GroundTruth PM
A report-first visibility and reconciliation system for remote land-based assets - so an absentee owner can see what is actually happening, not just what they were told.
The messy workflow
An owner controls a large rural estate - thousands of hectares - but manages it remotely, from another country. Updates arrive constantly: WhatsApp photos, videos, voice notes, verbal reports. The problem isn't a lack of updates. It's that the updates aren't reconciled. The owner sees activity, but not progress; receives information, but not control.
The root issue is information asymmetry: the people doing the work also control the evidence of the work. GroundTruth PM exists to break that pattern - calmly, and without accusation.
The core insight
WhatsApp updates are claims, not control. A photo or voice note can be useful, but none of it should automatically count as control-grade truth. The product treats every update as a claim until it is reconciled against independent evidence - not because anyone is assumed to be lying, but because remote operations require independent verification.
Inputs
- Satellite observations (the independent spine)
- WhatsApp images, videos and voice notes
- GPS-tagged field photos
- Receipts, labour records and PM notes
- Survey / boundary data and historical imagery
Output
- A weekly visibility brief - the product artefact
- Zone status with confidence levels
- Claims reconciled against available evidence
- Readiness gates for time-sensitive activity
- Decisions needed, with owner and deadline
The core mechanic
Every update is a claim; every claim needs evidence; every evidence chain carries a confidence level; every uncertainty becomes a next action. This mechanic runs through the UI, the data model and the report copy alike.
The AI layer - and its limits
AI is a productivity layer, not the brand. It parses WhatsApp updates into structured claims, summarises voice notes, extracts dates and locations, flags missing evidence, and drafts the weekly report. It does not make legal conclusions, determine ownership or legality, accuse anyone, or replace surveyors, lawyers, agronomists and mining experts. AI structures messy information; judgement stays human.
Guardrails
- Calm, evidence-first language - claims and observations, never accusations
- Confidence is a feature: high / medium / low / requires verification
- Satellite is independent context, not automatic proof
- No legal conclusions - outputs may support professional review
- Every figure in the build is clearly labelled demo data
Status
Working v0.1 build - report-first, with a dashboard, claims table, evidence inbox, a satellite lab of mock scenes, and a printable weekly brief. Private for now; a walkthrough is available on request.
What I learned
In a sensitive, contested context, a product is only trustworthy if it refuses to overclaim. The hardest design work wasn't the model - it was the discipline of language and scope: staying with visibility rather than reaching for control, and treating confidence levels and careful wording as first-class features rather than uncertainty to hide. That restraint is what turns noise into decisions a remote owner can actually act on.