A briefing for A2

Nothing here is adopted. The dictionary is closed at 300 roots. This page sets out what has been measured about the next 300 so that the decisions are made on evidence rather than on the order a wordlist happens to be in.

The standing plan is A2 at around 600 roots — double A1. That number is not what this page is about. What the next 300 should be made of is.

1. The thin families are thin for the wrong reason

The record said Chinese and Japanese contribute almost nothing because their words are short and the two- and three-letter space is full. That is true of the fifteen words already taken — hao, kan, lai, cang, yuki, nomu, suru and eight more — and false about the languages.

That count read eleven until September 6, 2026, which is the Sino-Tibetan figure alone; the sentence is about Chinese and Japanese, and the seven words it names are four of one and three of the other. Eleven plus four is what the record should have said.

Root length, measured across all 300:

Three of those rows were stale until September 5, 2026 — Austronesian’s 5+ column read 81 against 84, Turkic’s 61 against 65, Latin/Romance’s 46 against 47 — and two averages with them. The rows counted roots by reach, so a root added anywhere moves several of them at once, and nothing was recounting. The table is generated now.

Sino-Tibetan is the only family in the dictionary with no root of five letters or more. But modern Mandarin vocabulary is overwhelmingly two syllables, and so is much of Japanese — which is the length the four-letter rule wants. Ten ordinary words run through every rule in check.py: seven passed. mianbao, laosi, dianua, tomodaci, genki, gako, tabemono. The working is in balance.md.

So the door is open. The thinness is not something the phonology did. It is that the borrowing reached for the classical monosyllables and stopped.

2. What a balance floor would cost

If A2 is to fix the imbalance rather than inherit it, the cost is countable.

Together that is up to 100 of the 300 new roots — a third of the entire A2 batch — before a single word is chosen for what it means. The real figure is lower, because one word can reach several families at once and these count each separately, but the order of magnitude is right.

That is the trade-off in one line: A2 can be balanced, or it can be chosen freely by meaning, and it cannot be both. This is a decision, not a calculation.

3. What the writing has asked for

Sixteen gaps are on the list: thirteen found by trying to write something and failing, one by a sweep of sentence shapes, and two found a third way — by translating a sentence somebody asked for, and by somebody asking about kinship. They are recorded with the sentence that stopped in each case. The first 7, in the order they were found: pain, love as a noun, to miss someone, the clock, a word for then, standing up, a coin as against an amount.

That enumeration was printed with no count in front of it until September 6, 2026, directly under the word sixteen, so seven items read as the whole list.

Three of those seven are about how a person feels or where a person hurts. No thematic wordlist predicted that, and it is the strongest evidence available about where 300 roots are actually thin. Two more were claimed and withdrawn on checking, which is why the list is audited rather than appended to.

This section’s heading said thirty-four until September 5, 2026 and the paragraph beneath it said two, which is what the checker had been holding all along. A heading is the one line on a page that nobody rereads.

3b. Two roots have never been used

Being taught is not being used. Counted on September 4, 2026: 55 of the 300 roots appeared in no text at all — met once in a lesson’s word table, drilled in that lesson, and never written again.

What they were is the finding. Not obscure nouns but the joints of the language: o (or), kadar (as … as), kurang (less), lasim (must), berapa (how many), kaifa (how), fikir (to think). A rule can be exercised by a text while the word that carries it is not.

Three texts were written for them: 16 took twenty-one, 17 twenty-seven, and 18 the last five that could be written. Two are still unused, and check.py recounts the number so this paragraph cannot drift:

kadang, madad.

Neither can be written at all, which is why they are the two that are left. madad’s class is undecided, so any sentence using it would answer the open question by accident — the question has no evidence because the word has no use, and it can have no use until the question is answered. kadang is the same shape: where a frequency adverb stands is open, so the word cannot appear in a sentence without settling it.

The open questions are not only holes in what can be said. They are holes in what can be practised, and after two texts written to sweep the vocabulary they are what is left.

3c. The other method: a checklist, and what it says

Every gap in section 3 was found by writing. Text 19 showed the blind spot in that method — a wall and a floor are exactly what a thematic list would have contained, and nineteen texts went past them. So the other method was run once, kept separate, and written down: a hand-made list of 274 ordinary A1 concepts by domain, each looked up in the English index by check.py.

167 of 273 are present. A hundred and six are not.

Both halves of that sentence are checked now. The second was not, until September 6, 2026: the checker’s number parser stopped at ninety-nine, so a figure written out as a hundred and six was invisible to it, and this is the count the whole page turns on. A parser that cannot spell a number cannot test a sentence that spells it — and the shape of that failure is the one this repository keeps finding, a map returning nothing for what it does not know rather than saying it does not know.

One of those was counted present by an adjective. clean stands in the checklist twice — in actions, where it is the verb, and in qualities, where it is what a washed room is — and the English index answers clean with safi, which is the quality. The language has gasil, to wash, and no verb for cleaning. From September 5, 2026 a concept in actions counts as present only if the index holds it as to X, which is what an action is.

And work is set aside from both sides of the fraction. It is the only other action the index answers with a bare entry, and the reason is rabota’s undecided class. It was exempt from the rule and counted present until September 6, 2026 — which is the answer the exemption was written to avoid giving, because counting it present asserts that rabota is the verb and counting it missing asserts that it is not. Neither is neutral. It is now left out of the numerator and the denominator alike: 273 concepts counted, the 274th named and waiting. The totals moved by one each way, and the page and the checklist now say the same thing about the same word, which they did not.

