Upload a scanned plat. Get the calls, the math, and a CAD file.
It reads the bearings and distances off a scanned survey document, then checks its own reading with closure math. It reads. You review.
Drop a scanned plat here, or
The sheet stays in the field while it is read.
Your key
no key
Every read runs on your own Anthropic account and is billed there, not here. The key stays in this browser, goes to the reader with each request, and is not kept or logged by it. Get a key and set a spend limit in your Anthropic console.
How to read it
Which model reads it
A scanned survey sheet is far bigger than the reader can take in one picture, so a quick read shrinks the whole page to fit and the small lettering suffers. On a dense drawing that costs most of the sheet: measured on a 2004 subdivision plat, a quick read captured 1 to 2 figures and 7 to 23 calls, and the reader itself reported the interior labels as at or below reliable legibility. One quarter of the same sheet, sent whole, came back with 5 figures and 80 calls. A full read cuts the sheet into overlapping pieces small enough that the reader takes each one without shrinking it, and starts each request as soon as its piece is cut rather than waiting for the one before, so the reads overlap instead of running strictly in turn. It costs you several times more, because it is several reads, each billed to your key. Those requests go out together, and a brand-new Anthropic account's rate limits may turn some of them away; a turned-away piece is not billed, and the card names the rate limit as the cause. On a light single-parcel survey the two produce much the same thing, and a sheet already small enough is never cut up. A sheet so large it would take more than 36 pieces is read whole instead, and the card says so rather than pretending it got the full treatment. Two limits worth knowing. A PDF is rendered at 200 DPI before it is cut, stepping down to 175 and then 150 only when a page is too large to cut at the finer setting, so a page scanned finer than that is still read at 200. And a full read does NOT join a figure that runs across a cut. Deciding where one run ends and the next begins is survey judgment, so it is never guessed at. Nothing the pieces read is ever discarded: every reading of every figure is kept and drawn. (The whole-sheet pass is the exception, on purpose — it supplies the title block and the notes, and its geometry is set aside because it is the shrunken reading this mode exists to replace. If a piece fails, its figures over that ground are brought back and labelled.) Where two overlapping pieces read a figure under the same label in the same place, you are told so plainly, with what each one got, so you can delete the redundant one. Where the same label turns up in pieces that do not overlap, or in more than two of them, you are told that too, because it may be one figure or several. Separately, two figures from neighbouring pieces are flagged when they print an identical bearing and distance and their corner estimates land close together in the strip those pieces share, which is the usual sign of one line read twice. None of this is exhaustive: a duplicate the reader labelled differently and transcribed differently will not be caught, so compare against the sheet.
Reading
- Page rendered
- Reading the sheet
- Closure computed
- Drawing assembled
Nothing loaded. Run a stored reading to see the whole output without a key of your own:
A traverse that closes at survey grade is strong evidence the reading is right, because a wrong digit almost anywhere would have broken it. One that does not close gets flagged with evidence, never silently corrected. Both readings below came off the scan of a recorded plat.
Stuart Cay, Plat Book 16 Page 3. An 11-call creek traverse, closing far past the survey standard.
Lost River Bend, Plat Book 16 Page 2. The recorded right-of-way description misses closure by 584.75 ft — most likely a dropped call. Flagged, not corrected.