Research notes

Gap fill 2: TCF B1–B2 listening and reading items, coded by rule, and what fills Q20–29

Research date: 2026-11-04. Scope: TCF Canada listening (CO) and reading (CE), B1–B2 band. Goal: NCLC 7, which means CO at least 458 and CE at least 453.

What I did. I coded every B1–B2 item I could read from FEI-written sources with one written codebook (Section 2), so a second coder can repeat it:

• 136 reading items from all 17 TV5MONDE practice series. FEI says it wrote and validated these items. I read the document text, stem, options and key for each one.
• 136 listening items from the same 17 series. The stem is spoken and TV5MONDE gives no transcript, so I could code only the options, the key and the clip length.
• 8 official FEI/CIEP sample items at B1 level or above.
• 20 items from one circulating "real-order" paper that can be read without logging in: 10 listening and 10 reading, both at Q20–29.

Fully coded totals: 127 FEI-written items at B1 level or above (119 TV5MONDE reading items at the B1–B2 positions plus 8 official), 17 TV5MONDE boundary items (position 12), and 20 circulating items.

Limits, stated up front.

• The web-search budget was already used up before this run's first search, so I worked only from known URLs and the browser pane. More searching needs a follow-up message.
• FEI's example page sits behind an automated bot check. I did not try to get past it.
• No publisher (CLE International, Didier, Hachette FLE) currently lists a TCF book with a readable extract (Section 1).
• Every code is mine: one coder, with no second-coder check.

Confidence tags. [H] official or directly counted from the source. [M] counted, but resting on my coding or on a mapping assumption. [L] unofficial, single source, or provenance unknown.

0. Answer first: what dominates Q20–29

1. Reading Q20–29 is dominated by short informative press articles. The questions ask what the text says, or what you can infer from it.

   • In the 68 TV5MONDE items at the B2-band positions (series positions 8–11, mapped in Section 2.1), document types were:
     • informative press article: 38 (56%);
     • cultural presentation or review: 14 (21%);
     • portrait of a person: 9 (13%);
     • notice or offer: 4;
     • first-person testimony: 2;
     • message: 1.
   • Answer mechanics in the same 68 items:
     • inference: 26 (38%);
     • detail: 25 (37%);
     • purpose: 10 (15%);
     • change of plan: 5 (7%);
     • opinion: 2 (3%).
   • The circulating paper's actual Q20–29 shows the same pattern: 5 of 10 texts are informative articles and 6 of 10 answers are inferences. [M]
   • Sources: TV5MONDE series; circulating reading paper.

2. Opinion questions are a minority at Q20–29 in both skills. This contradicts the earlier [M] claim in exam-analysis.md, which lists "opinion or attitude" as the main listening mechanic at Q20–29 and "the author's position" for reading.

   • Reading, B2 band: 2 of 68 items under the strict rule, 4 of 68 counting every stem with an attitude word.
   • Listening, B2 band: only 1 of 68 TV5MONDE option sets is a set of attitudes, and 1 of 10 circulating Q20–29 items is an opinion question.
   • What dominates instead: "What do we learn / what is it about" (21 of 68 reading stems) and purpose questions (13 of 68). [M]

3. Wrong options mostly re-use words from the text. In the 68 B2-band reading items:

   • 90 of 204 distractors (44%) re-use a word, name or picture from the document but say something the document does not say;
   • 54 of the 68 items (79%) have at least one such distractor;
   • next come wrong combinations of true elements (swaps) at 21%, plausible but unmentioned options at 19%, and contradictions at 12%.
   • The key is usually a paraphrase that uses a more general word: vêtements (clothes) becomes l'habillement (the clothing sector); mise à l'eau (the launch of a boat) becomes "a stage in building a boat"; changé d'adresse (changed address) becomes a déménagé (moved).
   • So the cheapest skill for this band is spotting an option that re-uses the text's words. [M]

4. Reading gets harder through inference, not through longer texts.

   • From the B1 band to the B2 band, detail answers fall from 57% to 37% and inference answers rise from 20% to 38%.
   • Detail plus inference stays flat: 76% of B1-band items and 75% of B2-band items.
   • Portraits of a person fall from 33% to 13% of documents, and informative articles rise from 29% to 56%.
   • Every one of the 136 TV5MONDE texts was a single short passage of one to three paragraphs.
   • No key depended on reading or comparing a number. [M]

5. Listening Q20–29 in the one paper with transcripts is mostly dialogues and interviews about everyday life, not radio news.

