SRDP Reading Comprehension Task Generator (B1 / B1+)

• Gapped Text (Recommended — the core SRDP MM format)
• Matching Statements to Texts (B1+ variant)

If the user selects MM at B1 level, ALWAYS use Gapped Text — do not ask.

Question 3 — Topic & Style: "Any preferences for the topic or text style?"

• A specific topic (e.g., "social media and teenagers", "extreme sports")
• From the SRDP Themenpool (I'll choose)
• A specific text type (e.g., "personal blog", "magazine report", "interview-style")
• Surprise me

Question 4 — Creativity vs. Exam-Closeness: "How closely should this mirror an official SRDP task?"

• Very close — indistinguishable from official material (Recommended)
• Somewhat creative — authentic feel, but with a fresher angle on the topic
• More creative — still exam-valid, but with a more engaging or unusual text

If the user has already specified level, format, or topic in their initial message, skip asking about those — only ask what's still unclear.

Overview: What You Generate

Each reading comprehension task consists of TWO parts delivered together:

1. A reading text — an authentic-sounding English text on the requested topic
2. A task — one of four SRDP-approved formats with items, answer key, and distractor rationale

The Four SRDP Reading Task Formats

IMPORTANT — MM sub-type selection: The DEFAULT MM sub-type at BOTH levels is Gapped Text (continuous text with gaps, phrase list A–K, 2–3 unused distractors). This is the core SRDP MM format.

At B1+ level, an OPTIONAL additional sub-type exists:

• Matching Statements to Texts — see references/mm-format.md, Sub-type 2

NEVER use Matching Statements to Texts unless the user explicitly requests it. When the user says "MM" without specifying a sub-type, ALWAYS generate Gapped Text.

Step-by-Step Workflow

Step 1: Read the Appropriate Reference File

Before writing anything, read the reference file for the chosen format. This is non-negotiable — the reference files contain the reverse-engineered construction principles that make the difference between a passable task and an authentic one.

Also read:

• references/b1-vs-b1plus.md to calibrate text and item difficulty to the correct level
• references/item-validity-firewall.md to internalize ALL known item validity traps before constructing any items. This prevents generating items that have defensible alternative answers.

Step 2: Write the Reading Text

The text is the foundation. A poorly written text makes good items impossible. A text that "sounds AI" invalidates the entire task, no matter how good the items are.

THE AUTHENTICITY PROBLEM — READ THIS CAREFULLY

Previous iterations of this skill produced texts that were technically correct but felt "stale," "generic," and recognizably AI-generated. The specific problems were:

1. Telling instead of showing: "I felt excited but nervous" instead of showing the excitement through actions, dialogue, or specific sensory details
2. Too-perfect paragraph structure: Each paragraph covering exactly one topic with visible transitions ("However, I soon realized...", "Another benefit is...")
3. Generic voice: No humor, no vulnerability, no surprising details, no personality
4. Vocabulary feels "chosen for level": Words are correct but lack the organic messiness of real writing — no phrasal verbs, no colloquial asides, no register shifts
5. Wikipedia tone (especially for B1+ informational texts): Reads like an encyclopedia entry rather than a magazine article written by a human with a perspective
6. Emotional declarations: Characters state their feelings instead of revealing them
7. Missing specificity: Generic names, generic places, no cultural references, no sensory details, no humor

These are NOT acceptable. The text must feel like it was written by a real person for a real publication and then selected for an exam, not generated for an exam.

TEXT PERSONALITY PROFILES

Every text must have a personality — a voice, a perspective, a reason for existing. Before writing, decide which personality profile fits your topic and level. These profiles are derived from analysis of authentic texts in way2go! 5, way2go! 6, Best Shots, and Prime Time coursebooks.

Profile 1: The Personal Storyteller (best for B1 narratives)

• First-person voice with a distinctive personality
• Specific, vivid details: named people, exact places, precise numbers
• Includes at least ONE moment of vulnerability, humor, or surprise
• Structures like a story: setup → complication → reflection (not list of events)
• Uses dialogue or quoted speech from other people
• Register: informal to semi-formal, with occasional colloquial touches

Model openings from authentic B1 texts:

• "On a beautiful, sunny day in Cambridge, where I live, we were punting on the River Cam with some visitors. Five of us all together."
• "When I was little, I was pathologically shy. I was so shy I couldn't even trick-or-treat. My siblings would stand at a neighbor's door and point to the sad little Raggedy Ann at the curb to get candy for me."
• "I've just been on an awesome trip to New Zealand and almost came face to face with a hobbit!"

