Natural-Language Lesson Rules: L1–L7

Alongside the numbered programming courses P1–P13 governs, this site also generates standalone lessons in real human languages — Japanese, Hungarian, French. Those lessons follow a completely separate set of rules, language_rules.md's own L1 through L7, because a language lesson genuinely isn't shaped like a course chapter: there's no fixed chapter count, no Fundamentals/Intermediate/Advanced track, and a single topic (say, "question words") gets requested and generated on demand, in whichever language the learner actually wants it in.

L1 — The Languages Table & Learner Profiles

A small table tracks every language currently in use, each with its own level, its own per-country accent colour, and a short learner profile (goals, current knowledge, why they're learning it) that shapes how every lesson in that language gets written:

That "no kanji yet" detail isn't decorative — it directly shapes how a Japanese lesson gets written. Every kanji that does appear still needs a romaji reading right beneath it, since the learner can't yet read it unaided.

L2 — The <language>:<topic> Prompt Convention

A prompt like hungarian:question words means exactly what it looks like — generate a lesson, in that language, on that topic. The colon is the whole signal that distinguishes this from P5's course-chapter convention: hungarian:question words is a language lesson; rules1-4 (no colon) is a course chapter. No outline step is needed either way — a lesson generates directly the moment it's requested.

L3 — A PDF for Every Single Lesson

Unlike P4 (which only fires once, on a course's final chapter), every language lesson gets its own PDF immediately, via a reusable builder script that detects the language straight from the filename prefix and applies that language's own accent colour and flag automatically.

L4 — The Two Lesson Sub-Formats

Not every language lesson is shaped the same way. A plain topical lesson ("directions," "food," "question words") uses sub-format A: a six-card dialog grid demonstrating the topic in natural conversation, followed by a fixed 20-word vocabulary table. A lesson specifically about a verb tense or conjugation pattern (Hungarian's past definite tense, for instance) uses sub-format B instead — full conjugation tables and a pattern box come first, with the dialog grid and vocabulary table still following after.

L4a layers one more detail on top of either sub-format: every language gets its own single accent colour, derived loosely from its flag — Japanese red, Hungarian green, French blue, all visible in the language cards above — kept consistent across every lesson in that language so a glance at a page identifies which language it's in before reading a word of it.

L5 — Introducing a Brand-New Language

A lesson request in a language that isn't already in the L1 table doesn't generate silently. Per L5, the correct response is to ask first — is this genuinely a new language to add, or a one-off/typo? What's the learner's actual level? What accent colour fits, without clashing with a language already in the table? Only once those questions are answered does the language get added to L1 and the first lesson actually generated.

L6 — Verb Deep Dive Lessons

vdd:hungarian:lenni or vdd:japanese:to be requests something narrower and deeper than an ordinary topical lesson — a full treatment of one specific verb, built entirely around L4's own sub-format B (conjugation tables across the language's own relevant tenses, plus mini-dialogues using that verb and a labelled vocabulary list of related verbs). When a VDD is requested across several languages in sequence, each one is generated and then paused on, rather than all being produced back to back unprompted.

L7 — Song Lessons, and the Copyright Line That Splits Them in Two

song:title:artist:language produces a full, line-by-line lesson from real song lyrics — but only for lyrics that are genuinely copyright-free (public domain, traditional, or original material). songvocab:title:artist:language exists specifically for everything else: it never reproduces the lyrics themselves in any form, extracting only individual, decontextualized vocabulary words instead — which is legally safe even from a commercially copyrighted song, since isolated words and their translations aren't the protected creative expression the way the lyrics' actual sequence and phrasing are.

Coding Challenges