Evidence disclosure. No public material was collected and no identifiable example was retained. All records below are abstract synthetic fixtures.
Decision summary
Define the decision, population, permitted source boundary, period, inclusion rules and coding plan before selecting a tool. Keep automated classifications provisional until their validity is checked against a human-coded sample.
State the research decision
Replace "What are people saying?" with a bounded question: which recurring concerns in a declared public source set should inform the next service-information review? This turns retrieval into a defensible sampling and coding task.
Declare the source boundary
List included and excluded sources, access method, dates and relevant platform terms. Public accessibility does not erase privacy, expectation or research-ethics questions. Minimise collection and avoid publishing identifiable examples without a valid reason and review.
Version the query
| Record | Synthetic value | Reason |
|---|---|---|
| Research question | Which fictional delivery concerns recur in the example source set? | Connects collection to one decision. |
| Include | Generic phrase "example delivery" within a fabricated fixture. | Demonstrates a documented term. |
| Exclude | Jobs, investment discussion and duplicated fixtures. | Reduces known irrelevant material. |
| Version | Draft query version | Prevents silent query changes. |
Build a codebook
Define each label with inclusion, exclusion, mixed and uncertain examples. Preserve "uncertain" rather than forcing every record into positive, neutral or negative. If automated labels are used, compare them with a permitted human-coded sample and publish disagreement counts.
Synthetic record
The example record contains only the sentence "Example delivery timing was unclear." A demonstration coder assigns information clarity and mixed/uncertain sentiment. It represents no person, account, post or platform.
Plan independent review
Have a second trained researcher code a declared subset without seeing the first label. Record agreement, disagreement and resolution. Review should improve the codebook; it must not be used to hide difficult cases.
Methods
Preregister the research question, period, permitted sources, query version, deduplication, language, sampling, codebook, reviewer process, retention and publication limits. Keep raw material restricted and publish only content the release review permits. Record the decision owner and the precise condition that would cause the team to revise the query or stop collection.
Record retrieval failures and removals without converting them into sentiment observations. Train coders on the same versioned definitions, blind the independent subset where practical and retain disagreement before resolution. Publish counts for included, excluded, duplicated, uncertain and reviewed records so readers can see how the final sample was formed.
Limitations
Source populations are selective. Retrieval coverage can be unknown, deleted content may be missing and automated labels can fail on context, humour or mixed views. Findings describe the declared sample, not all consumers.
Sources
Evidence scope. No live post, account, platform export or participant record is a source for this methods article. A future study must document permitted source classes, access method, query version and rights basis without exposing identifiable material unnecessarily.
- Use a permitted, documented query and coding exercise on authorised public material.
- Have a second researcher review a declared subset.
- Redact or abstract identifiable examples where appropriate and complete rights and privacy review.
Update status
No material updates have been recorded since publication. A later query, codebook or coding change will identify the affected version, recalculate relevant counts and retain a visible change note.