Profile 2: The Magazine Reporter (best for B1/B1+ informational articles)

• Third-person or mixed perspective with a journalistic voice
• Attributed quotes from named experts or participants
• Specific data: percentages, dates, study names, institution names
• A clear editorial angle (not neutral encyclopedia — the writer has a perspective)
• Includes at least one surprising fact or counter-intuitive finding
• Register: semi-formal with accessible phrasing

Model openings from authentic texts:

• "Studies of student behavior and attitudes show that cheating by high school and college students has risen substantially in recent decades, and that successful students are just as likely to do it as others."
• "Irish parents are far more likely to show their appreciation of good behaviour. In fact, our research revealed that 64% frequently reward their children with praise — more than other EU countries."

Profile 3: The Advice-Giver / How-To Voice (best for B1 practical texts)

• Direct address ("you," imperative mood)
• Organized with implicit or explicit tips/steps
• Practical, actionable language with real-world examples
• Warm, encouraging tone — like a trusted older friend or mentor
• Register: informal, with some semi-formal structures for credibility

Model openings:

• "When you start a new class or school, you might feel anxious — everything is new and different. But here's the thing: almost everyone feels that way."
• "You may think your school rules of 'no running' or 'remove your hat in class' are a pain, but compared to some rules around the world, yours are pretty tame."

Profile 4: The Analytical Voice (best for B1+ argumentative/analytical texts)

• Third-person with author's evaluative stance visible
• Multiple perspectives presented, but the writer's position emerges
• Uses hedging language: "tend to," "evidence suggests," "while critics argue"
• Includes concession-rebuttal structures
• Data and expert views interwoven with analysis
• Register: semi-formal to formal

Model from authentic B1+ texts:

• Complex sentences with embedded clauses and fronted adverbials
• Visible irony or rhetorical sophistication
• Multiple perspectives balanced in a single paragraph
• Hedged claims rather than absolute statements

Profile 5: The Human Interest Feature (best for B1+ cultural/social topics)

• Focuses on a specific person, place, or phenomenon as a lens for a bigger theme
• Opens with a concrete scene or anecdote, then zooms out to the broader issue
• Named protagonists with backstory and quoted speech
• Narrative thread running through an otherwise informational text
• Register: varies — may shift between informal (when quoting) and formal (when analyzing)

OPENING STRATEGIES — NEVER START GENERIC

The opening paragraph determines whether the text feels authentic or AI-generated. Use one of these strategies. NEVER open with a generic topic sentence like "X is becoming increasingly popular" or "In today's world, many people..."

AUTHENTICITY MARKERS CHECKLIST

Before finalizing the text, verify it contains AT LEAST 5 of these 8 markers:

• Named people with at least one identifying detail (age, nationality, role, relationship)
• Specific places — real city names, street names, institution names, not generic "a city"
• At least one direct quote or dialogue from a person in the text
• A moment of humor, irony, vulnerability, or surprise — something unexpected
• Concrete sensory detail — what something looked, felt, sounded, tasted like
• A register shift — at least one moment where the tone shifts (e.g., formal text with a colloquial aside, or informal narrative with a reflective formal sentence)
• Specific numbers or data — ages, percentages, distances, costs, dates
• An imperfection or complication — something doesn't go as planned, an expectation is subverted, a nuance is acknowledged

AI-TELLS TO ELIMINATE

Scan your text for these patterns and remove or rewrite them:

UNIVERSAL TEXT REQUIREMENTS

• Length: 400–600 words (B1), 500–700 words (B1+)
• Paragraphs: 5–8, with varied lengths (not all the same length)
• Genre: Match what students encounter in authentic materials — magazine articles, blog posts, personal narratives, reports, advice columns, interviews, feature articles, diary-style entries
• Source simulation: Write as if the text appeared in a real publication (The Guardian, BBC, a teen magazine, a personal blog, a school newspaper). Include a title and, where appropriate, a subtitle or byline
• Paragraph structure: NOT every paragraph should have a visible topic sentence. Some paragraphs should open with an example, a quote, a continuation of the previous idea, or a shift in perspective. The structure should feel organic, not templated.

