SQL+LLM Dashboard Feasibility & Will McKinley Feedback Analysis

Date: 2026-02-17 Author: Claude (Opus 4.6) Project: PILOT Source: AZ1 Monday Check-In transcript + will-mckinley-action-prompts.md

---

1. SQL + LLM Dashboard Generation: Feasibility

Answer: Already Implemented

The generate-briefing.py (961 lines) implements a 5-phase SQL+LLM pipeline:

Phase	Engine	What It Does
1. Query	SQLite	15 parameterized SQL queries via BriefingDB class
2. Compute	Python	MetricComputer derives velocity, costs, quadrant status
3. Narrate	Claude API	NarrativeGenerator with SHA256 hash-based caching (Sonnet for decisions, Haiku for bulk)
4. Assemble	Python	JSXAssembler builds JSON manifest for JSX rendering
5. Validate	Python	BriefingValidator checks structural integrity

Architecture Details

SQL Layer (init-database.sql v3.0.0):

• 22 tables with ADR-119 multi-tenant columns
• 5 pre-computed views: v_sprint_summary, v_track_progress, v_velocity_metrics, v_decision_status, v_risk_summary
• 14 indexes for query performance

LLM Layer (NarrativeGenerator):

• Model selection: Sonnet for decision narratives, Haiku for bulk metric summaries
• SHA256 hash-based narrative caching in narrative_cache table
• Prevents redundant API calls when underlying data hasn't changed

Output Layer (JSXAssembler):

• Produces JSON manifest consumed by csuite-briefing.jsx
• Supports multiple audience presets: csuite, board, engineering, investor

Conclusion: The SQL+LLM pattern is architecturally sound and production-ready. No new infrastructure needed for the core pipeline.

---

2. Will McKinley Feedback: 15 Action Prompts

Source: will-mckinley-action-prompts.md (239 lines, 7 categories)

2.1 Coverage Matrix

#	Prompt	Priority	Status	Details
1.1	Incremental Deployment Pipeline	CRITICAL	NOT COVERED	CI/CD for stable-level promotion
1.2	Live Application Preview	CRITICAL	NOT COVERED	Persistent URL + feature flags
2.1	Executive Summary Dashboard (Top 5)	CRITICAL	COVERED	csuite-briefing.jsx + decisions table
2.2	Eliminate Dashboard Redundancy	HIGH	PARTIAL	JSX exists, needs redundancy audit
2.3	Session Log -> Dashboard Generator	HIGH	COVERED	generate-briefing.py Phase 1
2.4	Multi-Session Cumulative View	HIGH	PARTIAL	sessions table exists, need aggregation query
2.5	Category-Level Progress	HIGH	COVERED	v_track_progress view + category field
3.1	Kanban Column Reorder	MEDIUM	PARTIAL	Config change in existing component
4.1	Visual Annotation Feedback	HIGH	NOT COVERED	Canvas overlay + structured feedback
4.2	Feedback-to-Action Pipeline	HIGH	NOT COVERED	Multi-format classification + task gen
5.1	Authoring/Output Separation	CRITICAL	COVERED	ADR-209 D.5 mandates portable output
5.2	Idea Analysis Dashboard	MEDIUM	PARTIAL	Needs new route + data model
6.1	Multi-Tenant Dashboard Framework	HIGH	COVERED	ADR-119 columns on all 22 tables
7.1	Decision Brief Structure	HIGH	COVERED	decisions table: type, confidence, options, stakes
7.2	Development-Over-Time Trends	HIGH	PARTIAL	v_velocity_metrics exists, need Recharts components

Summary: 6 covered, 5 partially covered, 4 not covered

2.2 Gap Analysis

Fully Covered (6/15) - No new work needed

• 2.1, 2.3, 2.5, 5.1, 6.1, 7.1

Partially Covered (5/15) - Extension of existing work

• 2.2: Audit csuite-briefing.jsx for duplicate sections, consolidate
• 2.4: Add WHERE created_at BETWEEN ? AND ? to BriefingDB + time-window selector in JSX
• 3.1: Change column order config in Kanban component
• 5.2: New route + lightweight ideas table, reuse dashboard framework
• 7.2: Create 3-5 Recharts components fed from v_velocity_metrics view

Not Covered (4/15) - New capabilities required

• 1.1: Incremental deployment pipeline (CI/CD milestone mapping)
• 1.2: Live application preview (persistent URL, feature flagging, stub system)
• 4.1: Visual annotation (screenshot capture, canvas overlay, structured feedback schema)
• 4.2: Feedback-to-action (multi-format classification, task generation)

---

3. Priority Recommendations

Tier 1: Immediate (leverage existing SQL+LLM pipeline)

Action	Effort	Impact	Prompt
Refine executive summary (top 5 + decisions)	LOW	HIGH	2.1, 2.2
Add multi-session time-window queries	LOW	HIGH	2.4
Create Recharts trend components	MEDIUM	HIGH	7.2

Rationale: These deliver immediate value using existing infrastructure. The SQL views already compute the data; we just need better presentation.

Tier 2: Near-Term (moderate new work)

Action	Effort	Impact	Prompt
Kanban column reorder	LOW	MEDIUM	3.1
Idea analysis dashboard	MEDIUM	MEDIUM	5.2
Incremental deployment pipeline	HIGH	CRITICAL	1.1

Rationale: 3.1 and 5.2 are straightforward. 1.1 requires CI/CD work but addresses Will's strongest directive.

Tier 3: Strategic (new capabilities)

Action	Effort	Impact	Prompt
Live application preview	HIGH	CRITICAL	1.2
Visual annotation feedback	HIGH	HIGH	4.1
Feedback-to-action pipeline	MEDIUM	HIGH	4.2

Rationale: Will returned to visual annotation multiple times - it's a core product feature, not a nice-to-have. Live preview is his strongest overall directive.

---

4. Key Insight

Will's core frustration: seeing documents about software rather than the software itself.

The SQL+LLM pipeline already transforms raw project data into executive-grade intelligence rendered as interactive dashboards. This IS the mechanism Will wants (prompts 2.1-2.5, 7.1).

The gap is not architectural - it's deployment visibility (1.1, 1.2) and feedback loops (4.1, 4.2).

The SQL+LLM pipeline generates excellent dashboards. The question is whether stakeholders can see and interact with the live application as it's being built.

---

5. Recommended Execution Order

Phase 1 (Week 1): Tier 1 refinements
  - Dashboard redundancy audit + consolidation
  - Multi-session cumulative SQL queries
  - Recharts trend visualization components

Phase 2 (Week 2): Tier 2 extensions
  - Kanban column reorder
  - Idea analysis dashboard route
  - Incremental deployment pipeline design

Phase 3 (Week 3-4): Tier 3 capabilities
  - Live preview system architecture
  - Visual annotation prototype
  - Feedback-to-action pipeline

---

6. Files Referenced

File	Purpose
tools/web-publishing-platform/executive-briefing-new-component/generate-briefing.py	5-phase SQL+LLM pipeline (961 lines)
tools/web-publishing-platform/executive-briefing-new-component/init-database.sql	22 tables + 5 views + 14 indexes (551 lines)
tools/web-publishing-platform/executive-briefing-new-component/will-mckinley-action-prompts.md	15 action prompts from AZ1 Monday Check-In
tools/web-publishing-platform/executive-briefing-new-component/executive-briefing-generator.md	Master system prompt v2.0.0

---

Analysis Document Version: 1.0.0 Generated: 2026-02-17T05:29:41Z