Interscript
GitHub ↗
← All docs

Coverage and gaps

Interscript ships a registry of authority-backed transliteration systems. This page states what is covered today, what is not, and how the gap list is tracked. Every claim here links to a source you can check.

What is covered

The map registry contains 428 systems drawn from the maps repository — each an executable, test-carrying transliteration map with a named authority and source document, 12,337 embedded tests from those documents, all green. Browse them on the map catalogue; every system has its own page with its description, source document, scripts, mapping counts, invertibility class, a live in-browser preview, and its full rule chart.

Invertibility is measured per system and recorded in the coverage ledger: systems are classed as lossless one-to-one (reversible), lossy many-to-one, or context-only. The counts are re-derived from the maps themselves, not asserted.

What is not covered yet

The open queue is tracked as work orders in the models repository’s campaign ledger. The remaining systems, by source:

  • ISO — ISO 9 non-Russian Cyrillic coverage check, ISO 259 Hebrew letter table, ISO 7089/7098 Chinese pinyin, and the remaining ISO 15919 Indic scripts. The source documents are not yet in the project’s reference collection; acquisition is owner-held.

  • ALA-LC — the batch campaigns landed through 2026-11-09 (Brahmic/ SE-Asian, Arabic-script, Cyrillic, East Asian, and misc-script systems; 93 ALA-LC maps in the registry). The systems listed in the campaign ledger without a source document in the collection — moghol, bashkir, chuvash, udmurt, komi, yakut/sakha, N’Ko, tod-oirat — are recorded there, not encodable until a source is acquired.

  • ICAO and UNGEGN — the source-document diff against the registry is complete (2026-11-08), and the UNGEGN editions are acquired into the reference library.

Neural lines with open targets

The neural (learned) model lines and their open quality targets are tracked in the same ledger. Where a line’s next step depends on a license or a dataset only the project owner can grant — additional Hebrew vocalization data, commercial-license Urdu dictionaries, the SEAlang and full Chuon Nath sources for Khmer — that gate is recorded as owner-held and the line waits.

How to check any of this