Build G2 Cost Pre-Flight as Runwal-local guardrail inside fivetran-mcp…
Decision
Build G2 Cost Pre-Flight as Runwal-local guardrail inside fivetran-mcp, diverging from upstream Bible framework v19.1.3 which REMOVED G2. Remap scope doc’s database/schema/column references from aj-ea’s assumed RUNWAL_SNOWFLAKE_PROD.FIVETRAN_METADATA single-schema Platform Connector layout to the live RNWL_RAW.{FIVETRAN_METADATA, FIVETRAN_METADATA_FIVETRAN_PLATFORM} dual-schema reality discovered in Phase 1.0 pre-flight. Pause before Phase 1 DDL execution for explicit AJ authorization on production Snowflake modification.
Rationale
Phase 1.0 live-state probe via snowflake MCP caught 6 drifts from scope: (1) database RUNWAL_SNOWFLAKE_PROD does not exist — actual is RNWL_RAW; (2) Fivetran metadata is split across two coexisting schemas not one; (3) 4 of 6 target tables are named differently in the Platform Connector’s dbt-transformed layer (FIVETRAN_PLATFORM__* prefix); (4) column names diverge — MEASURED_DATE not measured_timestamp, CONNECTION_NAME not connector_id, requires CONNECTION table join; (5) rows_in_source_table column used in scope § 7.2 does not exist — must use MAX(TOTAL_MONTHLY_ACTIVE_ROWS) over 12mo as confidence=medium proxy; (6) mar_tier column does not exist — Fivetran migrated to flat-rate per-destination pricing. Upstream framework removal signal (contract_enforcer.py:17, contract.yaml:148-153, SKILL.md:258) was invisible to aj-ea when scope was drafted; AJ resolved in-session with explicit Runwal-local divergence directive. Scope § 6.1 DDL would have failed at first GRANT statement if executed verbatim. Live data validated scope’s Salesforce sanity-check ranges (46k rows/day sits inside 10k-500k) and surfaced a stronger G2 target: SAP HANA at 3.3M/30d (2.4x Salesforce). Pausing for AJ go-ahead before provisioning role/user/warehouse in production Snowflake respects auto-mode rule 5 (production modification needs explicit confirmation) and feedback_autonomous_by_default carve-out (strategic + irreversible boundary).
Alternatives Rejected
Outcome
Pending
Related
- docker
- freeze-bible-v190-as-authoritative-framework-for-all-future
- salesforce
- snowflake
- 2026-04-04-oracle-001-self-architecture-analysis
- blue-capability-stack-hybrid-mcp-primary-ai-surface-cli-v066
- proceed-to-phase-3-data-layer-with-three-explicit-bias-corre
- shipped-pure-observability-cost-observer-on-fivetran-mcp-opt
- researchclaw-v040-ship-14-tool-streamable-http-merged-server
- snowflake-mcp-upgrade-plan-approved-for-execution-pending-aj
- for-apex-test-class-deployment-via-salesforce-mcp-split-mono
- openclaw-gpt-55-migration-target-pivots-gpt-55-does-not-exis
- salesforce-mcp-v710-must-implement-r26-complete-capability-i
- adopt-phased-tableau-cortex-analyst-alignment-phase-0-1-defi
- recommend-creating-a-new-sfdc-formula-field-loanconsolidated