examines variability driven by its own unobserved random process. These models fit well and are considerably harder to estimate reliably.

Long Memory Research

Doctoral work examines dependence persisting across very distant observations. Long memory contradicts assumptions most standard models make.

Fractional Integration Research

Research examines models permitting noninteger degrees of differencing. Fractional formulations capture slowly decaying dependence naturally.

Nonlinear Model Research

Doctoral study examines relationships that linear formulations cannot capture. Nonlinearity is common and frequently ignored for convenience.

Threshold Model Research

Research examines models behaving differently above and below a boundary. Threshold behaviour appears widely in economic and physical systems.

Chaotic Dynamics Research

Doctoral work examines deterministic systems whose behaviour appears random. Sensitivity to conditions limits how far ahead prediction extends.

Multivariate Time Series

Research examines several series observed and modelled together jointly. Joint modelling exploits dependence that separate analysis discards.

Vector Autoregression Research

Doctoral study examines systems where each series depends upon all histories. These systems grow rapidly and demand careful parameter restriction.

Cointegration Research

Research examines wandering series that nevertheless move together over time. Cointegration reveals stable relationships beneath unstable series.

Error Correction Model

Doctoral work examines models pulling series back toward long run relationships. These models separate short run movement from long run structure.

Granger Causality Research

Research examines whether one series helps predict another beyond its history. Predictive precedence is weaker than genuine causal evidence.

Impulse Response Research

Doctoral study examines how systems respond to a single isolated disturbance. Response functions summarise dynamic behaviour rather interpretably.

Dimensionality Reduction

Research examines summarising very many series through far fewer components. Reduction makes extremely wide collections tractable to model at all.

Factor Model Research

Doctoral work examines shared drivers explaining movement across many series. Factor structure explains most variation in large collections.

Dynamic Factor Research

Research examines shared drivers that themselves evolve through time. Dynamic factors combine reduction with sequential modelling naturally.

High Dimensional Research

Doctoral study examines settings with far more series than there are observations. Standard estimation fails entirely within these wide settings.

Machine Learning Applications

Research applies learned models across sequential prediction and detection. Learned models require validation on genuinely held out periods.

Feature Engineering Research

Doctoral work examines constructing informative inputs from raw sequential data. Feature quality frequently matters more than the model chosen.

Lag Feature Research

Research examines using past observations as inputs to predictive models. Lag choice determines what history a model can actually exploit at all.

Rolling Statistic Research

Doctoral study examines summaries computed across moving observation windows. Window length determines which timescale the summary actually reflects.

Representation Learning Research

Research examines models learning useful summaries directly from sequences. Learned representations avoid manual feature construction entirely.

Sequence Embedding Research

Doctoral work examines numerical representations capturing sequence characteristics. Embeddings support search, clustering and comparison tasks.

Self Supervised Research

Research examines learning from sequences without any provided labels at all. Unlabelled sequential data is abundant where labels are very scarce.

Contrastive Learning Research

Doctoral study examines learning by distinguishing similar from dissimilar sequences. Defining similarity for sequences is genuinely nontrivial.

Recurrent Network Research

Research examines neural models processing sequences one step at a time. Recurrent architectures dominated sequential learning for a long period.

Long Short Term Memory

Doctoral work examines recurrent architectures designed to retain distant information. These architectures addressed problems that plain recurrence had.

Gated Recurrent Research

Research examines simplified recurrent designs with fewer internal components. Simpler gating performs comparably with substantially less computation.

Convolutional Model Research

Doctoral study examines convolution applied along the temporal data dimension. Convolutional models train considerably faster than recurrent ones.

Dilated Convolution Research

Research examines convolutions spanning long ranges without many stacked layers. Dilation captures distant dependence at modest computational cost.

Attention Mechanism Research

Doctoral work examines models weighting relevant positions within a sequence. Attention captures dependence regardless of separation distance.

Transformer Model Research

Research examines attention based architectures applied to sequential data. Whether these models beat simpler approaches remains genuinely contested.

Sequence Model Research

Doctoral study examines architectures designed specifically for ordered data. Architecture choice must respect the causal ordering that time imposes.