   • Document types: 4 dialogues, 5 interviews, 1 radio report.
   • Mechanics: purpose 4, detail 3, inference 2, opinion 1.
   • In the 68 TV5MONDE B2-band clips, the option sets break down as:
     • statements about a topic: 23 (34%);
     • names of things, people or topics: 19 (28%);
     • aims written as infinitives: 19 (28%);
     • a person's actions: 6 (9%);
     • attitudes: 1.
   • Clips average 61 seconds in the B2 band, against 55 seconds in the B1 band. They reach about 65 seconds at positions 10–12. [M/L]

6. Change-of-plan is a small but distinct mechanic.

   • The key depended on a stated change in 10 of 136 reading items (7%): 4 in the B1 band and 5 in the B2 band.
   • A further 8 documents contained a change but were coded under another mechanic: in 5 nothing used it, in 2 it related to the key, and in 1 only a wrong option used it.
   • When the key depends on a change, a wrong option often gives the old state: 6 such options across the TV5MONDE and circulating reading items. [M]

7. FEI re-uses material across forms, at shifted positions.

   • The same subject appears in two different series, with a different text or question, 4 times:
     • a news story about renting fake police cars, at position 5 in series 5 and position 11 in series 8;
     • one author-photographer's book;
     • the reform allowing courses taught in English at French universities;
     • the Expolangues language fair.
   • One story (anti-sun clothing sold in Brazil) appears to be used in both skills.
   • 2 listening items are exact repeats, moved one position later in a later series.
   • In the circulating paper, the recording at position 25 announces itself as question 24.
   • So learn item types, and expect positions to drift by about one. [H for what was observed]

8. The benchmark arithmetic checks out. The community points grid is unofficial [L].

   • At the earlier targets (95% on Q1–19, 80% on Q20–29, 30% on Q30–39), the grid gives 432–465, typically 439–451.
   • At 9 of 10 on Q20–29 it gives 453–486, typically 460–472.
   • With one B1 miss and only 8 of 10 on Q20–29, you need 4 of the last 10 correct to reach 458.
   • Use at least 9 of 10 (90%) as the Q20–29 drill target (Section 6).

9. Publisher material is not available.

   • Hachette FLE's adult exam catalogue lists DELF and TEF books only.
   • CLE International's TCF/TEF filter returns only ABC TEFAQ.
   • Didier FLE's "TCF" collection page shows 0 products.
   • The prep books named in the CIEP manual date from 2002 to 2014 and are not readable online.
   • The 680-question TV5MONDE bank is the largest FEI-written practice source. [H]

1. Sources used and what each gave

2. Codebook: the rules a second coder can repeat

2.1 Unit and level band

• Unit: one multiple-choice item, made of a document, a stem, 4 options and the key.
• The band rule for TV5MONDE. Each series has 15 items, and difficulty is assumed to rise through the series. TV5MONDE says the TCF listening test rises in difficulty (strategies page), and the series copy the test. Neither TV5MONDE nor FEI labels the level of single items.
• Mapping rule. Series position p maps to the TCF Canada position (p − 0.5) × 39/15 + 0.5. So:
  • positions 5–7 map to Q12–18 (the B1-band proxy);
  • positions 8–11 map to Q20–28 (the B2-band proxy);
  • position 12 maps to Q30 (coded separately as the boundary).
• Cross-check. Multiply the community grid's cumulative level shares (A1 4, A2 6, B1 9, B2 10, C1 6 and C2 4 items out of 39) by 15. B2 then covers positions 7.3 to 11.2, which gives the same 8–11.
• Sensitivity check. I re-ran every count with B2 = positions 9–12 (Section 3.1). [M]
• Listening structure. In all 17 TV5MONDE listening series, positions 5–15 are always "document + spoken question". Positions 1–4 hold 20 picture items, 41 question-and-reply items and 7 document items. TV5MONDE therefore packs the A-level families into 4 items, and position 5 already behaves like a TCF Canada Q11+ item. [H, counted]
• Circulating paper. Positions are taken as labelled. Its level labels differ by skill: listening puts B1 at 12–18 and B2 at 19–29; reading puts B1 at 11–19 and B2 at 20–29. [L]

2.2 Document type (reading; apply the first that fits)

Listening document types:

• DIAL: two or more speakers in an everyday exchange.
• INT: an interview, with a host or journalist and a guest.
• REP: a radio report or news item read by one voice.
• MONO: a presentation by one speaker.