The sets a language has to close

Kinship was not an isolated hole. Checked on September 4, 2026, four more closed sets are missing entire, and none was on the checklist either:

Three of the five may not need roots at all, and that is the interesting part. Chinese names its weekdays by number — 星期一 is week-one — and its months the same way, and a language built on economy should probably do the same: din du for the second day, mensis tri for the third month.

But it cannot, yet. Numbering a day or a month is naming an ordinal — first, second, third — and ordinals have no form. That open question, filed since text 6 wanted “on the eighth day”, turns out to block two whole vocabulary sets as well as a sentence. It is the first case where answering one open question would remove the need for nineteen roots: seven weekday names and twelve month names.

That figure read “a dozen” until September 5, 2026, while numbers.md and the ordinals briefing both said nineteen — and the briefing cited this page for it. Three pages, two numbers, and the one being quoted was the odd one out. The seasons are the set left out of all three: four more that a number could name, and mausim tri for autumn is not what anybody calls it, so they are counted as roots the language would still need.

The compass is the opposite: four directions that no number can build, and no root for any of them.

Kinship’s absences are thirteen of twenty-seven, and four of the thirteen are the hole. The people domain is the largest block of absences on the checklist, and the checklist now splits it: three of the thirteen are already sayable by possession alone — a grandmother is the mother of my mother — six are the no-gender rule doing what it was chosen to do rather than vocabulary that is missing, and four are roots the language does not have. Reading a settled decision as an absence is what the block was doing, and it made the domain look three times emptier than it is.

Clothing is two of nine — a shirt and a shoe — and is the thinnest domain in the language by a distance; nothing composes a coat. Kinship is fourteen of twenty-seven present, which is the same measurement as the paragraph above and not a second one: fourteen present, thirteen absent, and only four of those thirteen a root the language would have to add. This page said fourteen of twenty-seven and thirteen of twenty-seven four paragraphs apart, as though they were the same count disagreeing with itself, until September 6, 2026. The founder noticed the domain on September 4, 2026 by asking how to say uncle; the checklist had listed eighteen people-words and none of the nine he asked after, which is the second time a hole in the checklist has been found by somebody simply wanting to say something. The everyday adjectives are close behind: there is no slow, no angry, no tired, no hungry, no dark and no wet, so a person can be sick or sad but not tired, and a night can be cold but not dark.

The two methods agree where they overlap. Seven of the thirteen gaps found by writing — slowly, to stand up, cheap, the clock, then, but and a wall — are on this list too, which is the check that matters: the checklist reproduces what writing found and adds 99 more. What it cannot do is say which of the 106 matter, and that is exactly what writing does say. Neither list replaces the other and this page keeps both.

Those two figures read “about eighty” and “the eighty-four” until September 5, 2026, and the overlap read five rather than seven — then, but and a wall were missing from it and fluent was in it, which is one of the two gaps that came from a question rather than from writing. All three numbers are the checklist’s own arithmetic and are checked as such now.

4. What A2 content is blocked on

Three decisions are waiting, and lesson and text material for A2 will hit them immediately. Two more were settled on September 3, 2026 and no longer block anything:

• stress — settled: the second-to-last syllable
• the imperative — settled: a verb with no subject
• a mark for a name (briefing)
• “want to be” plus an adjective (briefing)
• rabota’s class — one word, twenty places

And on the open list, before, after, until and the ordinals were each hit by real writing: the letter was arranged into short sentences to avoid the first three, and text 6 could not say “on the eighth day”.

5. The vocabulary is thin in the wrong places

A1 is a level about concrete life — food, clothes, the body, the house, the weather, shopping, health. Counted by the dictionary’s own thematic groups, this one is not shaped like that.

64 of the 300 roots — 21% — are qualities and ideas. Health has two words, bimar and ilac; there is no pain and no fever. Clothing has two, a shirt and a shoe. Play has two. Feelings have three. Food has fourteen.

That is the shape of a dictionary built by writing texts about ideas, which is what happened, and it is the opposite of what a beginner needs first. The next roots should be concrete, and those first seven — pain, love as a noun, to miss, the clock, then, to stand up, a coin — all sit in exactly those thin groups.

Which raises the number itself

The front page has said 300 is enough for A1 because things can be described with what is already there. The stronger version of that claim — that compounding does the rest — was checked and is false. Estimates for A1 in a natural language run to about 500 words, and Basic English covers daily life in 850.

So 300 looks like an A1 core rather than a finished A1, and the honest reading is that A1 completes somewhere near 500 with the extra two hundred spent on concrete vocabulary. That would move A2 as well. It is a founder’s decision and this page does not make it; it records that the number now rests on a weaker argument than the one it was chosen under.

What is measured and what is judgement

Measured: the length table, the seven passing shapes, the +48 and +52, the gaps and where they were found, and the theme table — which check.py regenerates from the dictionary, so it cannot drift.

Judgement, and the founder’s: whether to set a family floor and where, whether the feelings are a deliberate theme for A2 or a coincidence of the data points, whether A1 finishes at 300 or nearer 500, and whether 600 is still the right number. Note that the reason once given for a small dictionary does not hold: mesin ambil foto korpo says X-ray with no new root, but it is a sentence, not a compound, and the language forms no compound words at all.