Why deeds don't close, and what closure actually proves

The check itself

A metes-and-bounds description is a walk: from the point of beginning, follow each bearing for each distance, course by course, back to the start. Recompute that walk from the recorded calls and the endpoint should land on the beginning. The distance by which it misses is the misclosure; divide it into the total perimeter and you get the closure ratio. A 400-foot loop that misses by four hundredths of a foot closes at 1:96,419, which is a real number from a real 2014 deed in our test corpus, and evidence of careful original work.

1:5,000 is a common floor for recorded descriptions: a one-foot miss over a mile of boundary. Modern instrument surveys routinely beat 1:10,000. Curve calls belong in the walk too, traversed as the arcs they are; a tool that approximates curves away is quietly rounding the very thing being measured.

Why deeds fail it

Because every step between the surveyor's field book and your screen is a chance to corrupt a number. The original survey may predate modern instruments. The drafter transposes digits in a bearing. The typist drops a course. The recorder's scan blurs a 3 into an 8, and whoever retypes the calls inherits it. A deed that fails closure is telling you one of those things happened; the ratio measures how badly, and the residual direction sometimes hints at which course to suspect. What closure never tells you is which error occurred, which is why the useful output is the misclosure, the suspect courses, and the source text together, not a red X.

What closure cannot catch

Closure is blind to any error that preserves the loop, and the important family of those is symmetric errors. Flip one quadrant letter on a near-cardinal rectangle and the figure mirrors: it closes exactly as well as the truth, and its area is identical too. We documented this while qualifying our easement corpus, where a course fabricated from a strip's own symmetry closed better than the recorded one. The arithmetic cannot see these. Two things can.

First, the deed's own stated acreage. A document that declares "containing 0.51 acres, more or less" has volunteered a second, independent measurement. Compute the area from the traverse and compare: two signals agreeing is a far stronger claim than either alone, and a boundary that closes beautifully while disagreeing with its own acreage is telling you something. This is why Easting's PASS requires both, closure within 1:5,000 and computed area within 2% of stated, and why the comparison runs in the document's own unit; a varas deed is checked in varas.

Second, a human looking at the drawn shape beside the source document. A mirrored rectangle passes every arithmetic check and fails a glance. That is why the GroundTruth certificate draws each tract to scale with its courses numbered against the call table: the reader becomes the check the math structurally cannot run.

Running the check

Easting runs all of it on the recorded document itself: reads the PDF (scans included), extracts every call with its verbatim sentence, recomputes the traverse, checks closure and stated acreage independently, and returns a PASS, REVIEW, or FAIL verdict with its reasons, the exact ratio, and the misclosure. A description that will not verify comes back as findings you can act on. Three documents are free, with no card, and failed extractions are never billed; app.easting.ai runs the check in your browser with nothing to install, the plugin installs from the QGIS plugin manager, and the guide walks the first extraction.