Training-spec drills

after_stage: ["identification","identification in company"]
after_stage_note: TWO prior stages, not one -- road.json's own step 6 needs both step 3 (identification) and step 4 (identification in company) first; after_stage was extended to hold a list for exactly this case (card road-chain-fixup-20260907, 2026-09-08).
symbols_known: 6
device: ["speaker"]

Before: identification and identification in company
After: minimal sequence reading

literacy-design.md §1 row 2 + interaction-model.md stage 2: script syllable displayed → pick its IPA among K numbered buttons (tap or digit key); correct flashes green and plays the syllable's audio as confirmation.
Shape: ? in, multiple_choice out

stage: minimal sequence reading
unit: syllable

after_stage: identification
symbols_known: 6
device: ["speaker"]

Before: identification
After: minimal sequence reading

Written syllable displayed; the K=4 candidate audios auto-play in sequence, each numbered button highlighting as its sound plays (the sweep); tap a button to hear it again and ARM it; tap the armed button (or press its digit twice) to commit. Space replays the sweep. Working-memory bound says K=4, not 6.
Shape: ? in, multiple_choice out

stage: minimal sequence reading
unit: syllable

after_stage: identification
symbols_known: 6
device: ["speaker"]

Before: identification
After: minimal sequence reading

literacy-design.md §1 row 2: CV cell text → pick matching audio from choices; confusors are same-C-other-V and same-V-other-C cells; itemSource = tglx CV worklists (devanagari/sets/cv-survey.tset)
Shape: ? in, multiple_choice out

stage: minimal sequence reading
unit: syllable

after_stage: minimal sequence reading
symbols_known: 12
device: ["speaker"]

Before: minimal sequence reading
After: extended sequence reading

literacy-design.md §1 row 3 + interaction-model.md §3 stage 3 (prereq: stage 2 cv-read certified + word meanings exist as PICTURES, no L1 text prereq): word text stimulus → pick meaning-picture among K; confusors = words sharing 2 of 3 phonemes (CVCVC near-homophones); itemSource = tglx lex/ + picture set.
Shape: ? in, multiple_choice out

stage: extended sequence reading
unit: word

after_stage: ["minimal sequence reading","sequence"]
after_stage_note: TWO prior stages, not one -- road.json's own step 7 needs both step 6 (minimal sequence reading) and step 5 (sequence) first; after_stage was extended to hold a list for exactly this case (card road-chain-fixup-20260907, 2026-09-08).
symbols_known: 12
device: ["speaker"]

Before: minimal sequence reading and sequence
After: extended sequence reading

Show the written word; pick the sound it makes among K audio buttons. Confusors: feature_nearest on c,v.
Shape: ? in, multiple_choice out

stage: extended sequence reading
unit: word

after_stage: minimal sequence reading
symbols_known: 12
device: ["speaker"]

Before: minimal sequence reading
After: extended sequence reading

Show the written CVCVC word; pick the sound it makes among K audio buttons, same doctrine as cvcvc-read.json (b158 verdict A). Confusors: feature_nearest on c,v (the word's own consonant/vowel skeleton).
Shape: ? in, multiple_choice out

stage: extended sequence reading
unit: word

after_stage: minimal sequence reading
symbols_known: 12
device: ["speaker"]

Before: minimal sequence reading
After: orthography

literacy-design.md §1 row 5 + interaction-model.md stage 5: hear the word (audio prompt) → pick the TRUE spelling among K, where same-rule trap spellings (nite, fone, sent) are preferred distractors. Proof P1. Misses re-show the rule decomposition.
Shape: ? in, multiple_choice out

stage: orthography
unit: spelling

after_stage: minimal sequence reading
symbols_known: 12
device: ["speaker"]

Before: minimal sequence reading
After: English orthography

Show the target word with the grapheme highlighted (PhaseRenderer's word_with_highlighted_sequence rendering, b143); pick the SOUND it makes here among audio choices. Confusors: same_written_form (the other sounds this exact grapheme makes elsewhere).
Shape: ? in, multiple_choice out

stage: English orthography
unit: spelling

after_stage: auditory exposure
symbols_known: 0
device: ["speaker"]