Sequence To Sequence Research

Research examines models mapping input sequences onto whole output sequences. This framing suits generating entire forecast paths in one step.

Foundation Model Research

Doctoral work examines large models pretrained across very many series. Pretrained models promise forecasting without any task specific fitting.

Pretraining Research

Research examines training on broad data before any specific application. Pretraining transfers structure that small datasets cannot supply.

Transfer Learning Research

Doctoral study examines reusing models across differing series and domains. Transfer works only where series share genuine underlying structure.

Few Shot Learning Research

Research examines forecasting series with extremely limited observed history. Short histories are common and defeat conventional estimation.

Zero Shot Forecasting

Doctoral work examines producing forecasts without fitting to the target series. Claims here require careful checking against contamination.

Meta Learning Research

Research examines learning how to select or adapt forecasting approaches. Meta approaches exploit accumulated experience across many series.

Ensemble Method Research

Doctoral study examines combining several models into one overall prediction. Ensembles reliably outperform each of their individual member models.

Boosting Research

Research examines sequentially built tree ensembles applied to forecasting. Tree ensembles remain extremely competitive on tabular sequential tasks.

Hybrid Model Research

Doctoral work examines combining statistical structure with learned components. Hybrid designs have won several major forecasting competitions.

Physics Informed Model

Research examines embedding known physical laws within learned models. Physical constraints improve behaviour where data is genuinely scarce.

Neural Differential Equation

Doctoral study examines learned models expressed as continuous dynamical systems. This formulation handles irregularly spaced observations naturally.

Continuous Time Model

Research examines models defined over continuous rather than discrete time. Continuous formulations suit data arriving at irregular moments.

Generative Model Research

Doctoral work examines models producing plausible new sequential observations. Generative models support simulation and uncertainty representation.

Diffusion Model Research

Research examines noise based generative methods applied to sequential data. These methods produce notably sharp probabilistic forecast distributions.

Synthetic Data Research

Doctoral study examines artificially generated sequences supporting model development. Synthetic sequences support testing where real data is restricted.

Data Augmentation Research

Research examines expanding training data through principled transformations. Augmenting sequences must respect the temporal ordering present.

Graph Based Model Research

Doctoral work examines models exploiting known relationships between series. Graph structure encodes dependence that raw correlation misses.

Spatiotemporal Model Research

Research examines data varying across both space and time simultaneously. Joint treatment exploits structure that separate analysis simply discards.

Spatial Dependence Research

Doctoral study examines nearby locations behaving more similarly than distant ones. Spatial structure is strong within environmental measurement.

Network Time Series

Research examines series observed at the nodes of a connected network. Network position influences how disturbances propagate between series.

Model Interpretability Research

Doctoral work examines making sequential model behaviour understandable to people. Interpretability determines whether forecasts are actually trusted.

Attribution Method Research

Research examines identifying which inputs drove a particular prediction. Attribution for sequences must respect temporal dependence structure.

Counterfactual Analysis Research

Doctoral study examines what predictions would follow from differing inputs. Counterfactual reasoning supports both explanation and planning.

Model Compression Research

Research examines reducing model size while retaining predictive capability. Compression enables deployment where computation is constrained.

Efficient Inference Research

Doctoral work examines producing predictions quickly and at very low cost. Inference cost matters greatly when forecasting very many series.

Edge Deployment Research

Research examines running sequential models on small local computing devices. Local processing avoids transmitting continuous measurement streams.

Streaming Model Research

Doctoral study examines models operating on continuously arriving observations. Streaming settings forbid revisiting the entire data history.

Online Learning Research

Research examines models revised as each new observation becomes available. Online revision keeps models current without complete refitting.

Incremental Learning Research

Doctoral work examines models absorbing new data without complete retraining. Incremental methods suit settings where refitting is impractical.

Concept Drift Research

Research examines relationships changing after a model has already been fitted. Drift silently degrades models that appeared entirely satisfactory.

Distribution Shift Research

Doctoral study examines deployment data differing from the training period. Shift is the commonest reason that deployed models quietly fail.

Model Monitoring Research

Research examines tracking deployed model performance across ongoing operation. Monitoring detects degradation before decisions are badly affected.

