Skip to content

knowledgestream-hub.boundary-model.md

Latest commit

 

History

History
86 lines (62 loc) · 4.06 KB

knowledgestream-hub.boundary-model.md

File metadata and controls

86 lines (62 loc) · 4.06 KB
Raw

Transcript Provenance And Retention Boundary Model

Purpose

This document closes public issue #3: Document transcript provenance and retention controls for private enterprise video knowledge.

It provides concrete agentic-enterprise evidence for KnowledgeStream Hub. The repo remains independent: it is not an AAIF project, not a Linux Foundation project, and not endorsed by either organization.

High-Risk Action Boundary

Private video ingestion, transcript generation, transcript correction, semantic indexing, search-answer generation, access-controlled sharing, retention changes, and recording export.

Actors

Actor Responsibility
Content owner Owns recording classification, audience, retention class, and sharing scope.
Knowledge reviewer Reviews transcripts, tags, summaries, and searchable snippets.
AI knowledge assistant Proposes transcripts, summaries, tags, and search answers.
Search/index provider Stores or retrieves transcript chunks under approved scopes.
Maintainer Reviews provenance, retention, provider fallback, and audit evidence.

Identity Boundary

  • Recording ownership, viewer access, transcript reviewer authority, and AI assistant capability must be separate.
  • The AI knowledge assistant cannot widen audience, bypass retention, or approve transcript publication.
  • Every recording and transcript must preserve source, owner, classification, retention, provider, and reviewer state.
  • Provider credentials, private transcript text, and restricted recordings must never be exposed in public prompts or examples.

Tool And Action Boundary

Action Boundary
ingest_recording Allowed only with declared owner, classification, source, and retention class.
generate_transcript Requires access check and provider-scope decision.
index_transcript Allowed only for approved transcript chunks and declared retention.
answer_from_video Must return provenance and respect viewer authorization.
export_recording Requires content-owner approval and retention/audience recheck.

Human Approval Gates

  • Before publishing transcripts or summaries for executive, HR, legal, customer, or regulated recordings.
  • Before widening video or transcript audience.
  • Before exporting recordings or transcript packages outside the workspace.
  • Before changing retention class, legal hold, or deletion policy.
  • Before changing transcript/search providers for restricted content.

Audit Event Model

Event Minimum Evidence
recording.ingested recording id, owner, source, classification, retention
transcript.generated recording id, provider, model, language, reviewer state
transcript.indexed transcript id, index id, chunk policy, retention class
video_answer.generated query hash, viewer, source chunks, authorization decision
recording.exported recording id, destination, approver, retention state

Local Proof Path

  • bun install to prove dependency resolution.
  • bun run lint to prove static project health.
  • bun run test for transcript, search, retention, and audit fixtures when available.
  • bun run build to prove the application compiles.
  • Use synthetic recordings and redacted transcript fixtures in public proof.

OSS And Self-Hosted Fallbacks

  • Use Whisper-compatible local transcription for public examples where feasible.
  • Use PostgreSQL, SQLite, or local vector stores for transcript/search fixtures.
  • Use S3-compatible storage such as MinIO for local recording storage.
  • Keep hosted transcription, search, analytics, and storage providers behind explicit adapters.

Contributor Follow-Up

Add a small transcript fixture that maps recording owner, classification, transcript provider, index policy, viewer authorization, and export approval.

Issue Closure Evidence

This document satisfies the issue checklist by separating:

  • identity or actor boundary
  • tool/provider/action boundary
  • human approval or escalation point
  • audit or observability events
  • OSS/self-hosted fallback direction
  • validation and static inspection path