mcs enrich workflow

Use this workflow when generated build evidence should become reviewed semantic-package metadata. The proposal queue is a review buffer between machine suggestions and confirmed annotations.

Hard rules

All annotation writes go through proposals. Do not bypass the proposal queue. Apply only proposals you have reviewed against table context, evidence, and the user's stated semantics. Reject proposals that are unsupported, ambiguous, or conflict with confirmed annotations.

Workflow

1. Inspect package coverage:

bash mcs -f json status --tables Check has_ai_context and columns_with_description per table.

1. Promote build-time column-role suggestions into proposals:

bash mcs package propose --from-suggestions

1. Propose ai_context, column descriptions, and reviewed column semantics

via YAML: bash mcs package propose --from-stdin <<'EOF' tables:

• table: orders

ai_context: "Each row is one customer order event." columns: status: {role: dimension, dim_type: categorical, description: "Order lifecycle state (pending/paid/shipped/cancelled)."} total_amount: {role: measure, agg: SUM, description: "Raw order amount to aggregate for revenue."} EOF

1. List all proposals (role + ai_context + description together):

bash mcs package list-proposals

1. Inspect each candidate:

bash mcs package show-proposal <id>

1. Apply or reject after review (omitting --reason is valid):

bash mcs package apply <id> mcs package reject <id> --reason "<short reason>"

1. Re-check coverage:

bash mcs -f json status --tables

Use mcs -f json ... when you need to parse command output or compare proposal payloads programmatically.

See references/enrich.md for review policy and examples.