3. The Studio Method (Session Lifecycle)
3.1 Pre-Production (1–2 weeks before)
Brief & PRD: goals, users, KPIs, constraints, branding.
Data Inventory: what exists, gaps, permissions, retention.
Context Pack Plan: sources, chunking strategy, metadata schema, redact rules.
Eval Plan: what “good” looks like (hallucination <1%; citation precision >0.8; latency <2 s for QPS 1; etc.).
Artifacts: PRD v1, Context Map, Eval Rubric, Session Agenda.
3.2 Production (Live Session)
Block A — Scaffolding: spin up base agents, connect tools (search, DVLA-like APIs, POS, etc.), verify RAG end-to-end.
Block B — Prompt & Policy: system prompts, tool-routing rules, tone/brand style, safety filters.
Block C — Multi-Agent: define roles (Researcher, Planner, Critic, Publisher), collaboration graph, stop conditions.
Block D — Evaluation: run harness on curated test set; iterate until thresholds pass.
Block E — “Print Take”: freeze Master Agent v1.0, sign hash, export stems.
Artifacts: Master Agent v1.0, Stems (prompts, tools, knowledge, configs), Eval Report.
3.3 Post-Production (Mastering & Release)
Security & Compliance Gate: bias checks, safety red-team, PII-leak tests.
Performance & Cost Tuning: quantization, cache policies, batching.
Documentation: runbook, admin guide, upgrade path, SLAs.
Artifacts: Mastering Certificate, Release Notes, Ops Runbook.
3.4 Tokenization & Rights
Mint Master Agent Token (ERC-721) with metadata (hash, version, lineage, changelog).
Mint Stem Editions (ERC-1155) if needed (prompts, tools, context packs).
Attach Royalty Spec (ERC-2981) to master and/or license wrappers; optional rental via ERC-4907.
Register in Feeda Marketplace (see Feeda 10k / Litepaper for full spec).
Last updated