fix(mock+safety): route MOCK by sample fingerprint + defensive fallbacks

feat(lora): train-only SFT experiment scaffold (cookbook recipe, ms-swift)

feat(loop): agent-trace draft (consent-gated), demo kit, robustness v2

fix(triage): deadline-pressure paraphrases count as urgency

fix(prompts): read non-English letters in original script or refuse honestly

feat(triage): Devanagari + Japanese scam vocabulary in heuristics

feat(eval): Hindi + Japanese letter-text coverage in the forge

fix(ui): easter egg on the live Space + phishing severity boost

feat(training): letter forge + labeled dataset + Modal prompt-A/B eval

feat(ui): Gen X Soft Club easter egg on Japanese + loader/UX polish

fix(safety): strip letter contact details from facts and reasons too

feat(modal): full GPU validation matrix on Modal

fix(live): single progress overlay + drop prompt-echo 'facts'

feat(ui): localized analyzing state with envelope-flap loader

feat(ui): one-tap example letters shared with the Modal harness

feat(ui): 1-2-3 postmark step strip, localized in all four languages

feat(ui): postage-stamp verdict badge in worry card

feat(ui): Kitchen-Table Post Office skin — postal CSS, forced-dark wiring

feat(i18n): Japanese (ja) as fourth language — strings, labels, mock

feat(ui): design tokens with WCAG AA contrast tests + forced-dark theme

Add json-repair as final JSON-parse fallback for LLM output slips

Salvage truncated model JSON; surface signal labels when reasons absent

Harden JSON parsing + synthesize explanation fallback (live-Space regressions)

Mystery-Mail Guardian: complete app for the Build Small Hackathon