Phase 1 · Architecture & blueprint · v0.1

The French Confusion Masterbook

The TEF Canada guide to the words, structures and expressions that get misread — and the clue that settles each one in about three seconds.

Levels A1 → C1 Parts Part 0 + I–XVIII Confusion classes 42 Status Phase 1 · architecture only Entries written 3 of ~430

📘 What this page is This is the blueprint, not the book. It sets out the taxonomy, the part order, the entry template, the page furniture and the three trainers, and it ships three finished entries so the format can be judged before ~430 more are written. Everything on this page has been through the three-pass check; nothing carrying a NEEDS REVIEW flag is on it.

💡 The unit of the book A page is not a word. A page is a contrast — two readings a sentence could have, and the single feature that decides which one is live. venir de + INFINITIF is a different entry from venir de + LIEU, and the entry is about the thing that separates them.

Part 0, then the eighteen parts, in the order they are meant to be read

The order is not alphabetical and not by grammar topic.
It runs from how to read a sentence at all to the confusions that only appear inside the exam's own subject matter.
Family counts are the planned size of each part, and they add up to the working total for the book.

Why this order rather than a grammar syllabus

A grammar syllabus groups by form: all the pronouns together, all the tenses together.
That is useful for a course and useless here, because a student misreading je viens de partir does not know they are in the tense chapter.
The order below groups by where the misreading starts: first the tiny words that carry meaning, then the words with several jobs, then the verb families, then the structures, then meaning-level traps, then sound, then the exam's own worlds.
A student can enter at any part and the entries still make sense.

Part 0 is short on purpose. It installs the four-step parse — find the verb, look at what follows it, check for a chunk, then decide — and every later entry refers back to it.

The confusion map — 42 classes in six super-classes

A class is a reason a French sentence gets misread.
A chapter is built out of several classes.
The brief supplied 31 starting categories, which consolidate into the 24 unmarked classes; the exhaustiveness audit added the 18 marked ADDED, each argued for in the architecture file.
Filter by super-class, or show only what the audit added.

What the audit added, and why

Endings you cannot hear, dropped ne, numbers by ear, intonation questions, audible participle agreement, cognate mismatch — the original list treated listening as one category. Six distinct failures live in there, and only one of them is liaison.
Auxiliary changes the meaning (je suis sorti / j'ai sorti la poubelle) — usually taught as an agreement rule, which is not what causes the misreading.
Reference tracking — a pronoun three sentences back is a reading-comprehension failure, not a grammar failure, and TEF reading tests it directly.
Nominalisations and administrative French — l'embauche, le préavis, un justificatif: the student knows the verb and does not recognise the noun on the form.
Canadian vs Hexagonal usage — the exam is TEF Canada; the student lives with both.
Register shift, elliptical French, discourse markers — the sentence is grammatical and still unreadable.
Word order and scope of focus particles — un ancien collègue vs un collègue ancien; what même is attached to.
Production false friends, kept separate from reading false friends: the two fail in opposite directions.

Three finished entries, in the final template

These are specimens: one RULE, one COMMON PATTERN with a real exception, one family of four verbs where everyday speech and careful writing disagree.
Judge the format here — the fields, the order, the amount of English, the honesty of the hedging — before the rest are written to match.

👀 How to read an entry The Student brain box shows the misreading before it is corrected. The 3-second clue is the one thing to look at. Everything below is evidence for that clue: examples, a contrast set, one exam-shaped sentence, the production trap, and — where the truth is not tidy — a nuance box that says so.

◆ On certainty vocabulary Every entry is labelled RULE, COMMON PATTERN, PREFERENCE, REGISTER, CONTEXT-DEPENDENT, EXCEPTION or REGIONAL. The words always, never, only and must are allowed inside a RULE block and nowhere else. Where usage is split — amener with an object, for instance — the entry says it is split and says where each version belongs, instead of calling one of them wrong.

Page furniture — the eight boxes

Eight box types, each with one job and one colour.
A reader should be able to tell what kind of information a box holds without reading it.
In print they carry the same left rule and the same label; on screen they are the components below.

📘 RuleThe statement of the pattern, with its certainty label. Blue. Never contains an example on its own — the examples sit under it.

💡 The 3-second clueThe one feature to look at, phrased as an instruction: look at the word right after de. Green. One per entry, never two.

⚠ Production trapWhat the student writes or says instead, struck through, with the correct version under it and a sentence on why the wrong version is tempting. Red.

◆ NuanceThe exception, the split usage, the thing that would be a lie if it were left out. Amber. If this box is missing from an entry that needs it, the entry fails the audit.

👀 Student brainSees / thinks / look again / why it fails / actual meaning. Grey. Capped at one per spread — used where the misreading is worth staging, not on every page.

🎯 TEFOne sentence in the shape the exam actually uses, with its section and situation named. Navy.

🎧 By earWhat happens to this structure when it is spoken at speed. Violet. Part XVII is built from these; elsewhere they appear where the spoken form is the trap.

🔑 Memory hookGold. Optional, and left out rather than forced — a bad hook is worse than none.

The formula strip

Set in mono, so the eye reads it as a shape rather than as a sentence.

venir (conjugated) + de + INFINITIF → a moment ago venir (conjugated) + de + LIEU → where you are from

Type scale

Role	Face	Used for
Display & French	Literata	Headwords, every French example, entry titles
Interface & English	Inter	Explanations, translations, labels, buttons
Formulas	JetBrains Mono	Structure strips, clue instructions, schema

French is always in the serif and always carries its English line. That rule holds in every box, every table, every trainer — including inside the machine.