2.3 Question stem family (by the surface frame of the question)

2.4 Answer mechanic (the six required codes, applied in this order)

1. RA, required action. The stem asks what someone must do, should do or is advised to do (doit, faut, conseillé, recommandé, demande de).
2. PU, purpose. The key states the writer's, speaker's or document's intent (to announce, propose, inform, invite), or the goal of an action, project, organisation or device.
3. OP, opinion. Two conditions must both hold:
   • (a) the stem contains an attitude word: pense, trouve, avis, opinion, point de vue, position, attitude, sentiment, juge, apprécie, aime, plaît, préfère, regrette, reproche, critique, dénonce, déplore, se plaint, s'oppose, réagit, satisfait, déçu, étonné, fait rêver, passionne;
   • (b) the key states an evaluation, feeling, stance or grievance.
   • A key that merely names the liked object ("drawing", "a nomadic life") is not OP.
   • I changed this rule once while coding: the first version used condition (a) alone. The final rule was applied to all 136 items. Under (a) alone, OP would be 12 of 136 instead of 6.
4. CP, change of plan. The document gives an earlier state, plan or rule and then replaces it, and the key is the new state or the change itself. Signals: annulé, reporté, déplacé, finalement, désormais, aujourd'hui versus autrefois, plutôt que, ne… plus, il y a encore un an… depuis.
5. IN, inference. No single sentence states the key. Getting it needs one of: combining sentences, decoding a metaphor or irony, generalising, or naming the topic. Naming the topic is marked "IN-g".
6. DE, detail. The key paraphrases one sentence or clause.

DE versus IN is the least reliable boundary, so I also report DE + IN merged.

2.5 Distractor pattern (each wrong option; apply the first that fits)

Flags on the document:

• CONC: a concession pivot (mais, pourtant, néanmoins, par contre) where the key follows the pivot or a wrong option uses the conceded part.
• CHG: a change is present in the document. k = it relates to the key but the item is coded under another mechanic; d = only a wrong option uses it; u = nothing uses it.

2.6 Listening items without transcripts (TV5MONDE)

The option set is coded by the form that at least 3 of the 4 options share:

Clip length comes from the audio file's metadata and includes the spoken question.

3. Reading results

3.1 TV5MONDE reading, 136 items, by band (main mapping and sensitivity check)

Source: the 17 series (index); lot IDs and positions are in Appendix A. [M]

Reading the table.

• Moving from the B1 to the B2 band, the document shifts from "a person's story" to "an informative article", and the answer shifts from detail to inference.
• Opinion stays rare until position 12.
• The sensitivity mapping does not change the ranking.

3.2 By position (TV5MONDE reading, 17 items per position)

3.3 Question stems by band (TV5MONDE reading)

No single stem family reaches a quarter of the B2 band. "What do we learn / what is it about" (LEARN + TOPIC, 21 of 68) and purpose questions (PURP, 13) together cover half of it. [M]

3.4 Circulating paper, reading Q20–29 (actual positions; provenance unknown)

Source: InfinLume reading Paper 01. All items are paraphrased. [L/M]

Totals (10 items):

• Documents: INF 5, OPI 3, CUL 1, LIT 1.
• Mechanics: IN 6, DE 1, RA 1, CP 1, PU 1, OP 0.
• Distractors: S 10, E 8, R 5, U 4, P 2, O 1.

Conflict with the TV5MONDE bank. This paper has 3 opinion texts and 1 literary text at Q20–29. In the 68 TV5MONDE B2-band items, opinion texts are 0, with 1 at position 12. TV5MONDE builds its texts from short TV5MONDE media pieces (FAQ), so it probably under-represents argued texts. Train both.

