Conductor product

Conductor is a knowledge base for Hector's multi-line organization. It provides descriptive guidance organized as markdown documentation with a minimal MCP transport for discovery. The compiler has been archived to archive/compiler-v1/. Conductor does not perform the work, own project state, or synchronize production tools.

Hector is President and the sole human policy authority. Agents act as staff in designated roles. Conductor has delegated authority to approve ordinary execution plans that comply with current policy. Exceptions, overrides, policy changes, and explicitly human judgment gates require Hector.

Part of the canonical org model. See Organization for the org hierarchy and data model.

Product boundary

Conductor does:

• Read Initiative and workstream context from Notion.
• Discover relevant policy, patterns, prior work, and unresolved context.
• Provide descriptive stage contracts and workflow guidance.
• Document business lines, roles, tooling, and standards.
• Govern documentation through normal PR review.

Conductor does not:

• Replace Notion as the Initiative system of record.
• Replace GitHub, editorial tools, client folders, or runbooks as production systems.
• Require duplicate work contracts in line tools.
• Perform authorized work or update Notion for agents.
• Approve policy changes, exceptions, or human judgment gates.
• Coordinate a whole multi-role workflow.

Design principles

• Portable context: an agent should receive the same governing context from any computer, phone, folder, or remote runtime.
• Solve once, retrieve thereafter: agents may reason about new problems but should not rediscover accepted decisions or repeat known failures.
• Compact by default: include only context that can affect the current plan; expand referenced units on demand.
• State over memory: agents are ephemeral. Durable Initiative state, evidence, decisions, and next actions must live in authoritative tools.
• Standardize repetition, not novelty: use accepted patterns when they fit; preserve an explicit exploration path when they do not.
• Slow down only to prevent rework: each required question, field, stage, or artifact must support authority, safety, evidence, or a downstream decision.
• Enterprise clarity without enterprise bloat: borrow useful separation of responsibility and definitions of done, not committees or ceremony.
• Learning remains governed: experience may create a candidate improvement; repetition and evidence justify policy, not one anecdote.