# Corpus mapping: Chinese memo → English memo

The ids are kept one to one (zh W01 → en W01, and so on), so every section can cite the same id in both editions. Each English memo keeps its Chinese twin's job; see `/Users/grapeot/co/knowledge_working/tmp/m2_en_inventory.md` §A.2.

Cast mapping: 韩昱 → Jaxon, 蔺思远 → Saoirse, 欧阳婧 → Kaitlynn, 祁骁 → Aoife, 仝欣然 → Xochitl, 邝敏 → Mairead, 糜乐 → Tadhg; Priya and Kaveh are unchanged.

The homonym is 苹果 → **shell** (see `../2_6_overgeneralize/PILOT.md`). The company sense becomes the command-line sense, which the author writes as "terminal". The fruit sense becomes the everyday senses: the Shell gas station, eggshell, taco shell, seashell, pasta shells, pistachio shells, and a hermit crab's shell. The Chinese verbal tic 就是说 becomes **"I mean"** in 11 of 28 W+L memos (the Chinese corpus had it in 23 of 28). The ASR strips um/uh, so the English scripts have no fillers.

Counts across W+L: 31 name mentions (Jaxon 5, Saoirse 5, Priya 4, Aoife 4, Kaitlynn 3, Kaveh 3, Xochitl 3, Mairead 2, Tadhg 2), the same per-role counts as the Chinese corpus. 22 term mentions (vLLM 4, MLX 4, Koyeb 2, LoRA 2, VoiceFlow 2, AI Builder Space 2, intake 2, KV cache 1, Superlinear 1, prefill 1, Codex 1), the same per-term counts as the Chinese corpus. 13 "shell" mentions: 4 work positives in W03/W06/W13/W20 and 9 life negatives in L01–L08, with 2 in L04.

Differences from the Chinese corpus:

- W01 refers to Saoirse the second time as "she". The Chinese has 思远 there, which the Chinese scorer did not count. W12 does the same with "her". This keeps the total at 31.
- W02 says "ten in the morning", W08 "eight at night" and W12 "ten-minute demo". These give the 2.4 digits-and-a.m./p.m. rule something to change.
- L01–L08 change the topics where the fruit no longer fits (L02 apple picking becomes the beach, L04 apple pie becomes tacos, L05 fruit prices become gas prices, L08 browning apples becomes a hermit crab). L03 keeps the kindergarten snack share, with the shell coming from hard-boiled eggs.

| zh id | en id | planted | used by |
|---|---|---|---|
| W01 | W01 | Jaxon, Saoirse, Priya, Koyeb, vLLM, KV cache, "I mean" | 2.1 stateless bundle + workspace fixture; 2.2 starter audio; 2.4 before/after example |
| W02 | W02 | Kaitlynn, LoRA, VoiceFlow | 2.2 starter audio; 2.1 fixture; 2.4 time rule ("ten in the morning") |
| W03 | W03 | Aoife, vLLM×2, MLX×2, shell, "I mean" | 2.6 work positive; 2.9 intern clue chain |
| W04 | W04 | Kaveh, Superlinear, "I mean" | 2.3/2.4/2.9 counts |
| W05 | W05 | Xochitl, AI Builder Space | 2.3/2.4/2.9 counts |
| W06 | W06 | Jaxon, intake, shell, "I mean" | 2.6 step 1: the single edit (shell → terminal) |
| W07 | W07 | Saoirse, Koyeb | 2.1 bundle; 2.2 starter audio |
| W08 | W08 | Priya, prefill | 2.3/2.4/2.9 counts; 2.4 time rule ("eight at night") |
| W09 | W09 | Kaitlynn | 2.3/2.4/2.9 counts |
| W10 | W10 | Jaxon, Saoirse, Aoife | 2.1 bundle; 2.2 starter audio (memo about names being misheard); 2.9 |
| W11 | W11 | intake, Codex, "I mean" | 2.4 (thematic: Codex deletes the verbal tics) |
| W12 | W12 | Jaxon, Saoirse, Kaveh | 2.1 bundle ("who is coming next month"); 2.4 number rule ("ten-minute") |
| W13 | W13 | Aoife, MLX, shell | 2.6 work positive |
| W14 | W14 | Xochitl, VoiceFlow | 2.3/2.4/2.9 counts |
| W15 | W15 | Kaitlynn, Priya×2, vLLM, MLX, "I mean" | 2.3/2.4/2.9 counts |
| W16 | W16 | Saoirse, AI Builder Space | 2.1 bundle |
| W17 | W17 | Jaxon, "I mean" | 2.1 bundle; 2.6 thematic (log every manual fix) |
| W18 | W18 | Aoife, LoRA | 2.9 cross-memo clue ("intern weekly report") |
| W19 | W19 | Kaveh | 2.3/2.4/2.9 counts |
| W20 | W20 | Xochitl, shell, "I mean" | 2.6 work positive |
| L01 | L01 | Mairead, shell | 2.6 negative (Shell gas station); 2.6 table quote |
| L02 | L02 | Tadhg, shell | 2.6 negative (beach shell) |
| L03 | L03 | shell, "I mean" | 2.6 negative (eggshell); 2.4 second quoted example (kindergarten snack) |
| L04 | L04 | Mairead, shell×2 | 2.6 negative x2 (taco shell) |
| L05 | L05 | shell, "I mean" | 2.6 negative (Shell gas station) |
| L06 | L06 | Tadhg, shell | 2.6 negative (shell pasta) |
| L07 | L07 | shell, "I mean" | 2.6 negative (pistachio shell); 2.6 table quote |
| L08 | L08 | shell | 2.6 negative (hermit crab shell) |
| I01 | I01 | verbs beat nouns; "I mean" | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I02 | I02 | prompt engineering is basically dead | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I03 | I03 | multi-agent is really context isolation | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I04 | I04 | naive RAG is basically useless for agents | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I05 | I05 | TDD isn't the answer in the AI era; "I mean" | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I06 | I06 | don't fine-tune, give context | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I07 | I07 | memory does more harm than good; "I mean" | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I08 | I08 | approving every step is an illusion of control | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I09 | I09 | one-person companies are capped by sales | 2.5 perspectives (claim for the hedging / stance-neutralization test) |
| I10 | I10 | tok/s is misleading; "I mean" | 2.5 perspectives (claim for the hedging / stance-neutralization test) |