3.5 Official reading samples at B1 level or above

Source: FEI 2022 manual pp.53–56. The documents are images, described in tcf-co-ce-items.md §4.1. [H]

4. Listening results

4.1 Official listening samples at B1 level or above, with transcripts

Sources: FEI 2022 manual pp.49–51 and 56; CIEP Québec 2018 manual pp.31–32 and 41. [H]

4.2 Circulating paper, listening Q20–29 (actual positions; transcripts look machine-made)

Source: InfinLume listening Paper 01. [L/M]

Totals (10 items):

• Documents: DIAL 4, INT 5, REP 1.
• Mechanics: PU 4, DE 3, IN 2, OP 1.
• Distractors: E 9, S 9, R 6, U 5, P 1.
• Several Q20–29 items here (bookshop, weather, travel agency) read like B1 service dialogues. This fits position 20 being the B1/B2 seam, and fits this paper's own label of B2 starting at Q19.

4.3 TV5MONDE listening, 136 items (options and clip length only)

Source: the 17 series (index). Coding rule in Section 2.6. [M/L]

Mean clip length by position:

• Options that signal a change (for example "she resumed training", "he stopped dancing and founded his company", "she decided to change direction", "the offer is cancelled"): 7 of 136 option sets, 4 in the B1 band and 3 in the B2 band.
• Exact repeats: two items in series 13 come back word for word one position later in series 15.

5. What dominates Q20–29, and where earlier claims disagree

Disagreements with existing reports.

1. exam-analysis.md (listening, rank 4) gives Q20–29 "what recurs" as "opinion or attitude, reason, who said what" [M]. The counted data do not support opinion as dominant. Reason (cause) and purpose items are common; opinion is 1–3%.
2. The same file (reading, rank 3) gives "the author's position" as a recurring Q20–29 stem. It is rare in the FEI-written bank (2 of 68 B2-band items) but more common in the circulating paper (3 opinion texts out of 10, though none had an opinion question). I keep it as a secondary type.
3. Band boundaries: the circulating listening paper labels B2 as Q19–29 and B1 as Q12–18, against the community grid's Q20–29 and Q11–19. So Q19–20 is a seam either way.

6. What to drill for Q20–29, ranked by work-to-reward

Points: Q20–29 is worth 21 points per item and 210 per skill under the unofficial community grid (paliertcf; see tcf-co-ce-items.md §2.1) [L]. Frequencies come from Sections 3–4.

Daily drill sets for this band.

• Reading: 10 items a day at 90 seconds each.
  • The 68 TV5MONDE B2-band items are positions 8–11 of all 17 series (index), plus Q20–29 of the circulating papers.
  • Mark each item's stem family and distractor codes after answering.
• Listening: 10 clips a day of 60–70 seconds, played once.
  • Use TV5MONDE positions 8–12. Training mode allows replays, but do not replay; the simulator plays once.
  • Then use circulating Q20–29 for transcript review.
  • Before each clip, use the option-set form to predict the question type:
    • aims mean a purpose question;
    • entities mean an identity or topic question;
    • statements mean a "what do we learn" or inference question.
• Benchmark: at least 9 of 10 (90%) on Q20–29 drills.
  • The grid arithmetic (checked): 95% on Q1–19, 80% on Q20–29 and 30% on Q30–39 gives 432–465, typically 439–451. That is below CO 458 and usually below CE 453.
  • At 9 of 10 on Q20–29 the range is 453–486, typically 460–472.
  • With one B1 miss and 8 of 10, reaching 458 needs 4 of 10 on Q30–39.
  • Each extra Q20–29 item is worth 21 points. [L, unofficial grid]

7. Item register (so a second coder can check)

Appendix A. TV5MONDE reading: 136 items

Series n° to lot ID: 1=58, 2=57, 3=56, 4=45, 5=52, 6=53, 7=47, 8=42, 9=43, 10=44, 11=50, 12=55, 13=60, 14=66, 15=68, 16=69, 17=70.

URL pattern: https://apprendre.tv5monde.com/fr/tcf/test-dentrainement-au-tcf?tcf_lot_id=LOT&competence=CE. Stems are paraphrased. "D" lists the 3 wrong-option codes.

Appendix B. TV5MONDE listening: option-set form and clip length (seconds), positions 5–12

Same URL pattern with competence=CO. Format: position:form:seconds.

Notes on the table.

• Exact repeats: series 13 positions 6 and 8 reappear as series 15 positions 7 and 9.
• Judging from its options, series 14 position 9 appears to use the same anti-sun-clothing story as reading item 1-8.

