# Multilingual / Low-Resource Ledger — Bills Draft (v0.1)

> **13 bills + 6 meta-costs + 3 escape gates**, ★ at 4, 7, 10.

| # | Bill | What gets paid | ★ |
|---:|---|---|:---:|
| 1 | Low-resource-language sample-density audit | Per-language training-corpus size disclosed; minimum threshold for clean evaluation. | |
| 2 | Tokenizer fertility per language | Tokens-per-word disparity (English vs low-resource). | |
| 3 | High-vs-low-resource gap audit | Quantified gap between top-10 and bottom-10 languages. | |
| 4 | **★ Low-resource deep-learning parity** | ≤500K-sentence language reaches ≥80% of high-resource performance. Predicted empty. | ★ |
| 5 | Cross-domain transfer (news → biomedical → legal → conversational) | Per-domain coverage across languages. | |
| 6 | Translation-vs-generation decoupling | MT capability separated from generation capability. | |
| 7 | **★ Cross-script generalization** | Same policy passes Latin + CJK + Arabic + Devanagari + Brahmic with ≤10pp gap. Predicted empty. | ★ |
| 8 | Dialect / register preservation audit | African-American English / Indian English / Singlish / Brazilian-vs-EU Portuguese / Maghrebi-vs-MSA Arabic. | |
| 9 | Post-training-language-drift audit | Does instruction-tuning erode multilingual base-model competence? | |
| 10 | **★ Universal multilingual coverage at frontier scale** | ≥150 of 200 Flores languages above 60% BLEU. Predicted empty. | ★ |
| 11 | Vendor-self-eval independence | Stanford HELM-Multilingual / MasakhaneNLP / SEACrowd / Aya-Eval reproduction. | |
| 12 | Held-out post-2024 language-benchmark construction | Flores-200 → Flores-Plus, AmericasNLP rolling refresh, Masakhane held-out. | |
| 13 | Anti-saturation construction (Multilingual MMLU, XTREME-R) | Anti-contamination by design, rolling refresh. | |

## Six meta-costs: pre-2024 / English-only-evaluation / Latin-script-only / single-task / single-domain / implementation-specific.

## Three escape gates: G1 / G2 / G3.

## Iteration plan (8 sweeps)
- 901: Frontier multilingual model cards (NLLB-200, Aya-Expanse, Llama 3.1/4 multilingual, Qwen 2.5/3, Gemini, Claude, Mistral Saba, Apertus)
- 902: Flores-200 / Flores-Plus / XTREME-R / Masakhane benchmark construction + audits
- 903: Cross-script generalization papers (Brahmic, Arabic, CJK, indigenous languages)
- 904: Low-resource MT (NLLB / MaLA-500 / Cendol / Aya / SEACrowd)
- 905: Dialect / register / post-training drift papers
- 906: Tokenizer-fertility / vocab-multilingual papers
- 907: Stanford HELM-Multilingual + independent multilingual audits (Common Voice, MasakhaneNLP)
- 908: Multilingual safety / refusal-rate-by-language audits + negative results