gp-data-platform

Use when a Serve-product data concept — EO activation, Serve engagement/MAU, People Served, the onboarding funnel, Serve Amplitude events, hygiene filters — needs to resolve to its governed metric, table, column, or definition before a query is written. Not for running analyses (that is the analytics-process skill).

Use when a Win-product data concept — engagement, activation, outcomes/win rate, viability, segmentation, Amplitude events, joins, PMF/satisfaction — needs to resolve to its governed metric, table, column, or definition before a query is written. Not for running analyses (that is the analytics-process skill).

Use when planning, executing, reviewing, or calibrating a product analysis, or when asked a product data question that should become one (the entry runbook routes by product — Win and Serve today). For the data facts themselves, use the product's knowledge skill (win-analytics-knowledge or serve-analytics-knowledge).

Use when investigating a Databricks or dbt Cloud cost/usage increase, attributing spend across warehouses, clusters, jobs, models, and tests, or answering "what is driving our compute cost". Produces an interactive HTML cost dashboard from system tables.

Audit and quantify gp_api product campaigns that have no matching DDHQ election result, for a given election-date window. Classifies every unmatched candidacy into a definitive reason (never ran / race not present in DDHQ data / never made the ballot / product wrong-office or wrong-date / DDHQ omission / genuine matcher false-negative) using a deterministic SQL pre-pass against DDHQ staging plus a web-verification subagent fan-out, then writes an external aggregate summary CSV, an internal per-record detail CSV, and a confirmed-winners list. Use when asked to understand or quantify why product users tied to a 2026 (or later) campaign are not matched to election results, or to refresh the prior DDHQ miss audit.

Match an external or ad-hoc table (HubSpot export, lead list, P2V sample) to ICP flags or other reference data in Databricks. Use when you have a table of people or races and need to attach int__icp_offices flags (icp_office_win / icp_office_serve, voter_count, etc.) but the table only has emails, BallotReady race IDs, or position IDs, not the ICP join. Routes by which ID you have: email (two-hop via campaigns), race ID (two-hop via race staging), or position ID (direct join). General Databricks matching reference; not Win-pipeline specific.

Audit entity resolution match results for quality — analyze coverage, false positives, and false negatives

Triage and remediate L2 uniform schema drift using `l2_uniform_schema_preflight` output. Use when dbt runs touching `int__l2_nationwide_uniform` or `int__l2_nationwide_uniform_w_haystaq` fail, or when `L2_PREFLIGHT|` lines report drift (`stg_minus_src`, `src_minus_stg`, `target_minus_src`, `relation_missing`). Parse preflight logs, generate deterministic safe commands and manual instructions, and produce operator-ready next steps.