Transform Power BI dashboards into executive intelligence cockpits that explain what changed, why it matters, and what to do next — with evidence, confidence, and semantic discipline.
Two operating modes
Start with pbi-exec-intel-orchestrator — it routes the request, validates
inputs, sequences the right specialist skills, and enforces the evidence contract end to end.
Use when you already have a Power BI dashboard and want to upgrade UX, KPI clarity, executive narrative, semantic-model readiness, performance awareness, action-center logic, and conversational BI behavior — audited, then re-architected into a cockpit.
Use when you have a business briefing and need a complete, implementation-ready dashboard blueprint from zero: business questions, KPI architecture, semantic requirements, page architecture, wireframes, conversational BI, action center, backlog, and acceptance criteria.
Evidence contract
No claim is asserted as fact without source, period, and confidence. Each insight,
finding, and recommendation attaches a structured evidence_object (schema v1.0).
{
"claim": "...",
"evidence_status": "supported | insufficient_evidence
| assumption_to_validate | illustrative_only",
"source": "...",
"period": "...",
"filters": {},
"confidence": "high | medium | low | insufficient",
"evidence_ids": [],
"notes": "..."
}Rule: no source ⇒ status must be
insufficient_evidence or assumption_to_validate. supported
requires source + period + non-insufficient confidence.
Evidence levels
Confidence is bounded by what you supply. The pack is honest about it — the verdict ceiling rises with artifact completeness.
| Quick Audit (screenshots-only) | Full Certification | |
|---|---|---|
| Inputs | screenshots + business objective + audience (KPI list recommended) | screenshots + page inventory + KPI definitions + semantic model metadata + Performance Analyzer export + data sources + business context |
| Useful for | UX, visual hierarchy, executive narrative, KPI clarity, page architecture + wireframe suggestions | the above + semantic certification, measured performance, grounded conversational layer — implementation-ready certification |
| Semantic / performance | certification blocked → insufficient_evidence; model ideas as assumption_to_validate | scored with evidence; AI-readiness certified on governed measures |
| Max QA verdict | Go for redesign / spec exploration (never client-ready build) | Go / Go-with-conditions / No-Go (client-ready possible) |
Full-page screenshots of every page; page inventory or PBIP notes.
Tables, measures/DAX, relationships, cardinality, RLS metadata.
Power BI Performance Analyzer export (JSON/CSV).
KPI list with definitions — formula intent + grain.
Audience/personas, decisions to support, data sources, refresh cadence.
Agreed targets/benchmarks — or they're flagged assumption_to_validate.
What you get
Scored audit per dimension with evidence + confidence; unmeasured dimensions marked insufficient_evidence.
Each call traced to a finding — clutter removed, plumbing kept, narrative rebuilt.
Executive Cockpit → Drivers → Diagnostic Triage → Detail → Action Center.
Per-page layout, visual choices with rationale, narrative zones. No KPI walls.
Star schema, governed measures, descriptions/synonyms, RLS — gated by certification.
Verified-answer catalog grounded in certified measures; refuses unsupported questions.
Recommended actions, thresholds, owners, alerts, decision log, outcome tracking.
Ranked items routed to owning skills/teams, with blockers made explicit.
Testable statements a BI/dev team can build and verify against.
Example narrative shape: "Blended ROAS 2.1x vs 3.0x target ILLUSTRATIVE_ONLY" — placeholder only; never used as a real baseline or recommendation.
Safety & privacy
ILLUSTRATIVE_ONLY.Get started (internal)
These are guidance anchors for the private engagement flow — not live processing.