Glossary
Purpose
Define canonical terms used across DSAMbayes modelling, runner, diagnostics, and release documentation.
How to use this glossary
- Use these definitions when writing or reviewing DSAMbayes documentation.
- Keep term usage consistent across
docs/modelling/,docs/runner/, anddocs/internal/. - If a term changes behaviour in code, update this page in the same change.
Terms
| Term | Definition | Primary location |
|---|---|---|
adstock |
Media carry-over transform that spreads spend effect over subsequent periods. | Stan media-transform templates and modelling docs |
allocation |
Post-fit budget optimisation stage (allocation.* in YAML). |
docs/runner/config-schema.md |
artefact |
File written by runner validate/run workflows. | docs/runner/output-artifacts.md |
baseline term |
Non-media explanatory term (for example trend, seasonality, holiday controls). | docs/modelling/diagnostics-gates.md |
blm |
Base DSAMbayes model class for non-pooled regression workflows. | R/blm.R |
blocked CV |
Expanding-window time-series cross-validation used for model selection. | R/time_series_cv.R |
boundary |
Lower and upper constraints on model parameters. | docs/modelling/priors-and-boundaries.md |
chain diagnostics |
MCMC quality diagnostics such as Rhat, ESS, and divergence indicators. |
R/diagnostics.R |
config resolution |
Process of applying defaults, coercions, path normalisation, and validation to YAML. | R/run_config_*.R |
CRE |
Correlated random effects approach using Mundlak-style within and between variation terms. | R/cre_mundlak.R |
decomp |
Post-fit decomposition of fitted response into predictor-level contributions. | R/decomp.R |
diagnostics gate |
Thresholded pass/warn/fail policy checks over model diagnostics. | R/diagnostics_report.R |
divergence |
Stan sampler warning indicating problematic Hamiltonian trajectories. | MCMC diagnostics outputs |
dry run |
Runner mode that validates config and data without Stan fitting (validate). |
scripts/dsambayes.R |
ELPD |
Expected log predictive density, used for predictive model comparison. | R/compare_runs.R |
ESS |
Effective sample size for MCMC draws. Higher is generally better. | Chain diagnostics outputs |
fit |
MCMC fitting path (fit.method: mcmc). |
R/run_from_yaml.R |
fit_map / optimise |
MAP optimisation path (fit.method: optimise). |
R/blm.R, R/hierarchy.R, R/pooled.R |
hierarchical |
Model class with grouped random effects (`(term | group)` syntax). |
identifiability check |
Diagnostic check for baseline and media term correlation risk. | R/diagnostics_report.R |
kpi scale |
Business-outcome scale used for reporting. For log-response models this is back-transformed from model scale. | docs/modelling/response-scale-semantics.md |
lognormal_ms |
Positive-support prior family parameterised by mean and standard deviation on the original scale. | R/prior_schema.R |
MAP |
Maximum a posteriori point estimate from optimisation. Not a posterior mean. | fit_map paths |
MCMC |
Markov chain Monte Carlo posterior sampling. | rstan::sampling paths |
Pareto-k |
PSIS-LOO reliability diagnostic for influence of observations. | loo_summary.csv outputs |
pooled |
Model class with structured pooling over configured grouping variables. | R/pooled.R |
posterior draw |
One sampled value from the posterior distribution. | get_posterior() outputs |
pre-flight checks |
Guardrails and model/data compatibility checks run before fitting. | R/pre_flight.R |
prior_only |
Fit mode sampling only from priors, excluding likelihood learning. | prep_data_for_fit.* interfaces |
PSIS-LOO |
Pareto-smoothed importance-sampling leave-one-out approximation. | R/diagnostics_report.R |
QG-1 to QG-7 |
Canonical release quality gates for lint, style, tests, package check, runner smoke, and docs build. | docs/internal/quality-gates.md |
response scale |
Scale used inside the fitted model (identity or log). |
docs/modelling/response-scale-semantics.md |
Rhat |
Convergence diagnostic comparing within- and between-chain variance. | Chain diagnostics outputs |
runner |
YAML/CLI execution layer around core DSAMbayes APIs. | scripts/dsambayes.R, R/run_from_yaml.R |
run_dir |
Output directory used by a runner validate/run execution. | docs/runner/output-artifacts.md |
staged layout |
Structured artefact layout with numbered folders (00_ to 70_). |
docs/runner/output-artifacts.md |
Stan cache |
Compiled model cache location, typically under XDG_CACHE_HOME. |
install/setup docs |
SMAPE |
Symmetric mean absolute percentage error metric used in fit summaries. | R/stats.R |
time-components |
Managed time control features, including holiday-derived regressors. | R/holiday_calendar.R |
tscv |
Time-series selection artefact prefix for blocked CV outputs. | 50_model_selection/tscv_*.csv |
warmup |
Initial MCMC iterations used for adaptation and excluded from posterior draws. | fit.mcmc.warmup |
Hill transform |
Saturation function spend^n / (spend^n + k^n) used in budget optimisation response curves. k is the half-saturation point, n is the shape parameter. |
R/optimise_budget.R |
atan transform |
Saturation function atan(spend / scale) mapping spend to a bounded response. |
R/optimise_budget.R |
log1p transform |
Saturation function log(1 + spend / scale) providing diminishing-returns concavity. |
R/optimise_budget.R |
adstock |
Media carry-over transform that spreads a spend effect over subsequent periods via geometric decay. Applied as a pre-transform in the data, not estimated within DSAMbayes. | Formula transforms |
conditional mean |
Bias-corrected back-transform for log-response models: exp(mu + sigma^2/2). Default in v1.2.2 for fitted_kpi(). |
R/fitted.R |
Jensen's inequality |
Mathematical property that E[exp(X)] != exp(E[X]) when X has non-zero variance. DSAMbayes avoids this bias by applying exp() draw-wise before summarising. |
docs/modelling/response-scale-semantics.md |