Before: auditory exposure
After: identification

literacy-design.md §1 row 1 + interaction-model.md stage 1: audio prompt → tap the glyph among K; no literacy prerequisite (entry-level drill).
Shape: ? in, multiple_choice out

stage: identification
unit: symbol

after_stage: auditory exposure
symbols_known: 0
device: ["speaker"]

Before: auditory exposure
After: identification

literacy-design.md §1 row 1: audio (tts/wav) → tap glyph among K; confusors from competence-model nearest confusable glyphs (visual + phonetic distance); itemSource = tglx sym/ + sets/
Shape: ? in, multiple_choice out

stage: identification
unit: symbol

after_stage: auditory exposure
symbols_known: 0
device: ["speaker"]

Before: auditory exposure
After: identification

Reverse of identify.json: show the glyph as text, offer K sounds as audio buttons, tap the one this glyph makes.
Shape: ? in, multiple_choice out

stage: identification
unit: symbol

after_stage: auditory exposure
symbols_known: 0
device: ["speaker"]

Before: auditory exposure
After: identification

literacy-design.md §1 row 1 + interaction-model.md stage 1: audio prompt → tap the glyph among K; no literacy prerequisite (entry-level drill).
Shape: ? in, multiple_choice out

stage: identification
unit: symbol

after_stage: identification
symbols_known: 3
device: ["speaker"]

Before: identification
After: identification

Runs on the RECORDED ipa syllable sets (cv 286, cvc 50, cvcvc 100 - all have Tom-recorded audio). The learner hears a syllable and picks its written form from choices. Every item CONTAINS a letter the learner has already met, so the drill is always about their own letters, never a cold symbol. Confusors are chosen to differ in exactly the target letter where possible, so getting it right REQUIRES hearing that letter, not guessing from the rest.
Shape: ? in, multiple_choice out

stage: identification
unit: syllable

after_stage: minimal sequence reading
symbols_known: 2
device: ["speaker"]
note: Offer only once the learner knows the two sounds in the syllable on their own; before that they are guessing the spelling.

Before: minimal sequence reading
After: production

Hear a syllable, build its spelling from letter tiles: one tile for the consonant, one for the vowel, among a few wrong ones. Nothing written is shown in the question, because the spelling IS the answer. Fills the last open stage of wiki learning-trajectory.md, production: of 28 drills 17 ask the learner to pick one of several and none asks them to write what they hear.
Shape: ? in, tile_construct out

stage: production
unit: syllable

after_stage: minimal sequence reading
device: ["speaker"]

Before: minimal sequence reading
After: closed-class paradigm

isolated-paradigm: show one member of a real, cited paradigm subgroup; pick its meaning among the OTHER members of the SAME subgroup (same_grid confusors), so a learner is never asked to tell el apart from an unrelated word -- only from the ones that could really sit in that same slot.
Shape: ? in, multiple_choice out

stage: closed-class paradigm
unit: article

after_stage: minimal sequence reading
device: ["speaker"]

Before: minimal sequence reading
After: closed-class paradigm

isolated-paradigm: show one member of a real, cited paradigm subgroup; pick its meaning among the OTHER members of the SAME subgroup (same_grid confusors), so a learner is never asked to tell ga apart from an unrelated particle -- only from the ones that could really sit in that same slot.
Shape: ? in, multiple_choice out

stage: closed-class paradigm
unit: particle

after_stage: minimal sequence reading
device: ["speaker"]

Before: minimal sequence reading
After: closed-class paradigm

isolated-paradigm: show one member of a real, cited paradigm subgroup; pick its meaning among the OTHER members of the SAME subgroup (same_grid confusors), so a learner is never asked to tell -a apart from an unrelated word -- only from the ones that could really sit in that same slot.
Shape: ? in, multiple_choice out

stage: closed-class paradigm
unit: case marker

after_stage: minimal sequence reading
device: ["speaker"]

Before: minimal sequence reading
After: closed-class paradigm

isolated-paradigm: show one member of a real, cited paradigm subgroup; pick its meaning among the OTHER members of the SAME subgroup (same_grid confusors), so a learner is never asked to tell -aḥ apart from an unrelated ending -- only from the ones that could really sit in that same case-and-stem slot.
Shape: ? in, multiple_choice out

stage: closed-class paradigm
unit: case ending

Before: not specified in this trspec's data
After: not specified in this trspec's data

Shape: ? in, record_compare out