set50-system/.hermes/plans/2026-08-25_dashboard-rebuild-multitheme-real-data.md

Dashboard Rebuild — Multi-theme + Real Thai Data

For Hermes: Execute task-by-task, commit per task, verify with backend tests + browser. Plan mode — no code until user approves.

Goal: Rebuild the SET50 dashboard so it reflects REAL multi-theme Thai data (tourism + auto + energy + NPL), not the current tourism-only fixture, per the user's 10-point review.

Architecture: Replace the single-vertical tourism_result-driven dashboard with a multi-theme real-data source; add per-symbol theme exposure (multi-source factors PER theme); update all UI sections to read real data, Thai-only, and remove the 5 obsolete KPI cards + paper ledger.

---

The 10 user complaints → fix

#	Complaint	Root cause	Fix
1	Theme Surprise ไม่เข้าใจ, ลบได้	KPI card from tourism-only	Remove card
2	สัญญาณที่ใช้งาน: ซื้อ 7 แต่โชว์ 8	total=8 (incl neutral) mislabeled	Show long/short/neutral correctly, Thai
3	สมุดบันทึก paper มีทำไม — ไม่มี paper trade แล้ว	legacy section	Remove (simulation replaces it)
4	คุณภาพข้อมูล — ข้อมูลจริงรึยัง?	TOURISM_SOURCE=fixture default	Switch to real BOT data; show real source+as_of
5	ประตู backtest (blocked) เอาออก	legacy gate	Remove card (research is exploratory/forward)
6	แต่ละธีมควรดูหลายแหล่ง (ประชากร ฯลฯ)	1 source/theme	Add multi-source factors per theme (see Data plan)
7	ตารางหุ้น + ตารางสัญญาณ ควรรวมกัน	split	Merge into one combined table
8	Theme surprise ไม่ครบ 3 ธีม	tourism-only	3 theme surprise (auto, energy, tourism)
9	ที่มาข้อมูล ควรเป็นตาราง (แหล่ง+เวลา)	prose	Table of all sources + last fetched
10	บันทึกการวิเคราะห์ มีแค่ 1 ธีม	tourism thesis	Per-theme thesis (3) + multi-source note

Overall: "ข้อมูลทั้งหมดตอนนี้ยังไม่ได้ใช้ข้อมูลจริงใช่ไหม?" → ใช่, dashboard ใช้ tourism_result (fixture default). Rebuild to real.

---

User-confirmed decisions (2026-08-25)

1. Real data ONLY — switch to BOT real (no fixture fallback); dashboard fails if no real data.
2. Theme surprise = z-score of each theme's headline factor (uniform across all themes).
3. Multi-source per theme (A+B): wire existing multi-source (auto = volume + NPL + production + export); AND add broader macro (population/GDP/inflation) — found BOT SDDS page (bot.or.th/en/statistics/sdds.html) is server-rendered and scrapable WITHOUT auth: provides GDP (Q2/2026 4,793.5bn), Private Consumption +4.9%, Private Investment +18.1%, Manufacturing −3.1%, Population 70,472k, Headline Inflation 1.95%, Unemployment 0.93%. This is the macro backdrop layer (complaint #6).

Macro backdrop factor (new) — BOT SDDS collector

• Create backend/app/macro_thai.py — scrape BOT SDDS HTML for: private consumption index %YoY, private investment %YoY, mfg production %YoY, population (thousands), headline inflation, core inflation, unemployment. Real Thai macro backdrop.
• Wire into a "macro" theme/context read + into the sources table.

Backend task list

Task A: Real multi-theme dashboard source module

• Create backend/app/dashboard.py — assembles REAL data from all collectors:
  • tourism (BOT real, via collector), auto_credit (Trading Econ + auto_npl BOT), energy (TOP), siamchart fundamentals, prices
  • Returns {themes: [{id, name, frequency, read_* , surprise}], board: [{symbol, combined_score, dividend, ...}], sources: [{from, source, as_of, fetched_at}], thesis_per_theme: {...}}
• Multi-source per theme (complaint #6): registry mapping theme → list of factor sources (not 1). E.g. auto theme = {new_car_sales_yoy (TradingEcon), auto_npl_pct (BOT), vehicle_production, auto_exports}. Add aggregatable fields.
• 3 theme surprise (#8): compute surprise per theme (z-score of the theme's headline factor), not just tourism.

