WaterRadar is a water-sector intelligence desk built on one rule: publish only what a named human editor has verified against its source — and show the reader exactly how confident we are.
WaterRadar covers treatment, desalination, infrastructure, aquaculture, water quality, policy, companies, products and research across every major jurisdiction — and turns that firehose into a queryable, sourced, confidence-scored record.
The professional reading WaterRadar is making a capital or compliance decision, so the product is engineered for one thing above all: a claim you can trust, with the provenance one click away and the confidence stated plainly. The compliance tracker is the moat — accuracy held over time, not a news stream.
Every published claim passes the same pipeline. This is the machinery behind the confidence badges you see across the site.
A writer may only assert facts present in the researcher’s dossier, and every factual sentence carries the ID of the claim it came from. A sentence with no claim ID is an orphan— caught mechanically and blocked, never published. With grounding, “is this hallucinated?” becomes a checkable property.
Three reviewers run in parallel, blind to each other, each prompted to refute, not approve, each told “default to refuted when unsure”.
Keeps verified only if it survives ≥2 of 3 lenses; majority-refute kills it. A split verdict is a forced human read.
Every claim carries a confidence value set from source trust and updated by the panel. It is data on the claim, not a phrase in the prose — so it can’t get lost as the claim propagates into the tracker, search and entity pages. Always icon + label + colour, never colour alone.
Unverifiable-but-newsworthy claims may run, but only explicitly hedged and labelledfrom a fixed vocabulary — never freeform. If more than 30% of a piece’s claims are unverified, it isn’t a story yet; it holds. Hedged claims never write to the entity database as facts and never appear in a headline.
A trust vertical whose parent sells into the space it covers carries an obvious conflict. Left implicit, it quietly discounts every verified claim. So we make it explicit.
An independent water-intelligence desk.
WaterRadar publishes at effluo.waterdoctor.com.sg, a commercial operator in the same water sector WaterRadar covers. Sales and commercial interests do not steer coverage or verdicts.
A persistent, plain line — “An independent water-intelligence desk” — sits in the masthead, here, and the footer of every page, alongside this editorial-independence statement.
The publisher, its partners, and its direct competitors get the identicalgrounding, sourcing and adversarial-panel treatment as any other entity — no softening, no boosting. The publisher’s own commercial space clears a higher newsworthiness bar, not a lower one.
Any published mention of WaterDoctor or a direct competitor is labelled as a related / parent party so the reader can weigh it — the same way a single-source claim is surfaced.
Any piece that materially praises or criticises the publisher or a direct competitor is a forced human read. The editor owns that conflict explicitly — it never auto-clears.
WaterRadar is AI-assisted, but a named human editor signs off every published piece and owns any correction. These are placeholder roles pending the final named team.
Owns the trust model, the forced-read backstop, and the masthead. Final human gate before anything publishes.
Signs off the compliance tracker — the moat. Every effluent limit, MCL and compliance date is a forced human read here.
Owns process-engineering plausibility — the person who double-checks a figure like “>99.9% mineralisation” before print.
Maintains the public corrections log. When WaterRadar is wrong, the fix is visible and attributed — not quietly edited.
We disclose the model openly: WaterRadar uses an agent team to research, draft and adversarially review, but the byline, the sign-off and the accountability are always human.
WaterRadar keeps a standing corrections log. Every material correction is dated, attributed and held on the record rather than quietly edited away — the same principle as the confidence badges: nothing important stays hidden.
Spotted an error? Write to the Corrections Editor and we will check it against the primary source. Read the corrections policy & log →