TEXT-LEVEL CALIBRATION

Read references/b1-vs-b1plus.md for the full calibration guide. Key differences:

B1 Vocabulary Markers (use these naturally):

• Phrasal verbs: "turn up," "hang out," "come across," "break down," "find out"
• Collocations: "make friends," "do homework," "take a break," "catch the bus"
• Evaluative adjectives: "awesome," "amazing," "boring," "weird," "cool"
• Informal connectors: "but," "so," "and then," "anyway"

B1+ Vocabulary Markers (use these naturally):

• Semi-formal expressions: "tend to," "according to," "evidence suggests," "it appears that"
• Hedging: "somewhat," "to a certain extent," "arguably," "in many cases"
• Topic-specific terminology: introduced in context with enough clues to infer meaning
• More distant paraphrasing between text and items

BANNED TERMS — Austrian School Context

These British school terms must NEVER appear in the text OR items:

BANNED → USE INSTEAD:

• "form tutor" → "class teacher" or "homeroom teacher"
• "form group" → "class" or "homeroom class"
• "Year 9" / "Year 10" etc. → "grade 9" / "grade 10" or just "class"
• "sixth form" → "upper secondary" or omit
• "GCSE" / "A-levels" → omit or use "final exams"
• "head of year" → "head teacher" or "year coordinator"
• "prefect" → omit
• "house system" → omit or explain in context

Scan the ENTIRE output (text + items + answer key) for these terms before finalizing.

Step 3: Construct the Items

Read the format-specific reference file and follow its construction principles precisely.

Universal Item Requirements

• Minimum 8 items (plus item 0 as example)

• Item 0 is always a completed example with the answer filled in

• Text-order is absolute: the answer to item 1 comes before the answer to item 2 in the text, and so on. Verify this by listing the paragraph number for each item — the sequence must be non-decreasing.

• Even paragraph distribution (HARD LIMIT: max 2 items per paragraph): Items must be spread across all paragraphs. No more than 2 items from the same paragraph — no exceptions. Every substantial paragraph should contribute at least 1 item.

  MANDATORY CONSTRUCTION METHOD — paragraph-by-paragraph generation: Do NOT write all items freely and then verify distribution post-hoc. Instead:

  1. Read through the text paragraph by paragraph (P1, P2, P3, ...).
  2. For each paragraph IN ORDER, write exactly 1 item targeting that paragraph.
  3. After the first pass, you will have ~6–7 items. Select the 2–3 richest paragraphs and add a second item to each until you reach 8 items total.
  4. Number items sequentially: item 1 targets P1 or P2, item 2 targets the next, etc. Because items are generated in paragraph order, text-order is guaranteed by construction.

• Item language uses paraphrase — never copy exact phrases from the text into stems

• Each item tests reading comprehension, not general knowledge or guessing

• Refer to characters as "the writer" in stems — not "he/she" or personal pronouns. This avoids gender assumptions and matches SRDP convention.

Step 4: Create the Answer Key

For every task, provide:

1. Correct answers for all items
2. For MC: Rationale for why each distractor is wrong (what textual hook makes it plausible, and what makes it ultimately incorrect). Label each distractor with its type (1–6).
3. For T/F/J: All acceptable first-4-word justifications (there may be multiple valid ones)
4. For 4W: All acceptable answer variants (different phrasings that are correct). Every listed alternative must fit within the 4-word limit. Never include 5+ word answers in the answer key. If a natural phrasing exceeds 4 words, either trim it to fit or omit it.
5. For MM: Explanation of why each distractor option doesn't fit any gap

Step 5: Mandatory Verification and Correction Procedure

THE TWO-PASS RULE: The task is written first as a DRAFT, verified, and then output as a corrected FINAL VERSION if any errors were found.

1. First pass: Write the complete task (text + items + answer key) as a DRAFT.
2. Verification: Run all 5 verifications below. Record every error found.
3. Second pass: If ANY errors were found, output the CORRECTED FINAL VERSION of the complete task — rewriting the items/text/answer key sections with all corrections applied. Do NOT just list corrections. Output the entire corrected task.
4. If no errors were found, the draft IS the final version.

If you skip the second pass, the task is invalid.

Verification 1: Paragraph Map

Number the paragraphs in the text (P1, P2, P3...). Then for each item, identify which paragraph contains the answer evidence:

Text paragraphs: [count] paragraphs total
Item 0: Paragraph X (example)
Item 1: Paragraph X — [quote the specific sentence or detail]
Item 2: Paragraph X — [quote the specific sentence or detail]
...
Item 8: Paragraph X — [quote the specific sentence or detail]

Sequence: X, X, X, X, X, X, X, X, X
Non-decreasing? YES/NO → if NO, list which items must be reordered
Para counts: P1=N, P2=N, P3=N, P4=N, P5=N, P6=N, P7=N
Any paragraph >2 items? YES/NO → if YES, list which and which item to move
Any substantial paragraph with 0 items? YES/NO → if YES, add an item for it

Count paragraphs by re-reading the text top to bottom, not from memory. Titles, subtitles, and bylines are NOT paragraphs.