The clue machine

The reel stops on a French sentence with the deciding fragment highlighted.
Two readings are offered; only one is what the sentence says.
You have five seconds, and answering inside three is worth double — because recognition that takes ten seconds is not recognition, it is translation.

The buttons under the machine are paylines: switch a confusion family off to stop it coming up, or leave one on to drill it alone.
Sentences you get wrong are put back into the pool twice as often.
Three correct in a row opens a bonus round — three sentences from one family, cleared in a single pass for the full bonus.

💡 Why a slot machine The exam gives you one pass over a text and no dictionary. Practising with unlimited time trains a skill you will not have on the day. The reel supplies the one thing a static exercise cannot: you do not know which family is coming, so you cannot pre-load the answer.

Scores stay in this browser. Nothing is uploaded.

Spot the clue

Same skill, one step earlier: before choosing a meaning, find the word that decides it. Click the single word in the sentence that settles the reading. The answer comes with the reason and the English.

👀 What this trains Most misreadings are not vocabulary gaps. The student knows every word in the sentence and is looking at the wrong one — at the verb they recognise instead of at the preposition that changed its meaning.

Where to start — a 20-question placement

Twenty questions, one or two from each part.
Every answer comes with its reason straight away.
When all twenty are done, the page ranks the parts you missed and gives you a reading order.
This is the short version; the book's own diagnostic runs 40 items and separates the four tiers.

Build notes

Database schema — one record per entry

The PDF and this page are both generated from the same records. Nothing is written twice; a correction is made once.

id slug, unique part I … XVIII chapter X.1 size FULL | COMPACT | ROW headword the contrast, not the word base dictionary form(s) gloss one line of English grammar what the structure is classes[] A1 … F3 confusedWith[] thinks the misreading, in the student's voice fails why the misreading is reasonable clue { look, a, b } formula mono strip certainty RULE | COMMON PATTERN | PREFERENCE | REGISTER | CONTEXT-DEPENDENT | EXCEPTION | REGIONAL register cefr skills[] score { freq, risk, tef, prod } / 20 examples[] { fr, en, note? } every fr carries an en contrast[] { fr, en, tag } tef { fr, en, context } trap { wrong, right, why } nuance listening memory related[] status VERIFIED | VERIFIED_NUANCE | NEEDS_REVIEW sources[] what was checked, never quoted

Build gate: the generator refuses to emit any record whose status is NEEDS_REVIEW, whose examples are missing an English line, or whose certainty label is absent. A flagged entry cannot reach a student page by being forgotten.

Three passes, every chapter

Linguistic audit — is the statement true, is the hedging right, is the exception present, is the certainty label the correct one?
Native-naturalness audit — would anyone say this sentence, in this register, in this situation? Invented sentences that exist only to show a rule are cut.
Student-misinterpretation audit — read the finished entry as a student who does not yet know the answer. Anything that could be read as a stronger claim than intended gets rewritten.

Twelve pre-publication audits

#	Audit	How
1	Accuracy of every rule statement	Independent reviewer
2	Naturalness of every French example	Independent reviewer
3	Register labels correct	Independent reviewer
4	Exceptions present where they exist	Independent reviewer
5	Translations faithful, not literal	Editorial
6	TEF claims justified	Independent reviewer
7	Every example has an English line	Automated
8	No NEEDS_REVIEW record in the build	Automated
9	No duplicate entry across parts	Automated + editorial
10	Cross-references resolve	Automated
11	Absolutes only inside RULE blocks	Automated
12	No promotional copy anywhere in the book	Automated

Anti-duplication

Forty confusable pairs already exist on this site in the P.A.I.R.E. trainer.
They are imported as records and cross-referenced — not rewritten into new prose.
Where a masterbook entry covers a pair that already has a trainer, the entry links to it and the trainer links back.
See open decision 5.

Five decisions still open

Audio. Part XVII cannot be finished honestly without recordings. Ship the book without it, or hold the part back to a later release?
Scope of Part XVIII. Exam-world vocabulary could run to a book of its own. Cap it at the confusions, or let it become a glossary?
Print length. ~430 families at the current density is a large book. One volume, or split at Part XII?
Canadian usage. Marked inline as a note, or gathered into its own section at the back?
The existing 40 pairs. Import as records and cross-reference (current plan), or rewrite them to the new template?

Sources consulted for the specimen entries

Le Robert · Larousse · CNRTL · Académie française (Dire, ne pas dire) · OQLF, Banque de dépannage linguistique · Grevisse, Le Bon Usage · Government of Canada terminology records.
Nothing is quoted; sources are checked, then the explanation is written from scratch.
The student edition carries no per-page citations — the list lives in the back matter.

Where the files live

confusion-masterbook/00-ARCHITECTURE.md the governing spec tools/masterbook/10-arch.js parts + classes tools/masterbook/11-entries.js entry records tools/masterbook/12-banks.js trainer banks tools/masterbook/30-engine.js rendering + trainers tools/masterbook/build.js writes public/tef/masterbook.html tools/masterbook/test.js the automated audits

French With Guri · The French Confusion Masterbook · Phase 1 blueprint v0.1 · English explanations, French examples, every French line translated.