Master release QA checklist
This checklist combines the official task architecture, the methodology learned from public Cambridge samples, and every design goal retained from this project. It is tied to this exact build.
North-star acceptance criteria
| Area | Release requirement | Status in this build |
|---|---|---|
| North star | Cambridge-faithful simulator with support off; intelligent listening coach with support on. Both routes keep the same B1 target and answer logic. | PASS |
| Official structure | Exam rounds use 7 / 6 / 6 / 6 answer points: 25 across the four Listening parts. | PASS |
| Originality | Public samples guide high-level architecture, pacing and distractor methodology only. No published dialogue is retained through simple word substitution; every script, question, option, character, explanation and cue card is original. | PASS |
| Sample-calibrated pacing | Broad guardrails based on the spoken content of official Sample Test 1: Part 1 70–105, Part 2 80–125, Part 3 250–340 and Part 4 390–480 words. Natural coherence takes priority over padding. | PASS — bank ranges 75–84 / 96–108 / 290–292 / 415–430 |
| Part 1 method | Seven independent short recordings in an exam round; three meaningful visual choices; all choices related to the audio; one answer resolved through correction, sequence, practical elimination or final decision. | PASS — 14 items / 42 cue cards |
| Part 2 method | Six independent dialogues in an exam round; three written options; the question targets the whole message, reason, advice, attitude or decision rather than isolated word matching. | PASS — 12 dialogues / 36 options |
| Part 3 method | One continuous monologue with six gaps in audio order. Canonical answers are no more than two words, a number, date or time; source evidence is exact and sensible variants are accepted. | PASS — 2 talks / 12 gaps |
| Part 4 method | One continuous two-person interview with six ordered questions. Questions sample the beginning, middle and end and mix detail, reason, attitude, advice, process, reflection and future thinking. | PASS — 2 interviews / 12 questions |
| Question logic | The recording alone supports exactly one answer. Distractors are plausible but rejected, corrected, less complete or tied to the wrong speaker/time. Explanations identify the decisive evidence. | PASS — full bank reviewed |
| Natural language | Believable school-age contexts, no test-aware/meta dialogue, no unrelated tail lines, no repeated generic filler, no malformed punctuation and no pronoun/name drift. | PASS |
| Speaker assignment | In every two-person recording, the first unique character receives the male voice and the second receives the female voice. Monologues may use either voice. | PASS |
| Part 4 pairing | Every interview uses a male Presenter and one female guest only. No narrator, third character, merged turn or voice switching. | PASS |
| Voice choices | Settings offer UK, US or any-English accent preference and separate male/female browser voice selectors. Missing voices fall back safely. | PASS |
| TTS source | TTS reads spoken turns only. It never reads titles, questions, options, punctuation names, speaker labels or interface instructions. | PASS |
| Transcript integrity | Audio and display use the same source data. Every turn has a bold label and separate row; monologues remain one speech block. | PASS |
| Transcript control | No transcript text appears before deliberate reveal or submission. Reveal state is explicit and resets on every new item or set. | PASS |
| Training mode | Adjustable speed, flexible replay, deliberate transcript reveal, explanations, methodology coaching, Part 3 prediction/hints and supported-attempt tracking. | PASS |
| Exam mode | Natural 1.00 speed, exactly two complete plays, no transcript/hints/method labels before submission and delayed correctness feedback. | PASS |
| Method integration | Each reviewed item explains its listening target, decisive evidence and distractor method. Part 3 shows an information map; Part 4 shows a full-interview skill map. | PASS |
| Visual alignment | Every Part 1 option has an original semantic cue card that matches its meaning. Training mode may show labels; Exam mode hides written option text so learners cannot solve the task by matching an audio noun to a printed noun. | PASS |
| Empty states | No task auto-starts. Each page clearly instructs learners to start a round or set, so an unloaded transcript cannot look broken. | PASS |
| Progress clarity | Score, completed items, streak and support use are tracked independently so assisted work is not confused with independent performance. | PASS |
| Appearance and access | System/light/dark appearance, text sizing, keyboard operation, visible focus, semantic controls, live feedback and skip navigation. | PASS |
| Responsive layout | No horizontal overflow at 390 px; all six navigation links remain reachable; Settings stays inside the viewport; desktop/tablet layouts remain usable. | PASS |
| Packaging | Static Cloudflare Pages build with no external runtime dependency. Official Cambridge links on the Method page are references only. | PASS |
Evidence generated from the exact release files
Automated structural and methodology QA: 12 pass groups, 0 warnings and 0 failures.
Browser/runtime smoke test: 132 checks passed, including Method-page wiring, empty states, transcript conceal/reveal/reset, methodology panels, Part 3 predictions, Part 4 skill mapping, two-play Exam limits, TTS speaker assignment and 390 px mobile overflow.
Human semantic pass: all 14 Part 1 items, 12 Part 2 dialogues, 2 Part 3 talks with 12 gaps, and 2 Part 4 interviews with 12 questions were reviewed for evidence, distractor logic, coherent endings, names/pronouns, level, pacing and naturalness.
Independent content-consistency scan: 15 pass groups and 0 failures, including exact duplication, known-speaker parsing, alternating turns, Part 3 source order, Part 4 full-arc sampling and repeated twelve-word passage detection.
The raw reports are included as QA_AUTOMATED_OUTPUT.txt, QA_RUNTIME_OUTPUT.txt, QA_CONTENT_CONSISTENCY_OUTPUT.txt, HUMAN_CONTENT_REVIEW.txt and FINAL_RELEASE_QA_REPORT.txt. The validators are included so the checks can be rerun after any future edit.
Definition of done
Structural QA: PASSMethodology QA: PASSHuman content QA: PASSRuntime QA: PASSOverall release status: PASSA saved report is not accepted as evidence by itself. The tests must pass again against the files extracted from the final deployable ZIP.