How-To Guides
Task-oriented guides for preparing evidence, running analyses, and reviewing outputs.
Task-oriented guides for preparing evidence, running analyses, and reviewing outputs.
The modelling workflow starts from a long-format tracker. Use one row per extracted effect estimate.
| Evidence | Default treatment |
|---|---|
geo_test |
Pool comparable effects on log_relative scale. |
bls |
Pool comparable effects on percentage_point scale. |
mmm, pa2, ROI, CPA, CPO, iROAS |
Treat as triangulation unless harmonised upstream. |
Resolve or explicitly accept:
(evidence_type, client, metric, period) candidates;ci_type or ci_level;p_value_sidedness=two_sided;The tracker and config enforce metric names and units, but they cannot prove that studies answer the same business question. Before pooling, confirm that population, measurement window, study design, and estimand are comparable enough for a pooled effect to mean something.
Use a YAML config when an analysis should be repeatable and reviewable.
For local repository use:
Config runs stop on directional and not_recommended readiness by default, and failed
sampler diagnostics are blocking by default.
Override these only for explicit exploratory work. Record the reason in project.notes
or the downstream analysis review.
data.tracker and outputs.directory are resolved relative to the config file when
they are not absolute paths.
Successful execution does not mean the result is reportable. Always review
run_status.reportable and the supporting artefacts before using pooled summaries.
Config runs write a compact artefact set intended for analyst review.
| File | Review question |
|---|---|
analysis_report.md |
What is the run status, readiness, and headline result? |
run_status.json |
Is reportable true? Why did the run complete, warn, block, or fail? |
readiness.csv |
How many comparable rows were eligible, and are they enough? |
diagnostics.csv |
Did sampler diagnostics pass configured thresholds? |
prior_diagnostics.csv |
Do priors imply warnings or implausible prior-predictive ranges? |
effect_preparation.csv |
Which rows were included, transformed, excluded, or scenario-based? |
ppc.csv |
Are individual studies out of line with in-sample posterior predictive checks? |
run_status.json may report:
completedcompleted_with_warningsblockedfailedA completed run is not automatically reportable. Treat reportable: false as a hard
prompt for analyst review before any downstream presentation.
directional or not_recommended.Small-sample Bayesian meta-analysis can be prior-sensitive and uncertainty-sensitive. For serious reporting, sensitivity is part of the result.
When prior_specs are omitted, the package uses a scale-aware grid:
log_relative: regularising, default, weakpercentage_point: regularising, default, weakReview prior_sensitivity.csv for pooled mean, probability positive, interval width,
diagnostics, and deltas versus the default prior.
Rows with source-derived uncertainty keep that uncertainty. Rows with
uncertainty_scenario use the configured low, medium, or high model-scale standard
error.
Scenario assumptions live in:
Report when reasonable sensitivity settings change the substantive conclusion.