SEP-073: Chart generation from structured data in deep research reports

[samudzi]

Summary

Generate matplotlib charts from Zorora's structured data (SAPP, Eskom, FRED, World Bank) and embed them in deep research output alongside text synthesis.

Motivation

Published research articles include 4-7 data visualizations per piece. Text-only reports cannot match this quality bar. Zorora already has 64+ time series and feasibility analysis generates matplotlib charts (SEP-045). The data and charting capability exist but are not connected to the deep research output pipeline. Visual-first output is core to Zorora's product vision.

Proposed Change

• When market context is injected (SEP-070), also generate charts for relevant series
• Use LLM-generated chart specs (title, series, date range, chart type) based on query
• Render charts as base64 PNG embedded in synthesis markdown
• Reuse matplotlib patterns from feasibility analysis (SEP-045)
• Support line (trends), bar (comparisons), stacked area (composition) charts

Acceptance Criteria

• Deep research reports include 2-4 embedded charts from Zorora's structured data
• Charts are contextually relevant to the research query
• Charts render in the web UI alongside synthesis text

Dependencies

• Requires SEP-070 (market data in synthesis) for data availability
• Benefits from SEP-068 (structured output) for section-chart alignment

Priority

P2

[samudzi]

Implemented in commit 6293eec. generate_research_charts() creates matplotlib charts from market data as base64 PNGs. _build_market_context_for_query returns (str, dict) tuple. synthesize() accepts market_summaries. 32 new tests + 79 existing all pass (111 total).