Data Integrity

Every number on this site, sourced. A financial dashboard is only as trustworthy as its honesty about where the data comes from — so here is the whole ledger: what is live, what is a saved snapshot, what is computed, what is an honest illustration, and what is a placeholder. Nothing here is hidden, and nothing fabricated is dressed up as real.

22 of 34 surfaces carry real data (live / cached / computed / yours) · ledger as of 2026-06-18

YOURSYour own real data, entered or imported. PII is scrubbed before it is stored.

Example 1 · your real book/money · Mine vs Dream

SOURCEFund SuperMart statement

METHODHoldings imported, PII scrubbed (no name/account/advisor)

AS OF2026-05-30

ENGINEportfolio-data.ts REAL_HOLDINGS

⚠ LIMITA point-in-time statement; NAVs move daily.

LIVEFetched fresh from a public API on request, then edge-cached briefly.

Fear & Greed IndexDESK · Pulse

SOURCEfeargreedchart.com

METHODScraped composite, no key

AS OF≈5 min

ENGINE/api/* + fetchFearGreed

⚠ LIMITThird-party composite; methodology is opaque and US-centric.

THB / USDDESK · status strip

SOURCEBank of Thailand API

METHODOfficial daily reference rate

AS OFDaily

ENGINEfetchThbRate

⚠ LIMITReference rate, not a tradable quote; lags intraday FX.

US macro — Fed rate, CPI, yieldsDESK · Intel, /plan cycle

SOURCEFRED (St. Louis Fed)

METHODOfficial series via free API

AS OFDaily–monthly

ENGINE/api/macro · fetchMacro

⚠ LIMITOfficial data is revised; release lag of days to weeks.

Global & regional indices, assetsDESK, /markets

SOURCEYahoo Finance (free tier)

METHODQuote + history fetch

AS OF≈15 min delayed

ENGINEfetchAllRegional / fetchAssetClasses

⚠ LIMITUnofficial free feed; rate-limited, occasionally gappy.

World events & sentiment/events, /newsroom

SOURCEGDELT DOC 2.0

METHODEvent + tone query

AS OF≈15 min

ENGINE/api/world-news · /api/events

⚠ LIMITMedia-derived; correlation with prices is not causation.

Economic calendarDESK · Intel

SOURCEPublic calendar feed

METHODUpcoming releases

AS OFHourly

ENGINE/api/econ-calendar

⚠ LIMITConsensus estimates shift; times are exchange-local.

Market regime classification/regime, DESK

SOURCESupabase regime table (← causal pipeline)

METHODClassifier over macro features; deterministic fallback summary when DB is unavailable

AS OFDaily run

ENGINEingestion/regime.py · /api/regime

⚠ LIMITA probabilistic label, not a verdict; falls back to a synthetic summary without the DB.

Thai bond curves, spreads & turnover/yields, /api/thaibma

SOURCEThai Bond Market Association API

METHODAuthenticated government/corporate curves, average credit spreads, and EOD trading summary; daily JSON cache fallback

AS OFDaily after ThaiBMA publication

ENGINEsrc/lib/api/thaibma.ts · ingestion/bonds.py

⚠ LIMITPaid/trial credentials are required; without them the surface is explicitly marked illustrative.

Mr. Market Mood gaugeDESK + /money + subway

SOURCEFRED (VIXCLS, T10Y2Y, BAMLH0A0HYM2, NCBEILQ027S, GDP) + yfinance SET50 + live Fear&Greed

METHODVIX, yield curve, credit spreads from FRED; Buffett Indicator = corporate equities ÷ GDP; SET P/E market-cap-weighted from yfinance; each value scored −2..+2 by documented thresholds → composite mood. /api/mood overlays live Fear & Greed so every surface (gauge, Graham Diagnosis, subway P/E) reads one aligned source

AS OFdaily + live F&G

ENGINEingestion/market_mood.py → market-mood.json · /api/mood · src/lib/market-mood.ts · useMarketMood()

⚠ LIMITFear & Greed shows the labeled illustrative snapshot when its feed is unreachable (only marked live on a genuine response); S&P forward P/E is always illustrative (no free analyst-estimate source). Every signal carries its own LIVE/CALC/EST tag + as-of date.

CACHEDA committed JSON snapshot produced offline by an ingestion script.

SET50 daily OHLCV/trade, /scan, /api/technical

SOURCEyfinance (.BK tickers)

METHODDaily ingest → prices.json, edge-imported

AS OFLast ingest

ENGINEingestion/prices.py · fetch_all_to_json.py

⚠ LIMITEnd-of-day only; refresh depends on running the cron.

SET50 fundamentals — P/E, P/B, ROE, EPS, revenue/scan, /trade · Graham/Buffett scores

SOURCEyfinance (.BK tickers)

METHODDaily ingest via yfinance → fundamentals.json; Graham Number, Buffett Score, Margin of Safety computed from these figures

AS OFLast ingest

ENGINEingestion/fundamentals_refresh.py

⚠ LIMITYahoo Finance free API rate limits or limits the breadth of historical ratios; some sectors like banking use specialized metrics.

