Spec: marketing-bot agent-readiness model + scorecard + sequencing (#71)
Grades the bot against the external 8-point "what an AI marketing agent needs to actually work" standard, reconciles it with the cockpit-not-doer doctrine, and sequences remaining work behind the real binding constraint.
Grades the bot against the external 8-point "what an AI marketing agent needs to actually work" standard, reconciles it with the cockpit-not-doer doctrine, and sequences remaining work behind the real binding constraint.
Current-state scorecard (origin/main, 2026-07-27): of the 7 cockpit-scoped requirements, 4 Strong (connected-data warehouse + GSC truth layer, durable Postgres action-memory, guardrails, statistical patience), 2 Partial (business context, observability), 1 Weak (outcome signal — North Star defined but its funnel terms uninstrumented). #3 (write/act) graded as by-design distributed, not a deficiency.
Thesis: cockpit-readiness is necessary but not sufficient — the constraint is upstream of all 8 (GDA is pre-traffic, ~3 clicks/28d; authority is the lever). Tier 0 = fuel (authority + price index + Street Delta, product-repo); Tier 1 = three cheap cockpit fixes (instrument the North Star series; alert the 2 silent crons; promote-gate floor + S4 re-check); Tier 2 = defer until post-traffic.
References existing specs, does not redefine them. Assessment doc, no code.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>