Triangulation over volume.
Sourcing that survives scrutiny — the method behind reports a law firm can rely on.
THE WATER · 01
A search engine will hand anyone thirty links about a Japanese company in thirty seconds. None of that is evidence. The uncomfortable rule of cross-border research is that volume is the cheapest thing to fake — and the first thing a serious reader discounts.
The failure mode: volume wearing the costume of diligence
Most "research" that crosses our desk for a second opinion has the same shape: a long list of links, a summary that states one source's claim as fact, and — increasingly — machine output passed along as if retrieval were verification. It reads impressive and collapses under one question: how do you know?
For a market-sizing memo the cost of that collapse is embarrassment. For due diligence — an acquisition, a counterparty check, a person you're about to trust — the cost is the decision itself.
Nothing goes out as fact on one source's word. Every load-bearing claim is triangulated across at least three independent sources — and where three don't exist, the claim ships labeled, not laundered.
What "independent" actually means
Three copies of the same press release are one source. Independence means the sources would have to be wrong for different reasons: a corporate registry filing, a regulator's disclosure, and a trade-press record can corroborate each other; three aggregator sites that scraped the same database cannot. The discipline is boring and it is the entire game — mapping where each "fact" was born before counting it.
The reconciliation is the work
Japan adds a layer most overseas teams underestimate: the Japanese-language record and the English-language record routinely disagree. Filings, 登記 records, domestic trade coverage and a company's own Japanese site carry detail — and corrections — that never make it into English databases. Working both languages isn't a nice-to-have; the discrepancies between them are where the findings live.
"Public records stopped at shell entities." That sentence, from our own case log, is where most research quits — and where triangulation starts earning its fee.
In the engagement behind that line, a US law firm needed the real people behind a set of aircraft. The registries gave entities; the entities gave nothing. Reconciling ownership chains across registries, corporate filings and both languages resolved 29 records down to named, contactable decision-makers — every identification carrying its source chain, so the firm could rely on the findings instead of taking them on faith.
Where AI belongs — and where it doesn't
We run research AI-accelerated, in parallel, and we're not shy about it: retrieval, translation drafts, structuring, cross-checking at a breadth one analyst couldn't cover. What AI never does here is adjudicate. Deciding whether two conflicting records describe the same entity, whether a source is independent, whether a claim is load-bearing — that judgment stays human, because the liability for being wrong is human too.
What it looks like on your desk
- Findings first — the answer, then the evidence, never a link dump.
- Every claim carries its source — you can verify instead of trusting.
- Uncertainty is labeled — "confirmed", "single-source", "unknown" are different words for a reason.
- Both languages reconciled — with the disagreements surfaced, not smoothed over.
Thirty links is what research looks like. Three sources per claim is what it is.
Need something in Japan verified?
Send a short brief. First findings inside 72 hours of kickoff, every claim carrying its source.