Trust & Data Health

How this dashboard is built, tested, and monitored — the trust ledger. Updated on every run.

At least one source is past its warn threshold (expected in CI with the pinned fixture). Live deploys are fresh — see generated_at.

`synthetic_sources = 1` — at least one API fell back to deterministic synthetic data. Commodity is always synthetic until the World Bank parser is wired. See ingestion manifest.

dbt PASS

dbt total checks

Lake snapshot

Field	Value
Generated at	2026-09-02T03:59:37.558611Z
Warehouse	ducklake
Ingestion status	synthetic_fallback
Synthetic sources	1
Freshness	raw_weather_forecast:pass, raw_fx_rate:warn, raw_commodity_price:warn, raw_weather:warn
Row counts (raw)	{"raw_weather": 360, "raw_weather_forecast": 21, "raw_fx_rate": 83, "raw_holidays": 8, "raw_commodity_price": 8}
dbt	92 / 93 PASS ( 0 error, 0 warn)

Every scheduled run appends this same ledger to the GitHub Actions Summary and, on failure, opens a deduplicated Issue. Lake snapshots are queryable at any past snapshot_id via DuckLake time travel.

Margin assumptions

The margin mart uses seeds/cost_assumptions.csv — fertilizer, harvest and transport costs per region plus extraction rate. Replace those seed values with your estate's actuals and the margin recomputes everywhere. Formula per effective harvest day:

revenue_per_ha = CPO_IDR_per_tonne * yield_t_ha * extraction_rate

margin_per_ha = revenue_per_ha - fertilizer - harvest - transport ; otherwise -fertilizer (you still fertilize even when you don't harvest).

See 7-Day Forecast for prescriptive outlook and Operations for historical planner.