Before asking a flow for review themes: check the table, denominator and evidence rows

[LMSlay]

Disclosure: I maintain TubeText. This note was prepared with AI assistance. It shares a review-analysis checklist, not a tested Langflow flow, component or integration.

If you already have a permitted app-review CSV and want a flow to propose themes, loading the file and establishing the denominator are different jobs.

The Langflow 1.13.x Read File documentation distinguishes Structured Content → Table from Raw Content → Message for a single CSV with advanced parsing disabled. Advanced parsing is not the CSV route. That output-type distinction does not establish which rows a downstream model actually receives or whether it sees the entire sample.

Here is the preparation and review checklist I would use:

1. Keep an unchanged source export and inspect a separate working table. Decide which records are actual reviews. App-summary, AI-insight or error records must not silently become review counts. Record exclusions such as missing text or dates. If you add a local evidence ID, retain its mapping to the original row; it is an analyst-added reference, not a product-provided universal review ID or proof of unique authors.
2. Fix the comparison before prompting for themes. Record the time window, language/filter choices, retrieval cap and actual eligible review count for each app. A capped “newest” sample can cover different periods for different apps. Inspect the rows and columns at the downstream boundary rather than assuming the input file equals the model's context.
3. Treat proposed themes as a draft codebook, then verify against source rows. Check each proposed example using the evidence ID and original text. Count a review at most once per theme; a review can have multiple themes, so percentages need not sum to 100%. Report theme mentions among the defined sample—not the proportion of all users affected, unique authors, or a confirmed product defect.

The existing Google Play review-theme comparison guide gives the manual coding method and a clearly fictional worked example. It can be used with an existing permitted export; using a TubeText Actor is not required. The guide is not a Langflow import or end-to-end compatibility test.

This is about checking the evidence behind a model-assisted comparison, not a claim that Read File generates themes, adds evidence IDs, deduplicates authors or automatically preserves an analysis sample.