Task B: Switch create_app to real data

• backend/app/__init__.py: default TOURISM_SOURCE → bot (real), or better: build multi_dashboard from collectors at startup, keep tourism as one theme. _load_default_snapshot no longer the dashboard root.
• dashboard/summary, /signals, /factors endpoints → read from dashboard.py real assembly, not tourism_result.

Task C: Sources table endpoint

• dashboard.py collects sources list: each {from (e.g. "TradingEconomics"), source, url, as_of, fetched_at, frequency}. Expose in /api/v1/themes + summary.

Task D: Per-theme thesis (#10)

• dashboard.py builds a short deterministic thesis per theme from that theme's factor readings (not a hardcoded tourism sentence).

Frontend task list

Task F: Remove 5 obsolete items (#1,#3,#5)

• Delete cards: Theme Surprise (#1), คุณภาพข้อมูล (#4 replaced by honest sources), ประตู backtest (#5), สมุดบันทึก paper (#3) → card 2 (สัญญาณ) only + 3-theme panel.
• Remove paper-ledger UI + backtest-gate UI blocks (already partially removed paper ledger; remove backtest KPI).

Task G: Correct signal count (#2)

• Show long / short / neutral as separate Thai chips; value = long only, with "+X ซื้อ −Y ขาย" clearly, no off-by-one.

Task H: Merge stock + signal tables (#7)

• Single combined table: [สีสัญญาณ, symbol, theme(s), combined_score, dividend, PE, EPS YoY, Yield, P/BV, ROE] — one table, sortable, dividend filter.

Task I: Sources table UI (#9)

• Render sources as a <table>: ตัวจาก, แหล่ง, ข้อมูล, as_of, fetched_at, frequency.

Task J: 3-theme surprise + thesis (#8,#10)

• Theme panel shows 3 themes, each with its own surprise + read + thesis.

Task K: Thai-only consistency + visual verify

• Audit all remaining English; keep only technical terms. Browser verify at 1440/1024/768. Screenshot for user.

---

Multi-source factor registry (complaint #6) — where to add data

For "แต่ละธีมดูข้อมูลรอบด้านของประเทศ" — extend themes to multiple Thai sources:

• Auto: new_car_sales_yoy (TradingEcon) + auto_npl_pct (BOT) + vehicle_production + auto_exports (TradingEcon) [already in collector]
• Energy: TOP net_profit/ebitda + Thai retail fuel/EPPO + (optional) import/refining
• Tourism: BOT tourism surprise + arrivals + hotel occupancy
• Macro backdrop (new): Thai population / GDP (BOT API — needs auth, mark deferred) per user's "จำนวนประชากร"

Note: Full multi-source (population/GDP via BOT portal API) requires API key registration — mark as a separate later task; this plan wires the already-available multi-source readings per theme.

Verification

• Backend: full suite passes; live /api/v1/themes shows 3 themes with real Thai readings + sources table + per-theme thesis; no fixture.
• Frontend: npm run build; browser shows combined single table, 3-theme surprise, sources table, no removed cards; Thai-only.
• Screenshot evidence for user review before push (push gate).

Open questions / risks

1. Real-vs-fixture default: Should dashboard run on BOT real data by default (needs a fetch), or keep a cache so it works offline? → Recommend: real BOT via daily cache, fallback to last-good.
2. Population/GDP macro: needs BOT portal API key (user registers). Defer unless user provides key.
3. Theme surprise definition: need concrete per-theme surprise metric — z-score on each theme's headline factor (auto YoY, energy net profit trend, tourism surprise). Confirm with user if ambiguous.
4. Push gate: after visual verify, user approves before push (per repo convention).