Routing · measured live

See which engine each field of your documents actually needs.

Three of the five fields need no model at all. A text pattern reads them, free.The name and the address go to a model, each to the smallest one that holds up.$1.91 per 1,000 documents, not $8.00.

Try it free on 100 of your pages
1,000documents routed
$1.91spent, assumed prices
94.4%mean accuracy per field
The measured relief, state 01: 94.4% per field, 76.7% per file published routing: 94.4% accuracy per fieldper-file rate: 76.7%, 92 of 120 fileshuman tier: 85% assumed, not sampled
finding 0117.7 points apart

The same published routing reads 94.4% accuracy averaged per field and 76.7% as the per-file rate, 92 of 120 files, 17.7 points apart.

94.4%mean accuracy over the five fields
76.7%files with all five fields correct (92 of 120)
The measured relief, state 02: File-aimed routing: 3 files gained, none lost, of 120 files file-aimed routing: cheaper only on an assumed pricepublished routing: the pick it replacesa tier both routings share
finding 02on an assumed price

Aiming at the file gains 3 files and loses none of the 120 files: too few to separate the two rates. The cost falls 3.5× only if the large model is billed at an assumed price per call.

$191Published routing, per 100,000 documents, with the large model at an assumed $1.60 per 1,000 calls.
$54File-aimed routing, same volume and assumption. Priced at machine time, it is the dearer of the two.
The measured relief, state 03: Abstention: 85 wrong values removed, 12 correct values withheld an emptied cell: a blank held for re-readingabstention: 85 wrong values removed, 12 correct withheld
finding 03after abstention

On 30 documents chosen for being hard, abstaining removes 85 values that were wrong and withholds 12 values that were right. Of the values still returned, 62.3% are right, against 30% before.

30%Accuracy when each value is returned, right or wrong, on the 30 documents of the hard corpus.
62.3%Accuracy of the values still returned after abstention; 97 of 150 values go to review.
The measured relief, state 04: Identical counts across two passes, and unstable durations withheld two runs, identical countsdurations vary 16 to 60%, so withheld
finding 04withheld

Two passes produce identical counts. Durations vary, so they’re withheld.

identicalToken counts match exactly across both runs.
16–60%Durations moved between 16% and 60% from one run to the next, so they are withheld; a cost built on a duration carries that spread.
The measured relief, state 05: All 16,807 routings computed, none sampled all 16,807 routings tested, none sampledhuman tier: 85% assumed, not sampled
finding 05the full span

The solver computes all 16,807 routings and prints the winner; the record carries the count. None is sampled.

16,807All 16,807 routings computed, end to end.
120 filesThe 120 records held out for scoring them, frozen with a content hash.

The extraction cost audit

Find the cheapest extractor for each field, measured on your own pages.

  1. You labela sample of your own documents, field by field.
  2. You runthe audit on your machine: your extractor, the ones you compare, our local models.
  3. You receivea sealed record and a signed report: the cheapest source per field, the saving at your volume. Check the signature with node src/verifier-rapport.mjs.
First page of the sample report: 100 real receipts, 8 sources on 3 fields, all routed to gemini-flash, saving $95,710 a year at 1,000,000 documents.
100 real receiptsrecord ac7d0adbe4907caf
signed page beside the PDF
Download the sample ↓

Start with receipts

100 real receipts,
graded in the open.

The public run grades two vendors and our local tiers on 100 real receipts. To grade your own, write to us: the free test below.

96.8%gemini-flashvendor
93.7%google-expensevendor
81.1%gen-4bours, local, needs Ollama
62.1%largeours, large encoder
2.1%smallours, small encoder

Share of totals read right. The local tiers read a text that our macOS OCR produced from the images, so their rates include that OCR's errors.

Snapshot

$490up to 1,000 pages

One document type, measured once and sealed, back within 48 hours.

  • Up to 10 fields and 4 extractors
  • The routing for each field, and the saving at your volume
  • What the sample is too small to decide
  • Paid in two halves, on signature and on delivery

Audit

$1,900up to 10,000 pages

Up to 3 document types, sealed, back within 72 hours.

