Add France: full-text curriculum source, Français literacy track, French history/civics + cross-language crosswalks

[dahbiahmed]

Add France: full-text curriculum source, Français literacy track, French history/civics + cross-language crosswalks

What this adds (three layers, three commit groups)

1. A new 🟢 full-text curriculum source — the French programmes d'enseignement for cycles 1–3 (école maternelle → CM2/6e), as in force for the 2026-2027 school year:

• 3,179 standards extracted at statement level from the official BO texts (2020–2026), in French, including the renewed 2024–2026 programmes (français, mathématiques, EPS, histoire-géographie, sciences et technologie, langues vivantes, EMC, EVAR, plus the 2026 maternelle programme).
• 2,446 topic↔standard links across 923 of the original 1,590 micro-topics, added under a conservative genuine-match rule.

2. A native Français subject — 252 new micro-topics: French-as-first-language literacy as a parallel track to English, in 9 domains mirroring the English literacy architecture where the science is shared and diverging where the languages diverge (décodage syllabique, orthographe grammaticale, conjugaison…). Every topic is distilled from the French standards (carries its fr-programmes-2026:* codes), has evidence criteria, an assessmentPrompt, and sits in the prerequisite graph (614 new edges, including cross-domain chains like grammaire → orthographe agreement).

3. French history & civics growth of existing subjects — 54 History topics (two new domains: French History: Origins to Revolution / Revolution to Today) and 26 PSD topics (Citizenship & la République). Shared periods (Rome, Greece, world wars) reuse the existing topic nodes — e.g. Gaule romaine takes Daily Life in a Roman Town as a prerequisite instead of duplicating it.

Plus data/crosswalks.json (new file, schema, and validator support — additive, optional): 139 same-skill-different-language links pairing Français ↔ English literacy topics, tagged transfer: high|partial. Rationale: mastery is observed per language, but transfer is real and asymmetric (Cummins' interdependence) — comprehension strategies and composition transfer nearly fully; decoding partially; conventions barely. Two nodes + a typed link preserves per-language mastery state while letting an app treat mastery in one language as a strong prior in the other. The same pattern extends to any L1 (Spanish, German…).

validate.mjs passes (structure, referential integrity, cycles, checksums, crosswalk integrity). centrality for new topics is approximated (normalized descendant count) pending your canonical recomputation.

Licensing

The ministry publishes these texts under Licence Ouverte / Open Licence (Etalab) 2.0 — attribution-only, commercial reuse permitted, no share-alike (education.gouv.fr: "Sauf mention contraire, tous les contenus de ce site sont sous licence etalab-2.0"). Same shape as the existing UK OGL v3 source, so full text ships. Required notice added to PROVENANCE.md.

Design decisions (feedback welcome)

1. Codes are minted for this dataset (C2.Fr.CP.Lect.3 = cycle 2, français, CP, lecture, item 3) — the French programmes define no official statement codes, mirroring how uk-nc-2013 codes were minted.
2. Go-forward texts only. France is mid-way through a rolling reform (2026 texts apply to CP/CM1 first, reaching remaining classes by 2027-2028). This ships the new texts and omits superseded transitional programmes (e.g. Questionner le monde). Per-subject BO references are in each standard's data.bo and in PROVENANCE.md.
3. Granularity follows the source. The 2024–2026 French programmes are annualised and much more granular than the 2013 UK NC — hence 3,179 standards vs 1,117 for the UK. Statements repeated per year in the source are kept per year (as the source declines them).
4. Mapping is deliberately sparse where curricula genuinely differ: Anglo-centric history topics link only where the same period/concept is taught; French primary defers plate tectonics, formal algebra, and negative numbers to cycle 4, so those topics carry no French links.

The English ↔ Français question

The taxonomy's English subject is English-as-first-language literacy; the French programmes contain French-as-first-language literacy (Français) and English-as-a-foreign-language (langues vivantes). No pair is the same construct, so links follow a skill-transferability rule:

• Linked to Français: language-agnostic literacy skills — comprehension strategies, text composition, oral discussion, print/genre concepts, vocabulary-learning strategies. A learner who can summarize or structure a narrative carries that skill across both languages.
• Deliberately NOT linked: English-specific mechanics — phonics, spelling, English grammar/punctuation. Décodage graphème-phonème is not the same teaching content as English phonics, even though it is structurally parallel.
• Linked to langues vivantes only where a topic matches early EFL functional content (greetings, question words, simple descriptions) — a small set.

Net effect: 135/286 English topics carry French links. This PR now also ships the cleaner long-term answer — the native Français track (layer 2) — with the crosswalk expressing the English↔Français relationship explicitly at skill level. The English-topic→Français-standard links from layer 1 remain as a complementary, coarser signal; happy to re-cut or drop either layer if you'd rather keep a single mechanism.

Provenance of the extraction

Extracted from the official PDFs published by éduscol / education.gouv.fr (two sciences PDFs had no text layer and were OCR'd). Extraction and topic mapping were LLM-assisted with human-defined rules and spot-check review; happy to share the intermediate artifacts or re-run against different granularity choices if the maintainers prefer.