REBALANCING CONSTRAINT: If you need to move an item to a different paragraph, you must ALSO move its POSITION in the item sequence. After any move, renumber ALL items to match text order.

Verification 2: Answer Distribution

For MC — list correct answer letters:

Items 1–8: [letter], [letter], ... → A=N, B=N, C=N, D=N
Each letter at least 2? YES/NO. Any letter >3? YES/NO.
Target: 2-2-2-2. Acceptable: each letter 2–3 times.

For T/F/J — count T/F split:

True count: N. False count: N. (Must be 3–5 of each for 8 items)

Verification 3: Item Validity — THE DEVIL'S ADVOCATE TEST

MANDATORY: Read references/item-validity-firewall.md before running this verification. It catalogs every known trap type with real examples and prevention rules.

THE DEVIL'S ADVOCATE TEST — run for EVERY item regardless of format:

For each item, argue AGAINST the intended answer. Try to make the case that:

• A FALSE item is actually TRUE (or defensibly TRUE)
• A TRUE item requires too much inference to confirm
• An MC distractor is actually correct
• A 4W gap has multiple valid completions

If you can make a reasonable case → the item is invalid → REWRITE IT.

For T/F/J — the full battery:

Item X ([T/F]):
  Intended answer: [T or F]
  Devil's advocate — argue the OPPOSITE:
    [1-2 sentence argument for the opposite answer]
  Is the opposite argument reasonable? YES/NO
  → If YES: REWRITE the statement

Specific sub-checks for EVERY FALSE item:

1. "Technically True" test: Is this statement true at ANY level of description? (proximate cause, literal reading, different perspective, different timeframe) → If yes at ANY level → REWRITE
2. Quantifier test: Does the statement use "mainly," "mostly," "often," "sometimes"? → If yes → replace with specific, verifiable information
3. Timeframe test: Is this true at a different point in the text's timeline? → If yes → add explicit temporal anchor ("Before X," "After X")
4. Attribution test (multi-perspective texts): Could this be true from a different person's/expert's perspective in the text? → If yes → attribute the claim ("According to Dr. X, ...")
5. Single-falsehood check: Does the statement contain exactly ONE point of falseness?

   The falsehood is: [specific error]
   Text says: [what text actually says]
   Justification sentence: "[first 4 words]" — does this DIRECTLY show the falsehood? YES/NO
   

Specific sub-checks for EVERY TRUE item:

1. Explicit support test: Can you point to ONE sentence that makes this clearly true? → If you need 2+ sentences combined → too much inference for B1 → SIMPLIFY
2. Inference overreach test: Does confirming TRUE require a logical leap? → At B1: not acceptable. At B1+: acceptable only if strongly supported.

For MC — the NEGATION TEST + DEVIL'S ADVOCATE:

For each distractor, run two checks:

1. Negation test: Is this the opposite/negation of what the text says?

   Item X, option Y: "[distractor text]" — NEGATION? YES/NO → if YES, REPLACE with LATERAL SHIFT
   

2. Devil's advocate: Argue that this distractor IS the correct answer.

   Item X, option Y: Best case for this being correct: [argument]
   Is this case reasonable? YES/NO → if YES, REWRITE the item
   

Name the most tempting distractor per item, explain why a student might pick it, and why it's clearly wrong upon careful reading.

For 4W — word count + multi-answer audit:

Item X: "[answer]" = N words ✓/✗
  Alternative completions from OTHER paragraphs: [list any] → if found, stem is too vague → add specificity
  Stem-answer redundancy check: [insert answer into stem, read aloud] → redundant? YES/NO

Delete any answer exceeding 4 words entirely. If alternative completions exist from different paragraphs, rewrite the stem.

For MM — cross-fit audit:

For each option, test it in every gap:

Option X: fits gap [N] ✓. Also fits gap [M]? YES/NO → if YES, add disambiguation to text

For each distractor: tempting in at least one gap? YES/NO → if NO, replace distractor.