Retraining Strategy Research

Doctoral work examines deciding when models should be fitted entirely afresh. Refitting too often wastes effort and too rarely permits decay.

Continual Learning Research

Research examines models learning across a long succession of separate tasks. Continual settings demand retaining capability while acquiring more.

Catastrophic Forgetting Research

Doctoral study examines models losing earlier capability while learning anew. Forgetting is severe within sequentially trained neural models.

Scalability Research

Research examines methods handling very many series simultaneously together. Operational settings routinely involve millions of separate series.

Distributed Computation Research

Doctoral work examines spreading sequential computation across many machines. Distribution is necessary once data exceeds any single machine.

Hardware Acceleration Research

Research examines specialised processors accelerating sequential model computation. Hardware choice determines what model sizes remain practical.

Forecasting Research

Doctoral study examines predicting future values of an observed data series. Forecasting is the dominant purpose of sequential data analysis.

Point Forecast Research

Research examines producing single best estimates of a future value. Point estimates are convenient and convey no uncertainty information whatsoever.

Probabilistic Forecast Research

Doctoral work examines forecasts expressing the full range of possibilities. Decisions require uncertainty that single estimates cannot provide.

Interval Forecast Research

Research examines ranges expected to contain the eventually observed value. Reported intervals are frequently narrower than reality would warrant.

Quantile Forecast Research

Doctoral study examines predicting specified points of the outcome distribution. Quantile forecasts suit decisions with asymmetric consequences.

Density Forecast Research

Research examines predicting the complete distribution of a future outcome. Full distributions support any decision rule a user might apply.

Multistep Forecast Research

Doctoral work examines predicting several future periods ahead simultaneously. Error accumulation makes distant prediction progressively harder.

Forecast Horizon Research

Research examines how far ahead useful prediction remains genuinely possible. Predictability limits differ enormously between differing systems.

Forecast Combination Research

Doctoral study examines merging predictions from several separate sources. Combination is among the most reliable findings within this field.

Forecast Reconciliation

Research examines making related forecasts consistent with one another. Reconciliation improves accuracy as well as merely ensuring coherence.

Hierarchical Forecast Research

Doctoral work examines forecasting series organised within nested structures. Hierarchies appear throughout retail, energy and administrative data.

Grouped Series Research

Research examines collections aggregated along several crossing dimensions. Grouped structures are more complex than simple nested hierarchies.

Intermittent Demand Research

Doctoral study examines series containing many periods of zero observation. Intermittent patterns defeat methods designed for continuous data.

Sparse Series Research

Research examines series with very few nonzero or observed values present. Sparsity demands methods that ordinary averaging cannot ever supply.

Cold Start Research

Doctoral work examines forecasting series with essentially no observed history. New products and newly installed sensors both create this problem.

Cross Learning Research

Research examines learning shared patterns across many related series. Cross learning permits forecasting series individually too short to model.

Global Model Research

Doctoral study examines single models fitted across an entire collection. Global fitting frequently outperforms fitting each series separately.

Local Model Research

Research examines fitting a separate model to each individual series. Local fitting suits collections of genuinely dissimilar observed series.

Forecast Evaluation Research

Doctoral work examines judging how good a set of forecasts actually was. Evaluation design determines which methods appear to perform the best.

Accuracy Metric Research

Research examines measures quantifying the size of observed forecast errors. Metric choice can entirely reverse a comparison between methods.

Scale Free Metric Research

Doctoral study examines measures comparable across series of differing magnitude. Scale free measures permit averaging across whole collections.

Forecast Calibration Research

Research examines whether stated probabilities match observed outcome frequencies. Poor calibration makes probabilistic forecasts actively misleading.

Forecast Sharpness Research

Doctoral work examines how concentrated predictive distributions actually are. Sharpness matters only alongside demonstrated good calibration.

Backtesting Research

Research examines evaluating methods against historical observed outcomes. Backtests routinely overstate the performance later achieved live.

Temporal Cross Validation

Doctoral study examines validation respecting the ordering of observations. Ordinary random splitting is entirely invalid for sequential data.

Data Leakage Research

Research examines future information improperly entering model training. Leakage produces spectacular results that never survive deployment.