  • Up to 30 fields and 10 extractors
  • A sample sized to separate sources a few points apart, when they are
  • The signed report, its PDF, and the sealed record
  • Paid in two halves, on signature and on delivery

Quarterly audit

$4,900a year

The audit measured again 4 times a year, as vendors change their models and prices.

  • 4 sealed reports a year
  • Each one says what moved since the last
  • Paid a year in advance; if you stop, it runs to the end of the year paid, no refund
  • Offline. The measurement runs on your machine.
  • What reaches us. One sealed record: counts, a right, wrong or blank verdict per case and field, your file's name and its hash, and the prices and volume you declared. No document, no value read from one.
  • Cloud extractors. Testing one sends pages to that vendor under your own account, as it does today; the local tiers keep each page on your machine.
  • Vendor fees. For the pages you run through cloud extractors, billed to you by those vendors, on your own keys.

How the free test goes

Five steps, all on your machine, once we have sent you the tool.

Format
A CSV: an id, the text, then one column per field, named with its type: id,text,total:amount,date:date,vendor:free-text
No type
Exact text: "$1,234.50" and "1234.50" differ. Such a column is refused unless you pass --exact.
Blanks
A dash where a document has no such line. Empty when you do not know the value.
Text
What your vendor already returns 1, or your images read offline: once 2 (5.2 MB of language files), then 3.
Output
Either one writes your-with-text.csv, used in steps 2 and 3.
Needs
Node 24 or newer. Up to 3 extractors.
1$ npm run text-from-exports -- --cases=your.csv --vendor=<textract|documentai|azure> --exports=<folder>
2$ npm run tessdata -- --prime
3$ npm run text-from-images -- --cases=your.csv --images=<folder> --ocr=tesseract --lang=eng
No vendor call
You run each extractor you use or want to compare on your pages, then grade its outputs here.
From values
A JSON { "<id>": { "<field>": "<value>" } } 1
From exports
Raw Textract, Document AI or Azure: --vendor=<textract|documentai|azure> --exports=<folder> --mapping=mapping.json in place of --values.
Output
Verdicts only, no value.
1$ npm run grade -- --cases=your-with-text.csv --name=<vendor> --values=<its outputs> --price-per-thousand-pages=<your price> --out=<vendor>.json
Margin
The one you accept, 1.
Files
One --sorties per file, or a comma list.
Vendors only
Add --no-encoders: nothing downloaded. Without it, our local models download once.
1$ npm run measure:yours -- --cases=your-with-text.csv --sorties=<vendor>.json,<vendor>.json --current=<the one you run today> --margin=2 --pages-per-year=<your volume>
Send
The file ending in -measured.json beside your CSV, to contact@crusetra.com.
It holds
Counts, a right, wrong or blank verdict per case and field, your file's name and its hash, the prices you declared.
Never
No document, no value. It is the one file you may attach.
Contract price
A price under a vendor contract can be replaced with a list price before you send it.
Back
A one-page PDF by email within 48 hours, from that record. The record carries a content hash.
Free test
This report is not signed: the free test stays an internal evaluation under the thirty-day grant.
Paid tiers
The Snapshot and the Audit come back signed, and include the right to act on the recommendation in your own operations.
Sample size
On our 100-receipt sample, two vendors 3.2 points apart on totals could not be told apart (7 disagreements, p = 0.45). 100 pages separate vendors far apart; the paid tiers size the sample for close ones.

DownloadThe sample reportVerifyThe signed report

Try the Routing instrument on our public test set.

Crusetra · RoutingOpen the live instrument See every field, tier, accuracy and cost, live from our public test set. Set the budget and watch the tool choose.
See Pricing
  1. Measured on. 1,000 held-out records for the rules, small and large tiers; 120 for the generative tiers. A ring marks the published routing of each field.
  2. Corpus. Records we wrote, about 166 characters each.
  3. Caveat. The generative tiers' prompt was tuned on the held-out half: their figures are optimistic, by an amount not yet measured.
  4. Human. Accuracy assumed at 85 % until you measure your own reviewers.

Crusetra, proven in 90 seconds.

The five Routing findings