8. Limits, conflicts and open questions

• Single coder. No second-coder agreement was measured.
  • The weakest boundaries are DE versus IN (so DE + IN is also reported) and OP (reported under both rules).
  • Recoding a 20-item sample (for example, items 1-5 to 3-8) would give a quick agreement check.
• Level mapping is an assumption. TV5MONDE does not label item levels. Its results grid is a PDF that downloads as a file, so it was not used (FAQ). Shifting the B2 band by one position leaves the ranking unchanged (Section 3.1).
• TV5MONDE is not the live test. Its items are FEI-written but built from TV5MONDE media, and TV5MONDE says simulator questions are never used in a real TCF (support page). The topic mix leans towards culture, travel and the French-speaking world, and argued texts are probably under-represented.
• TV5MONDE listening has no transcripts, so its document types, stems and wrong options could not be coded. The listening document-type claims rest on 10 circulating items and 5 official items.
• The circulating paper is unofficial (provenance unknown). Its transcripts look machine-made, and its positions can be off by one: one recording announces "question 24" at position 25. Its CO level labels differ from its CE labels.
• Not reached: FEI's examples page (bot-protected); publisher extracts (none listed); TV5MONDE PDF booklets (they download as files; not opened). Web search was unavailable for this run; a follow-up with search could look for archived extracts of ABC TCF (CLE 2014) or 250 activités (CLE 2003).
• The grid is unofficial. FEI says the conversion weighs item difficulty and differs by version (CIEP Québec 2018 p.20). The 90% benchmark is a planning rule, not an FEI rule.

Sources

• TV5MONDE TCF practice index and series pages: https://apprendre.tv5monde.com/fr/tcf (lot IDs in Appendix A)
• TV5MONDE FAQ (item authorship, results grid): https://apprendre.tv5monde.com/fr/article/foire-aux-questions-du-tcf
• TV5MONDE booklets page (680 questions, FEI validation): https://apprendre.tv5monde.com/fr/article/les-livrets-dentrainement-au-tcf-r-avec-tv5monde
• TV5MONDE TCF strategies (difficulty rises; question heard after the audio): https://apprendre.tv5monde.com/fr/article/strategies-pour-le-tcf
• TV5MONDE support page: https://tv5monde.freshdesk.com/en/support/solutions/articles/61000311553-how-can-i-practise-for-the-tcf-with-tv5monde-
• FEI Manuel du candidat, Version D (2022), mirror: https://www.unicaen.fr/wp-content/uploads/2023/02/Manuel_du_candidat_TCF-1.pdf
• CIEP Manuel du candidat TCF pour le Québec (2018): https://elf.uqac.ca/wp-content/uploads/2018/10/manuel-candidat-tcf-quebec1.pdf
• CIEP Manuel du candidat TCF Canada (©2018): https://www.afsf.com/exams/tcf/Manuel_du_candidat_TCF_Canada.pdf
• FEI examples page (blocked): https://www.france-education-international.fr/test/exemples-epreuves-tcf
• InfinLume practice library and Paper 01: https://www.infinlume.com/tcf-canada ; https://www.infinlume.com/tcf-canada/listening/test-101?mode=review ; https://www.infinlume.com/tcf-canada/reading/test-101?mode=review
• OPAL series (login required): https://tcfcanada.opal-ca.com/courses/comprehension-ecrite/ ; https://tcfcanada.opal-ca.com/courses/comprehension-ecrite/quizzes/comprehension-ecrite-serie-160/
• tcf-canada.ca (paid): https://tcf-canada.ca/comprehension-ecrite/
• Hachette FLE adult exams catalogue: https://www.hachettefle.com/adultes-diplomes-et-tests/
• CLE International certifications (TCF/TEF facet): https://www.cle-international.com/recherche?sort_by=field_parution_date&items_per_page=50&search=&facets_field_taxo_product_univers%5B167%5D=167&facets_field_taxo_product_certif%5B41%5D=41
• Didier FLE TCF collection: https://didierfle.com/categorie-produit/collections/tcf/
• Community points grid (unofficial): https://paliertcf.com/tcf-canada
• Earlier team reports: research/tcf-co-ce-items.md ; exam-analysis.md