Ask a question the corpus can answer

Start with a denominator and a release#

POD 1.6.0 reports 215,769 canonical organs, 143,880 organs with stop accounts and 3,604,579 public stop rows. The ratio 143,880 / 215,769 is about 66.7 percent. It describes the proportion of that release’s canonical organ records with stop accounts. It does not measure the percentage of organs worldwide that have been documented or the completeness of each recorded disposition.

The unit matters just as much as the denominator. A stop row is a published data row, not necessarily one unique physical stop across all sources and historical states. A source account, canonical instrument, state, location and workshop are different counting units. State which one your query uses, how duplicates are handled and how a join can multiply observations.

Three productive research directions#

Stop nomenclature is especially revealing because a familiar label can conceal different construction or usage. Preserve the original lexeme alongside a proposed normalized concept. A pitch designation or division can disambiguate a local account; a regional or historical convention may still require expert interpretation. A corpus can expose patterns and candidate comparisons without proving that all occurrences describe acoustically equivalent pipes.

Read a map as a map of documentation too#

A dense cluster on a map may reflect a well-documented collection, an active source provider or easier geographic reconciliation. A sparse area may reflect a gap in source availability rather than a historical absence of instruments. Keep source and regional coverage visible, and distinguish an instrument’s location from the location of a source, workshop or repository. Unknown coordinate accuracy should remain unknown.

Temporal coverage has a similar problem. An exact-looking year can come from a broad estimate, a later catalogue or an event that concerns a rebuild rather than initial construction. Preserve date precision and event roles. Excluding records with uncertain dates may make a chart cleaner while systematically removing less-documented regions or periods; report that selection rather than treating the remaining cohort as neutral.

A reproducible query plan#

• Identify the exact POD release and distribution used. Retain its published identity and the relevant local file digest.
• Write the research question and unit of observation in a sentence. Define the cohort, date interpretation and inclusion rules.
• Inspect the supplied schema before joining tables. Check whether source accounts, states or repeated relations multiply rows.
• Report the starting population, missing values, exclusions and final analyzed cohort. Keep unresolved identities and mappings inspectable.
• Save the query or analysis script, vocabulary version and resulting table. Cite the corpus release as well as the notebook that explains the method.

What a defensible result looks like#

A useful result might describe the distribution of a source spelling among a defined set of stop accounts in one release, with the number of accounts lacking date or location information. That is narrower and more reproducible than a statement about the universal prevalence of a stop. A follow-up can inspect representative source fragments, compare alternative mappings and test whether one provider dominates the pattern.

Navigator supports discovery of promising relations and cases. A fixed dataset distribution supports a recoverable analytical cohort. Use both: move from exploration to an explicit query, and from an aggregate back to individual evidence. The result then explains not only what the corpus contains, but how the available documentation supports the historical question.