Agent Fan-Out Ledger — Parallel Exploration & Lineage-Tracking Protocol for Multi-Agent Systems

Give any parallel agent fan-out a manifest, an isolation boundary, and a lineage trail — so the branches that lose leave a paper trail instead of vanishing. A portable framework for Claude Code subagents, CrewAI crews, AutoGen group chats, LangGraph parallel nodes, or any orchestrator that runs more than one agent at a time.

The problem this solves

2026's agent tooling made parallel fan-out trivial — try three fixes at once, research four sub-questions at once. What it didn't solve is keeping track of why the branches that lost, lost. Most setups keep only the winning output and discard the reasoning behind every rejected branch, so the same dead end gets explored again next quarter because nothing recorded that it already failed.

What this kit does

• Defines the Branch Manifest — every parallel branch declares its hypothesis and success criteria before it runs, so results are comparable after the fact instead of judged in hindsight
• Specifies the Isolation Boundary — each branch works in its own scoped state so concurrent branches can never clobber each other or the trunk
• Defines the Lineage Record — every branch's decisions and dead ends are logged, not just its final answer, so rejected approaches stay searchable
• Includes the Reconciliation Pass — a fixed procedure for merging N branch outputs: dedupe agreement, flag unconfirmed findings, and surface contradictions instead of silently picking one
• Ships two worked examples (a parallel research fan-out, a parallel bug-fix exploration) and a named list of anti-patterns

Built for

• Anyone using Claude Code subagents, CrewAI, AutoGen, or LangGraph parallel branches to explore more than one approach at once
• Teams whose agents research, prototype, or debug in parallel and currently keep only the "winning" output
• Agencies productizing multi-agent workflows for clients who need a real answer to "how do we know the other branches were actually worse, and not just first to fail"

What's inside

• AGENT-FANOUT-LEDGER.md — the full methodology: the Branch Manifest, the Isolation Boundary, the Lineage Record, the Reconciliation Pass, both worked examples, and the anti-pattern list
• branch-manifest-template.md — a fill-in-the-blanks manifest format to open every branch with
• isolation-boundary-checklist.md — what "isolated" means for filesystem-based, git-worktree-based, and pure-context-based branches
• lineage-record-schema.json — a JSON Schema for the append-only lineage record
• reconciliation-pass-playbook.md — the dedup / flag-unconfirmed / surface-contradiction procedure, worked step by step
• quick-start.md — one-page mappings for Claude Code's Task tool and subagents, CrewAI, AutoGen, and LangGraph

How you get it

Your full kit — every section, every template, copy-paste ready — is hosted at ukiyoprod.com/pages/agent-fanout-ledger. Bookmark it; nothing else to download.

Author

Built by Ukiyo Productions — an original methodology developed in response to 2026's wave of parallel-agent exploration tools, including the git-worktree-based systems that popularized isolated parallel branches with lineage tracking for coding and research agents specifically. This kit generalizes that pattern to any multi-agent fan-out workflow, regardless of orchestrator. Not affiliated with, and not derived from the source code of, any specific open-source project. Part of a 100+ product marketplace built for founders, operators, and agencies working with AI agents.