The method
How we read the record
Watermark reconstructs a deliberately thin public record — across every site, across every domain — from primary documents and live public data into reviewed, cited, structured findings. This page explains the method once, for the whole network.
The evidence grammar
Every figure wears its standing
The tags below are fixed and never recolored. They are the whole trust model in one glance — before you read a number, you know what kind of number it is.
Read from a cited record or a live source — confirmed and reproducible.
A labelled reading or derivation from cited inputs — reasoning, not a record.
An outside published specification or dataset — authoritative, but not a record about this site.
Withheld, unresolved, or not yet checked — a question, not a verdict.
The pipeline
Ingest → extract → analyze
Figures are read from the 300 DPI scanned image, never the garbled OCR text layer — then checked against the layout a human would see.
Primary documents enter as bytes: county recorder scans, OCR'd PDFs, live public-data feeds. Nothing is summarized before it's stored, and the source bytes are never altered.
A structured read is pulled from the 300 DPI page image — tables, figures, signatures, redaction bars — and logged against the exact page and coordinate it came from. The garbled OCR text layer is a hint only.
The read is reconciled, labeled with its evidence tag, linked into the entity graph and timeline, and set beside its cited source. What isn't corroborated is published anyway, marked open.
Chain of custody
The corpus is treated as litigation evidence.
The original scan or filing, hashed and stored exactly as received — never renamed or edited in place.
Every structured read is checked against its source page before it is published.
Anything pulled from a live public feed can be re-run and re-verified at any time.
Across the network
Twelve domains, one grammar
Each domain reads its own kind of document or feed, but every one publishes through the same evidence grammar and the same chain of custody.
The method layer
Seven skills, one spine
Underneath every domain sits the same research agent, working through seven discoverable skills. Each hands its output to the next with the citation attached — nothing moves forward unlabeled.
That chain rests on the evidentiary-discipline spine: ingest immutably, extract legibly, analyze transparently, and never let a modeled figure travel without its tag. It's the same spine whether the domain is hydrology, the entity graph, or grid & regulatory — read each domain's page for how the spine bends to its own kind of source.