Methodology · version 3.0
Tags should be useful, traceable, and correctable.
DaoReads uses structured automation to build the catalog, but it never treats a model guess as an unquestionable fact.
- Source the bookGoogle Books is the primary metadata source. Curated Open Library records are the fallback, and optional Hardcover data enriches ratings.
- Run two grounded tag passesEach pass must choose from the public vocabulary and copy an exact supporting quote from the source description. Unknown labels and quotes absent from the source are rejected.
- Write the trope viewThe model writes a short explanation of the relationship and plot dynamics rather than republishing the generic synopsis. It may use only the supplied source description.
- Publish only complete profilesA book enters discovery pages and the sitemap only after it has a grounded summary and at least three supported trope tags.
- Keep the trailEvery book–trope edge stores its evidence URL, source, confidence, prompt version, and verification time.
Review official source notes separately
We also collect official publisher and author pages, preserving synopses and content notes as separate source documents. A reviewer checks the book identity and edition before accepting a document, then reviews each quoted fact separately. Accepted facts appear under Verified source notes with a link to the original source. A source revision or withdrawal hides dependent facts until they are reviewed again.
This source archive is separate from the automated trope-tagging process. Collecting a new source does not automatically change a book’s tags, heat, or warnings. Publication dates identify a specific edition unless first publication has been verified separately.
The evidence archive today
436 verified facts across 218 current source documents, covering 151 books.
Counts include only currently displayed, validated facts on discoverable book profiles. Historical revisions and documents without displayed facts are excluded. Every accepted fact keeps its source link, exact quote, and review record.
What the labels mean
Confidence measures how strongly the available description supports a tag; it is not a reader rating. Spice level runs from 0 to 5 and may remain blank when evidence is insufficient. Content notes require exact source quotes and agreement between both automated passes. A missing note does not establish that sensitive content is absent. Legacy heat and warning claims without recorded evidence are held for review.
The heat scale
0: explicitly no on-page sex. 1: kissing only. 2: closed door or fade to black. 3: non-graphic on-page sex. 4: explicit sexual content. 5: frequent explicit sex. A description calling a book “sexy” or “sizzling” is insufficient. Unknown heat stays unverified.
Limits and corrections
Blurbs omit plot details, editions differ, and subjective tropes have fuzzy boundaries. Every book profile includes a correction form. Suggestions enter a private review queue and never change a profile automatically. A completed human review replaces the full current tag set, records human as its source, and keeps an auditable batch record; uncertain records may be quarantined instead of published.