Evaluation

Measure extraction accuracy, then ship the change

A schema change can improve one field and quietly break another. Klaaro shows you which before it goes live: change a schema, run it against records your team has already approved, and see field by field where the new extraction matches and where it does not. Your live pipeline stays unchanged until you keep the edit.

Talk to our founder

Sarah Stemmler · Founder & CEO

Try for free

100 pages · no credit card

A pull request for your schema

If you know GitHub, you know the flow: edit the schema in a draft, test it on the invoice your reviewer just approved, and promote it to main only if it holds up.

01 · Like a branch and commit
Edit in a draft
The total gets a clearer description. The edit lands in a draft with its own commit, while main stays live for the API.
02 · Like CI checks
Run the evaluation
Play runs the draft against approved records and compares every field with the value your team confirmed — the ground truth, so to speak.
03 · Like a merge
Promote to main
The score holds, so the draft goes live. From now on, every new document uses it.
Next: Integrations

Invoices/Schemamain @ c91b0e4

Invoice

supplierCompany that issued the invoice

invoice_noInvoice number as printed

issue_dateDate the invoice was issued

totalInvoice total

Config commits

draft @ e7b3d90just now · Read total as the line-item sumMatchtotal13,950.00Promote to main

91%main @ c91b0e43d ago · Initial schema

01 · Like a branch and commit
Edit in a draft
The total gets a clearer description. The edit lands in a draft with its own commit, while main stays live for the API.
Invoices/Schemamain @ c91b0e4
Invoice
supplierCompany that issued the invoice
invoice_noInvoice number as printed
issue_dateDate the invoice was issued
totalInvoice total
Config commits
draft @ e7b3d90just now · Read total as the line-item sumMatchtotal13,950.00Promote to main
91%main @ c91b0e43d ago · Initial schema
02 · Like CI checks
Run the evaluation
Play runs the draft against approved records and compares every field with the value your team confirmed — the ground truth, so to speak.
Invoices/Schemadraft @ e7b3d90
Invoice
supplierCompany that issued the invoice
invoice_noInvoice number as printed
issue_dateDate the invoice was issued
totalSum of the line items, as printed
Config commits
draft @ e7b3d90just now · Read total as the line-item sumMatchtotal13,950.00Promote to main
91%main @ c91b0e43d ago · Initial schema
03 · Like a merge
Promote to main
The score holds, so the draft goes live. From now on, every new document uses it.
Invoices/Schemadraft @ e7b3d90
Invoice
supplierCompany that issued the invoice
invoice_noInvoice number as printed
issue_dateDate the invoice was issued
totalSum of the line items, as printed
Config commits
97%draft @ e7b3d90just now · Read total as the line-item sumMatchtotal13,950.00Promote to main
91%main @ c91b0e43d ago · Initial schema
Next: Integrations

Test your schema edits without sacrificing extraction quality

Branch the schema

Edits go into a draft, a branch of the class's schema. Main keeps serving the API, so nothing you try reaches production unless you instruct it.

Scored on approved documents

A run extracts the class's approved documents again with the draft, all of them or a sample you choose, and checks every field against the value your team confirmed.

No manual re-checking

You do not re-run the whole class and proofread the output after every edit. The score shows which fields got better, and if your edits broke others.

Better? It goes live

If the draft beats main across the class, promote it and every new document uses it straight away. If not, fix it and run again.

Continue the product tour

Product capabilities

Quality & validation

See the full product overview

…Or let's talk personally!

Tell me which documents slow you down by booking your 20-minute slot in my calendar. Together, we can find out if Klaaro is the right tool for you.

Sarah Stemmler

Founder & CEO

Book your slot Try for free

Data extraction accuracy FAQ