Tier 01
Pilot
500 – 2,500 pages
One document family, one eval question, answered end to end. Built for a team that wants to see the gate report before committing to a ladder.
Fixed fee from $1,500
Synthetic document fixtures
Verifiable synthetic test data for document AI teams
Not a dataset vendor. A fixture vendor.
01
Approach
We don't sell data. We ship the recipe, the quality gate, and a measurable lift in your document-parsing eval — zero LLM calls in the pipeline.
-
a
The recipe
A declarative fixture spec: document archetypes, degradation ladders, and the label schema your parser already emits. You keep the generator. We keep it reproducible — same input, same pack, byte for byte.
-
b
The quality gate
Four machine checks run before anything leaves the building, and every file lands with a sha256 manifest you can verify yourself. A pack that fails a check is not shipped, not discounted, not "fixed in the next batch".
-
c
The lift
You buy a before-and-after number on your own parser and your own eval set — not a leaderboard placement, not a volume of pages. If the delta doesn't hold on your metric, the fixture didn't work and we say so.
02
Tiers
Three ways to buy the same thing: fixture pages, gated and hash-verified. Scope is set by page count, not by seat count.
Tier 02
Standard
5,000 · 10,000 · 25,000 pages
The recurring fixture pack your CI runs against. Fixed volumes, fixed schema, re-gated on every regeneration so a green run means something.
$0.12 / $0.07 / $0.05 per page
Tier 03
Enterprise
25,000+ pages
Continuous fixture maintenance: new archetypes as your intake changes, quarterly re-gating, and a standing eval delta you can hand to a reviewer.
From $5,000 per month
Priced in USD, quoted per page. Volumes above 25,000 pages are quoted individually.
03
Contact
Send the four things below and you get a fixture scope back, not a sales sequence.
- Page volume you need per gate
- Document families and their failure modes
- The parser and metric you already trust
- The lift you would call a win