Implemented
Data contracts and fixed runs
Discharge, clocks, provenance, nulls and one snapshot across map, chart and export.
System Design v2 · implementation status
What works, and what comes next
MekongIQ’s goal is actionable river-flood intelligence. A traceable source and careful reporting workflow provide the first foundation.
Evaluation adapter
External forecast
Source adapter, central cache and immutable manifests are built. The public release uses synthetic data; live retrieval is disabled pending approval. Reach matching and local skill remain unverified.
Fixture rehearsal
Citizen reports
Local walkthrough plus authenticated owner-scoped fixture records. Real image intake closed.
Built; hosted testing pending
Private storage and account boundaries
Native D1/R2 and dispatch identity. Reviewer allowlist starts empty.
Disabled
Safety and image AI
No external provider calls or real photo processing.
Not implemented
Local flood timing, extent and depth
Requires basin data, model calibration and independent validation.
Disabled
Warnings and operational actions
Requires responsible institutional authority.
Later phase
Irrigation
Needs water accounts, crop demand and delivery constraints.
One evidence trail
- Allowlisted source and bounded fetch
- Schema, units, clocks and grid checks
- Immutable raw archive and run manifest
- One fixed snapshot for maps, charts and explanations
Citizen fixtures stay in a separate private path. A manual evidence review cannot publish a report or update a forecast. Local scientific models and optional AI are future extensions, with their own approval and validation gates.
No deployed “Private Brain” or hidden forecasting engine is claimed by this release.