Benchmark Research

Doctoral work examines shared datasets used to compare forecasting methods. Benchmark composition strongly shapes which methods appear superior.

Forecasting Competition Research

Research examines organised contests comparing methods on common problems. Competitions have repeatedly favoured simple and combined approaches.

Baseline Research

Doctoral study examines simple reference methods against which others compete. Many published methods fail to beat very simple naive baselines.

Reproducibility Research

Research examines whether reported forecasting results can be repeated. Undisclosed preprocessing choices obstruct attempted reproduction badly.

Anomaly Detection Research

Doctoral work examines identifying observations departing from expected behaviour. Detection supports monitoring across industry and infrastructure.

Outlier Research

Research examines individual observations lying far from the surrounding values. Outliers may be errors or the most interesting events present.

Novelty Detection Research

Doctoral study examines recognising behaviour never previously observed. Novelty differs from anomaly and demands differing detection approaches.

Rare Event Research

Research examines events occurring too seldom for conventional learning methods. Rarity means examples are scarce and consequences are severe.

Fault Detection Research

Doctoral work examines identifying equipment problems from measurement streams. Early identification prevents failures that become far costlier.

Early Warning Research

Research examines detecting signals preceding a serious system transition. Warning value depends entirely upon how much lead time is provided.

Sequence Classification Research

Doctoral study examines assigning whole sequences to defined categories. Classification supports diagnosis, recognition and quality assessment.

Segmentation Research

Research examines dividing sequences into internally consistent portions. Segmentation precedes analysis where behaviour changes across time.

Sequence Clustering Research

Doctoral work examines grouping series exhibiting similar behavioural patterns. Clustering reveals structure within very large series collections.

Similarity Measure Research

Research examines quantifying how alike two sequences genuinely are. Similarity definition determines every clustering and retrieval result.

Dynamic Time Warping

Doctoral study examines comparing sequences that differ in timing or speed. Warping matches patterns that direct comparison would entirely miss.

Motif Discovery Research

Research examines finding repeated patterns within long observed sequences. Recurring motifs frequently correspond to meaningful real behaviours.

Discord Discovery Research

Doctoral work examines locating the least typical portions of a sequence. Unusual portions frequently mark the events analysts actually want.

Event Detection Research

Research examines recognising defined occurrences within measurement streams. Detection converts continuous signals into interpretable event records.

Causal Inference Research

Doctoral study examines establishing causal relationships from sequential observation. Temporal ordering helps and does not establish causation alone.

Intervention Analysis Research

Research examines estimating the effect of a known event upon a series. Intervention methods separate genuine effect from ordinary variation.

Counterfactual Forecast Research

Doctoral work examines estimating what would have happened without an event. Counterfactual estimates underpin most impact evaluation work done.

Synthetic Control Research

Research examines building comparison series from unaffected observed units. Constructed comparisons substitute for genuine experimental control.

Missing Data Research

Doctoral study examines gaps within otherwise regular observed sequences. Gap handling choices propagate into every subsequent analysis step.

Irregular Sampling Research

Research examines observations arriving at unevenly spaced moments. Irregular spacing defeats methods assuming a fixed observation interval.

Measurement Error Research

Doctoral work examines imprecision within the recorded observations themselves. Measurement error attenuates relationships and inflates apparent noise.

Signal Filtering Research

Research examines separating meaningful signal from unwanted recorded variation. Filter choice determines what information survives into analysis.

Denoising Research

Doctoral study examines removing noise while preserving genuine signal features. Aggressive denoising destroys the sharp transitions that matter.

Resampling Research

Research examines converting series between differing observation frequencies. Frequency conversion loses information in one direction always.

Series Alignment Research

Doctoral work examines synchronising series recorded on differing time bases. Misalignment silently corrupts any joint analysis that is attempted.

Multivariate Fusion Research

Research examines combining measurements from several differing sources. Fusion exploits complementary information that single sources lack.

Sensor Data Research

Doctoral study examines series produced by continuously recording instruments. Sensor streams are enormous, noisy and frequently incomplete.

