Male-described speakers were mapped to the female voice achernar (and a few
females to schedar) — 210 VOICE swaps across the four regenerated books.
Also: book-11 'Two Letters' now labeled Jane:/Ann: with female voices
instead of one Narrator (Yana's report), and junk auto-declared 'VOICE: ID'
lines dropped.
Co-authored-by: Cursor <cursoragent@cursor.com>
Full reconversion from book markdown + tape transcripts (probe-first, eval
vs baseline), silence-aware clip slicing, Fable master assembly. All 30
track lessons and all 10 masters validate at 0 errors / 0 warnings; every
master uses 100% of its cassette clips.
Co-authored-by: Cursor <cursoragent@cursor.com>
Trailing years no longer cut mid-word and starts no longer carry the previous phrase (tester reports on 1c-f2, 1c-f4, 1d-f1).
Co-authored-by: Cursor <cursoragent@cursor.com>
Remove markdown dividers that were parsed as spoken dialogue and caused lesson TTS requests to time out.
Co-authored-by: Cursor <cursoragent@cursor.com>
Regenerate every LLA tape lesson from original markdown+transcripts with the
post-review pipeline, select the cleanest candidate per lesson, deterministically
refine/split ranges, and slice fresh clips. All 30 per-track drafts validated at
0 errors / 0 warnings before assembly.
Reassemble weekly masters with Claude Fable (not Gemini), apply gated final
structure repair, and restore any drifted clip-bearing lines from source tracks.
All eight changed masters validate at 0/0, pass stage-master clip identity, and
use 100% of their source cassette clips. Includes the systemic fixes behind
Yana's reports: no Number/Repeat cue leaks, no doubled pronunciation models,
clean split boundaries, audible cloze setup context, apparatus-free feedback,
and properly distributed vocab teaching beats.
Co-authored-by: Cursor <cursoragent@cursor.com>
Converted, refined, and re-sliced both LLA tracks from the tape with the
tester-round rules (passage->DIALOGUE+REFERENCE splits, vocab on the first
teaching line, sentence-sized read-only cards, merged-beat detection with
best-of-two sampling, repair loop hard-gated on figure coverage).
Book 10 masters now include tape lessons 2C-4D — ten lessons of audio content
the earlier masters never shipped (they used only 1A-2B). All ten masters
lint clean, 100% cassette coverage on every taped lesson.
Co-authored-by: Cursor <cursoragent@cursor.com>
- Bob and Dave / Martha and Sally listening stories become their own dialogue
activities; the What-Didn't-You-Hear questions REFERENCE them (the whole
passage no longer sits inside the question prompt)
- Brutus the Cat: the full story is a dialogue activity, and the follow-up
questions REFERENCE it (testers had to answer from memory)
- vocab tags moved to the first line that actually teaches the word
(gate, dog/cat/bird, run/climb inflections, sky)
- long read-only paragraph lines split into sentence-sized cards
(airports, circle definitions, winter yard)
- self-study wording: removed classroom "this week" phrasing
Co-authored-by: Cursor <cursoragent@cursor.com>
- Restore Linda's Mistake as the printed word-bank task: shared SELECT options,
gapped sentence templates, completed-sentence feedback, and repeat.
- Restore The Other One to DIALOGUE. The source says 'Repeat the sentences';
conversion invented a PRODUCE task and printed the full response in TEMPLATE.
Co-authored-by: Cursor <cursoragent@cursor.com>
- MISSING_ALTERNATIVES (63/64 applied): uncontracted/contracted forms, digit
variants (05|5), synonyms and full-sentence variants, via ACCEPT lines and
inline gap alternatives; skipped 'better/fine' for the Opposites drill (not
opposites of "sick").
- NOT_DERIVABLE (16 fixed, 10 false alarms, 1 skipped): item-level fixes
(visible word hints, missing premise, speaker clarifications "Joe (to Jim):",
emphasis/cue corrections, un-gapped unguessable words, rounded-number
accepts); converted the two book-10 paragraph clozes (Mary / The Mall) from
CHECK: exact to CHECK: llm with logical-completion rubrics; cross-book
recycled vocab (pilot, teller, barber, variety store, ...) left alone.
- UNCLEAR_TARGET (10): instruction/intro corrections (Food "Type the word for
the food.", Explain-with-Too intro), cue/emphasis fixes.
- Rubric rewrites (57): applied audit-suggested grader rubrics (now effective
with the pre-item RUBRIC parser fix); moved the two per-item rubrics of
book-11 "Write the Main Idea" inside their items.
No clip-bearing line changed; all masters validate (0 errors, 0 new warnings)
and pass stage-master clip-identity checks. Full decision log:
reactor-module-tools module-convert/answerability-audit/APPLIED.md
Co-authored-by: Cursor <cursoragent@cursor.com>
Conversion kept only the tape's confirmation pass ("Now check your
sentences...") as RESPONSE clips, so the gap answers were never audible
during the exercise. Add a PROMPT per item carrying the stimulus pass
("Listen and write", ~12-148s of LLA 1C Figure 6): three short sentences
plus the combined sentence, with clips cut from the figure audio and
ASR-verified against the transcript.
Co-authored-by: Cursor <cursoragent@cursor.com>
Those items show a group of different shapes and want the CATEGORY word, so a
learner naming the individual shapes — perfectly good English — was marked wrong
with no idea why (Elena). The grader is now told to say so and hint that one word
covers them all.
Co-authored-by: Cursor <cursoragent@cursor.com>
A word tagged VOCAB in one activity was tagged again in later activities of the
same session, so the learner got the same chip repeatedly — most visibly in the
pronunciation drills, which re-listed every word from the dialogue before them
(reported: book 7 session 1-2). Rule applied: drop a VOCAB whose word was
already tagged in an EARLIER activity of that session. Repeats INSIDE one
activity are kept — those teach distinct senses ('stuck to his finger' vs
'stuck the milk in the fridge', 'shop' as verb then noun). Also dropped chips
for plurals of already-known words in book 7 session 1-2, where the drill
teaches the singular/plural pair rather than new vocabulary. No session lost
all of its vocabulary (verified).
Co-authored-by: Cursor <cursoragent@cursor.com>
The word-list activity had 'Side. Sides.' on two consecutive lines: the second
reused the SAME audio range as the first but carried the NEXT item's picture, so
the learner saw a triangle and heard 'Side. Sides.' (Elena). The tape says it
once, with silence until 'Triangle.' — the line was invented in conversion.
Scanned all 12 books for the signature (same text + same clip range + different
image): the other 9 hits are legitimate drills asking one repeated question
about different pictures.
Co-authored-by: Cursor <cursoragent@cursor.com>
The converter tagged the section's THEME word on every line — 'job' four times
was the only VOCAB in the whole session — while the words each line actually
teaches (doctor, nurse, teacher, hospital, mechanic, travel agent, waiter) went
untagged, so the learner got the same chip three times in one dialogue and
again in the next.
Co-authored-by: Cursor <cursoragent@cursor.com>