Thai macro — GDP growth, inflation, current account/plan · macro context, /signals

SOURCEWorld Bank Open Data (public API, no key required)

METHODWeekly ingest via World Bank REST API → macro snapshot; complements FRED with Thailand-specific official figures

AS OFLast ingest

ENGINEingestion/macro.py

⚠ LIMITWorld Bank data lags by quarters to a year; use for structural context, not near-term signals.

Efficient frontier & optimal weights/portfolio · Optimizer

SOURCERiskfolio-Lib over 3y SET50 returns

METHODMean-variance + HRP, Ledoit-Wolf cov → allocation.json

AS OFOn run

ENGINEingestion/optimize.py

⚠ LIMITOptimizers trust their inputs blindly; a reference, not advice.

Portfolio proxy history/money · Mine vs Dream

SOURCEyfinance proxy ETFs

METHODDaily proxy NAV → dream-history.json

AS OFLast run

ENGINEingestion/track_dream.py

⚠ LIMITProxy approximates the fund's driver, not its exact NAV.

Market Web — causal relationship graph/signals · MarketGraph

SOURCECurated market-graph.json + granger.json + market-states.json + signal-edge.json + narratives.json

METHODStatic trader-knowledge graph enriched offline: Granger edges to SET proxy, KMeans regime on cycle node, narrative clusters on tickers; trader lens filters day/swing/value

AS OFLast cache ingest

ENGINEingestion/market_graph.py · src/components/Desk/MarketGraph.tsx

⚠ LIMITCurated edges are conceptual; Granger ≠ causation or tradable edge. Regime chips are analogues, not forecasts.

COMPUTEDCalculated from real inputs — indicators, compounding, optimization.

Technical indicators/trade

SOURCESET50 OHLCV (above)

METHODEMA/RSI/MACD/BB/VWAP/ADX… computed in-app

AS OFPer render

ENGINEsrc/lib/technical.ts · trade-insights.ts

⚠ LIMITIndicators describe the past; they do not predict it.

Retirement projections/plan

SOURCEYour inputs (age, salary, savings)

METHODMonthly compounding + inflation + VUCA haircut

AS OFLive

ENGINEuse-plan-state.ts

⚠ LIMITNominal assumptions; real life is not a smooth curve.

Market States + today's analogue/patterns

SOURCEFRED T10Y2Y + yfinance ^GSPC/^VIX (2004–now, daily)

METHODKMeans (k=5, unsupervised) on z-scored VIX, yield curve, S&P 20d return and 20d vol → market regimes; today's analogue = nearest neighbours in feature space, with the S&P 500's actual forward 30/90-day returns

AS OFdaily

ENGINEingestion/signal_history.py → market-states.json · src/lib/market-states.ts

⚠ LIMITAnalogues are statistical neighbours, not forecasts — small samples, and the future need not rhyme. Credit spreads are excluded from the model (FRED truncates the free series); the SET is not in the feature set (it tracks global states via US data).

Granger causality — what statistically precedes SET?/patterns

SOURCEFRED (VIX, yields, Fed rate) + yfinance (S&P, Oil, DXY, BTC, Gold, Copper, JPY, THB, Nikkei, HSI)

METHODGranger F-test at lags 1–10 for each series vs SET returns; optimal lag by AIC; significant at p<0.05. Structural context, not signals — Granger causality ≠ tradable edge.

AS OFdaily

ENGINEingestion/granger.py → granger_latest.json → src/lib/data/cache/granger.json

⚠ LIMITGranger causality is a statistical property of the historical sample. Relationships break as markets adapt. Does NOT imply a live trading edge.

Forecast calibration — does the 80% band cover 80%?/signals · ForecastCalibration

SOURCE^SET.BK historical prices (yfinance)

METHODWalk-forward ARIMA(1,1,1) pseudo-backtest: 120 windows × 1yr training → 5-day forecast → actual. Coverage80 = fraction of actuals inside the 80% interval.

AS OFon-run

ENGINEingestion/calibrate_forecast.py → src/lib/data/cache/calibration.json

⚠ LIMITRetrospective backtest, not live out-of-sample tracking. Model re-fits on each window. Coverage can change as market volatility regimes shift.

Signal Edge — does the signal work?/patterns

SOURCEFRED T10Y2Y + yfinance ^GSPC/^VIX (2004–now, daily)

METHODConditional forward-return analysis — the S&P 500's actual 30/90-day forward-return distribution by market state and VIX / yield-curve bucket, measured against the unconditional baseline; edge = conditional mean − baseline mean (pp)

AS OFdaily

ENGINEingestion/signal_edge.py → signal-edge.json · src/lib/signal-edge.ts

⚠ LIMITHistory, not a forecast. Where a bucket ≈ the baseline it carries no usable edge — and that is shown, not hidden. Small buckets (n<30) are noisy; read the dispersion (p25–p75), not just the mean.