Streaming Infrastructure Research

Research examines systems ingesting and processing continuous data flows. Infrastructure design determines what analysis remains feasible live.

Time Series Storage Research

Doctoral work examines databases designed specifically for sequential measurement. Specialised storage handles volumes general databases cannot.

Compression Research

Research examines reducing storage required for very large series collections. Compression choice trades storage against later query performance.

Indexing Research

Doctoral study examines structures enabling rapid search across many sequences. Indexing makes similarity search feasible at very large scale.

Query Research

Research examines languages and methods for interrogating sequential data. Query capability determines what questions analysts can readily ask.

Visualisation Research

Doctoral work examines graphical presentation of sequences and their forecasts. Visual design strongly influences what viewers actually conclude.

Annotation Research

Research examines marking meaningful occurrences within recorded sequences. Annotation quality bounds what any supervised method can achieve.

Labelling Research

Doctoral study examines obtaining reliable labels for sequential observations. Labelling sequences is far harder than labelling isolated items.

Dataset Research

Research examines collections used to develop and compare sequential methods. Available datasets poorly represent many real application settings.

Privacy Research

Doctoral work examines protecting individuals within detailed measurement streams. Fine grained sequences reveal behaviour with striking clarity.

Federated Learning Research

Research examines learning across sites without centralising the raw data. Federation supports collaboration where sharing is entirely prohibited.

Energy Application Research

Doctoral study examines forecasting within electricity and energy systems. Energy systems require accurate prediction in order to stay balanced.

Demand Forecast Application

Research examines predicting requirement for goods, services or capacity. Demand prediction underpins planning across nearly every single sector.

Financial Application Research

Doctoral work examines sequential methods within financial market analysis. Market series are extremely noisy and only very weakly predictable.

Retail Application Research

Research examines forecasting sales across very many products and locations. Retail collections are enormous, hierarchical and highly intermittent.

Supply Chain Application

Doctoral study examines prediction supporting inventory and logistics decisions. Forecast errors propagate and amplify along the whole chain.

Manufacturing Application

Research examines sequential monitoring within industrial production processes. Process signals reveal deviation long before products fail testing.

Predictive Maintenance Research

Doctoral work examines anticipating equipment failure from measurement streams. Anticipation permits repair before any unplanned breakdown occurs.

Healthcare Application Research

Research examines sequential methods applied to clinical measurement data. Clinical series are irregular, incomplete and consequential to interpret.

Physiological Signal Research

Doctoral study examines continuously recorded signals from the human body. Physiological signals combine high frequency with substantial artefact.

Epidemiological Application

Research examines forecasting disease occurrence within whole populations. Reporting delay and revision complicate this forecasting substantially.

Climate Application Research

Doctoral work examines sequential methods within climate and weather science. Climate series demand handling of very long dependence structures.

Environmental Monitoring

Research examines continuous measurement of environmental conditions. Monitoring networks generate series with substantial missing observation.

Hydrological Application

Doctoral study examines forecasting river flow, rainfall and water availability. Hydrological prediction supports both flood and drought response.

Transport Application Research

Research examines forecasting traffic, movement and transport network demand. Transport series combine strong spatial and temporal dependence.

Telecommunications Application

Doctoral work examines forecasting network load and communication traffic. Load prediction supports capacity planning and fault identification.

Web Traffic Application

Research examines forecasting online activity and platform usage patterns. Online series show sharp bursts that models struggle to anticipate.

Astronomical Application

Doctoral study examines sequential analysis of astronomical brightness measurements. Astronomical series are irregular and extremely large in volume.

Fairness Research

Research examines whether sequential models perform equally across groups. Unequal performance concentrates forecast error upon some populations.

Governance Research

Doctoral work examines oversight of models informing consequential decisions. Governance determines what automated prediction may actually decide.

Workforce And Skills Research

Research examines expertise required across sequential analytics practice. Combined statistical and engineering capability remains genuinely scarce.

Economic Evaluation Research

Doctoral study examines value delivered by improved forecasting capability. Accuracy gains translate into decision value rather inconsistently.

Implementation And Adoption

Research examines why forecasting advances reach practice or fail to do so. Organisational judgement frequently overrides the model output entirely.