A spoken-Tamil learning tool — hear a story you'd read anyway, with colloquial Tamil woven in (diglot-weave, Burling 1968). 3 new words per ~20-min block; known words carry forward and accumulate across a series. Built for Harry; later Hermes.
Suzanne Palmer's Hugo-winning novelette — more bots, after you loved Lost Ark. 4 ~14-min blocks. A second no-new-words episode: all 24 words you know, woven back in to grind the level (pair with the Anki deck). Back to the familiar narrator and calm, clear voices.
Kelly Link's Hugo+Nebula story, 3 ~13-min blocks. A no-new-words reinforcement episode: all 24 words you know, woven through a new story to hammer the level while you drill the Anki deck.
Naomi Kritzer's Hugo+Nebula winner, 4 ~18-min blocks. First run of the matured pipeline: your 12 known words recur (reinforcing) while 12 new ones are added, 3 per block. Full-cast voices.
Suzanne Palmer's Hugo novelette in four ~20-min blocks. 3 new words/block, density-capped, full-cast voices. The build that locked the working model.
The 24 words from the stories as a spaced-repetition deck (download → import into Anki). Two audio cards each: hear-Tamil→meaning, and English→say-it. The deliberate-practice adjunct so the words stop washing out.
Earlier experiment: structure-only weave. Washed over (i+0) — the negative result that led to the real series.
An Aesop fable woven four ways to see where each chunking strategy breaks.
Syntax-first: why, for Tamil, word order and case/tense morphology are one coupled knob.
Where it stands: the pipeline is live and proven — weave Tamil into novelettes/essays Harry reads anyway, full-cast voices, density-capped. 24 words introduced so far across two stories. All Tamil draft, pending Deepa.