ILLUSTRATIVEAn honest model or baseline. A shape to reason with, never a promise.

Forward View — 90-day cones/signals, /money hover

SOURCEReal 1y proxy prices

METHODDrift + volatility cone (median + 10/90 band); upgrades to TimesFM on the M3

AS OFOn run

ENGINEingestion/forward_view.py

⚠ LIMITA statistical range from past drift/vol — not a learned forecast yet.

Dalio Big-Debt-Cycle stages/signals web, /plan

SOURCERay Dalio, Big Debt Crises (2018)

METHODPattern match across ~16 historical cycles

AS OFStatic framework

ENGINECycleIntelligencePanel · plan-data

⚠ LIMITStructure recognition, no fixed duration; not a market call.

SET 5-day forecastDESK · forecast

SOURCESET history + macro covariates

METHODdarts LightGBM (ARIMA fallback), 80/95% bands

AS OFDaily run

ENGINEingestion/forecast.py

⚠ LIMITBacktested accuracy varies by regime; bands are wide for a reason.

Kronos foundation model forecast (SET)/signals · KronosForecast

SOURCEKronos-mini (NeoQuasar/Kronos-mini, 4.1M params, pretrained on 45+ global exchanges)

METHODZero-shot ARIMA-comparable forecast on SET50 tickers. 400-day context, 20-day prediction horizon. 3 sample paths averaged. Bands are vol-cone (95% CI), not model uncertainty.

AS OFdaily

ENGINEingestion/kronos_forecast.py → src/lib/data/cache/kronos-signals.json

⚠ LIMITZero-shot transfer — not fine-tuned on SET-specific dynamics. Not yet validated against SET forward returns. Treat as one additional perspective alongside the vol-cone baseline, not a prediction.

REFERENCEA historical fact or static reference figure.

2024 world index returns/plan

SOURCEPublished full-year 2024 figures

METHODStatic reference table

AS OFFY2024

ENGINEplan/page WORLD_2024

⚠ LIMITHistorical context only; past returns ≠ future.

Dream · Ideal portfolio/money · Mine vs Dream

SOURCEHindsight long-term allocation

METHODEducational scenario with real fund codes

AS OFStatic

ENGINEportfolio-data.ts HOLDINGS

⚠ LIMITHindsight allocation — what one *could* have held, not a record.

PENDINGPipeline built and ready — awaiting credentials or the first data pull.

Earth Signal — satellite alt-data/signals

SOURCEAlphaEarth Foundations (DeepMind) via Google Earth Engine

METHOD64-dim embedding year-over-year shift over Thai economic sites → SET tickers

AS OFAwaiting GEE auth

ENGINEingestion/alphaearth.py

⚠ LIMITNo numbers shown until real embeddings are pulled — never fabricated.

Usage analyticsapp-wide

SOURCEFirebase / Firestore

METHODPage-view + trade events

AS OFAwaiting env keys

ENGINElib/firebase · AnalyticsObserver

⚠ LIMITRuns in harmless mock-key mode until NEXT_PUBLIC_FIREBASE_* is set.

MOCKSynthetic placeholder so the app renders with zero API keys. Never trade on it.

Research committee preview/research

SOURCELocal agent contract registry + deterministic brief engine

METHODMacro/value/tape/event/portfolio/risk roles compose a research workflow simulation from mock data

AS OFPer render

ENGINEsrc/lib/research-agents.ts · /api/research/brief

⚠ LIMITWorkflow preview only: it uses mock data and does not call a live model or live market feed.

Scanner fundamentals/scan

SOURCEMOCK_STOCKS

METHODSynthetic Graham/Buffett metrics

AS OF—

ENGINE/api/scanner (no Supabase)

⚠ LIMITPlaceholder until the Supabase fundamentals feed is connected.

Pulse breadth / indicesDESK · Pulse fallback

SOURCEMOCK_INDICES / MOCK_SET

METHODSynthetic breadth + index levels

AS OF—

ENGINE/api/pulse

⚠ LIMITFallback when the live feed is unavailable.

Graham Diagnosis — portfolio autopsy/money · Mine vs Dream

SOURCEHardcoded Dream-vs-Real comparison

METHODFive Graham checks (entry mood, P/E<15, diversification, Fear&Greed, defensive hedge) → 5/5 vs 0/5

AS OF—

ENGINEDreamPortfolio.tsx GAP_ANALYSIS

⚠ LIMITHand-set figures to teach Graham's margin-of-safety lesson — not measured from the live books.

Maintained by hand in src/lib/provenance.ts — keeping it honest is a deliberate act, not an automated one. Research preview · not financial advice.