Contents
Corpus boundary
Only two primary comparison objects are in scope: the screenplay comparison record and the pinned Greek Odyssey TEI covering Books 1–24. Other translations are semantic controls only. No web evidence was used. The screenplay attribution, completeness and public-reuse rights are not established by this comparison.
Screenplay reconstruction
The comparison uses OCR-derived visual-line units. Columns were processed separately to avoid cross-column mixing, and reading order was checked separately from association claims. Stable research-generated line and block IDs and source-version aliases preserve the comparison’s citations. These IDs and generated scene identifiers are research locators, not published pagination or authorial numbering.
OCR blocks join adjacent visual lines only within one page, column, and generated scene when the vertical gap is no more than 25 pixels. The full auditable unit remains the visual line. A visual line below OCR confidence 50 is U even if its neighboring block is classified.
Association threshold
Rules are bounded by reading-order ranges and lexical/event signals, but they generate candidates only. A–D requires an explicit keyed panel adjudication with exact evidence-line IDs and Greek targets. When close reading corrects or narrows a rule's proposed locus, the adjudication must preserve the proposed targets and supply a written target_override_reason; target drift without that reason is rejected. N1 and N2 mean no specific association located under this method, not proof of absolute non-relation. A departure can coexist with an association when one screenplay block mixes a Homeric episode with a novel detail.
Greek context
Every accepted association rule points to exact book.line targets. For every target, the builder emits up to 25 extant base verses before and after, clips at book boundaries, and merges overlapping windows. Greek is parsed namespace-aware from TEI element content, never from physical XML lines or regex. Raw IDs and raw order are retained. The five-entry anomaly ledger prevents silent interpolation or global reordering.
Coverage contract
The current comparison covers 4406 eligible visual lines and 2436 OCR blocks. Every eligible line is exactly one of A/B/C/D/N1/N2/U. The public tables supply public comparison reference codes; screenplay cues and the transcript are not republished. This coverage statement does not establish source authenticity or republication rights.
Interpretive limits
Pattern rules only generate candidates. A positive edge enters the dossier solely after a keyed panel adjudication names its exact evidence-line IDs, strength, targets, rationale, and reviewer roles. Unadjudicated candidates remain U. These decisions do not establish authorial intent or exclusive dependence. The screenplay attribution and completeness are not independently authenticated. The dossier does not supply a full history of scholarship across all 24 books.