Method and editorial standards
How 457 dream records were chosen, checked, tagged and matched, and what this site cannot do.
What gets in
The corpus has 457 records from 84 works, spanning roughly 1950 BCE to 1920 CE (records with a stated date).
A work enters only with a rights-evidence link and the jurisdictions checked; the build fails if either is missing. Every work and its evidence is listed on the Sources page.
How excerpts are checked
Each excerpt should appear verbatim in its source text. A script re-checks excerpts against the public texts, and a duplicate audit stops the same dream appearing twice in different wording.
Tags and uncertainty
Themes, imagery, feelings, settings and story patterns come from editors and from keyword rules the editors wrote; no language model decides them. Every record also carries an uncertainty level saying how firm its date, place or attribution is.
Current spread: 53 low, 231 medium, 173 high. High usually means legend, retelling or an unknown date.
Translations and AI
Chinese titles and translations are editorial drafts. 122 records carry a note that the Chinese was drafted with AI and has not been checked line by line by a person. Originals are always shown separately from translations, and the original governs.
Overall review status of the corpus: prototype — editorial review still in progress.
How matching works
The baseline matcher runs in your browser and compares the excerpt you chose with each record’s words and editor-assigned features, limiting how much one source can fill the results. The optional on-device language model adds a semantic signal.
Results show labelled features two records share. They are not a percentage of story similarity and not an interpretation of your dream.
What it cannot do
The corpus leans toward cultures and periods whose texts survive in quantity, so it does not represent “human dreams” as a whole. Contemporary research records are mostly short reports. It cannot tell you what your dream means.
Corrections
If you find a mistake, an attribution problem or a rights concern, tell us through the feedback form.