v1 architecture sketch

`Upload (audio/video/transcript)` → `Consent capture + Local Contexts Labels at ingest` → `ASR front-end (Elpis/wav2vec2 for partner languages; Whisper-class for English/Aboriginal English)` → `Human correction UI for community language workers` → `LLM-assisted structured extraction → TNFD/GRI/GBF disclosure template + SavCAM/ACCU-relevant practice flags` → `Community review & approval gate` → `Governed export to corporate buyer`. Indigenous-controlled data store throughout; buyers never touch raw TEK.