Opportunities Trust Infrastructure

Status: Implemented into schema.

9 new tables + modifications:

scholars — identity layer with ORCID, institution, expertise annotation_runs — 3 new columns: scholar_id, method, publication_ref tokens — stripped to positional anchor (removed form, reading, sign_function, damage, reading_confidence) token_readings — competing transliterations per token position 4 evidence tables with real FKs (token_reading_evidence, lemmatization_evidence, sign_annotation_evidence, translation_evidence) 3 decision tables with real FKs and supersedes chains (token_reading_decisions, lemmatization_decisions, translation_decisions) 4 new migration steps added

Schema Update Summary

Changes

9 new tables, 3 new columns on annotation_runs, 5 columns moved off tokens into token_readings Table count: ~27 (v1) → ~36 (v2 pre-trust) → ~45 (v2 with trust) The 9 trust tables are structurally repetitive but simple (5-7 columns each, all with real FKs)

Tradeoffs

Tokens store reading data? No — always join to token_readings. Single source of truth; avoids staleness. Chose over cached consensus on token. Polymorphic FKs? No — separate tables per entity type. Real FK enforcement for a multi-user scholarly system. Chose over one flexible evidence_links table. Scholar identity? Dedicated scholars table. Prevents "J. Smith" vs "John Smith" at scale. Chose over freetext scholar_name. Evidence/decision scope? 4 evidence + 3 decision tables (7 total). Database-enforced integrity; self-documenting. Chose over 2 polymorphic tables.

What This Enables

Every reading, lemmatization, sign annotation, and translation can now answer:

Who said this? How did they determine it? What evidence supports it? Who chose it as consensus, and why? What did it replace?