Verification 4: Authenticity Check (NEW)

Re-read the text and check against the Authenticity Markers Checklist (Step 2). Count how many markers are present:

Named people: YES/NO — [list them]
Specific places: YES/NO — [list them]
Direct quote/dialogue: YES/NO — [quote it]
Humor/irony/vulnerability/surprise: YES/NO — [describe]
Sensory detail: YES/NO — [quote]
Register shift: YES/NO — [describe]
Specific numbers/data: YES/NO — [list]
Imperfection/complication: YES/NO — [describe]

Score: X/8

If fewer than 5/8, rewrite the text before finalizing. This is the most common failure mode.

Also scan for AI-Tells:

• Does any paragraph open with "However," "Moreover," "Furthermore," "Additionally"?
• Does the text open with "X is becoming increasingly popular" or similar generic statement?
• Does every paragraph have a visible topic sentence?
• Do characters state emotions directly instead of showing them?
• Is there dialogue or quoted speech?

If any AI-Tell is found, rewrite the affected section.

Verification 5: Final Consistency

• Gender: scan all stems AND answer key for "he ", "she ", "his ", "her ", "son/daughter" → replace with "the writer" or "the writer's"
• T/F/J wording: task instruction must say "more than one sentence that supports your decision" (NOT "more than one correct answer")
• Banned terms: scan for form tutor, form group, Year 9, sixth form, GCSE, A-levels, etc.
• No editing artifacts: scan for crossed-out text, "[CORRECTED]" markers, draft remnants

Output the Final Version

After all verifications, output the complete corrected task. Label it clearly:

## FINAL VERSION
[Complete task with all corrections applied]

SRDP Themenpool (Topic Areas)

B1 Topics (5. Klasse / Year 5)

• Sports and leisure: personal experiences, hobbies, fitness, extreme sports, school sports
• School and education: new school experiences, boarding schools, school systems, study habits
• Nutrition and health: healthy eating, sleep, fitness, mental health, food culture
• Communication and relationships: social media, friendships, body language, family dynamics
• Rules, laws and regulations: school rules, strange laws, parenting styles, ethical dilemmas
• Transport and tourism: travel experiences, backpacking, city guides, transport modes

B1+ Topics (6. Klasse / Year 6)

• Arts and culture: galleries, exhibitions, street art, music scenes, cultural traditions
• Traditions, trends and lifestyles: fashion history, cultural comparisons, changing lifestyles
• Nature and the environment: ecotourism, climate change, conservation, urban green spaces
• The English-speaking world / Intercultural issues: regional Englishes, cultural differences
• Media: news consumption, social media impact, media literacy, journalism
• Science and technology: AI, digital innovation, medical technology, tech ethics

Custom topics are fine — apply the same level calibration and format principles.

Output Format

Present the task in this structure:

## [Topic] | [Level] | [Format]

### Reading Text

[Title]
[Subtitle/byline if appropriate]

[Text paragraphs]

### Task

[Rubric/instructions — use authentic SRDP phrasing from the reference files]
[Items with answer spaces]

### Answer Key

[Complete answer key with rationale]

### Verification Log

[All 5 verifications with results]

For a docx output: If the user asks for a Word document, use the docx skill to create a professionally formatted document matching the visual style of Austrian Matura practice materials.

Core Principles

Authenticity is the craft. Your texts must sound like they were written by a real person for a real audience and then selected for an exam. They must NOT sound like they were generated for an exam. This means specific details, named people, humor or vulnerability, dialogue, register shifts, and organic paragraph structures. If a text could have been written by any LLM with a generic prompt, it has failed.

The distractor is the art. Anyone can write a correct answer. The art lies in distractors that are plausible to a careless reader but clearly wrong to a careful one. Every distractor must have a "textual hook" — something in the text that makes it tempting. Direct contradictions (negations of the text) are NEVER acceptable as distractors.

Paraphrase is the bridge. The SRDP tests reading comprehension, not pattern matching. Question stems must paraphrase the text. If a student can answer correctly just by matching words without understanding meaning, the item has failed.

Level is in the details. B1 and B1+ differ systematically — not just in vocabulary, but in sentence structure, inference demands, topic abstraction, perspective complexity, hedging language, and the sophistication of distractors. B1 is personal and explicit. B1+ is analytical and partially implicit. Read references/b1-vs-b1plus.md carefully.

Variety is non-negotiable. If generating multiple tasks in a session, vary the text personality profile, opening strategy, text type, and topic angle. Never produce two texts that sound like they were written by the same generic author.