An OCR pin is a choice of stack
A house note says to pin a pure-Rust OCR crate so an agent does not default to Tesseract or PaddleOCR. The pin is a dependency choice. It is not a measured accuracy win.
The short version
No. The October 24, 2026 house note argues for pinning a pure-Rust OCR crate so a coding agent does not default to Tesseract or to Python PaddleOCR. Tesseract and PaddleOCR are real projects with their own licenses and their own results. The note does not publish a shared exam that ranks them.
The decision the note is about
The October 24 house note is about the dependency a coding agent adds when a task needs text read out of an image. Its argument is to pin a pure-Rust OCR crate, Carmenta, so the agent does not default to Tesseract or to Python PaddleOCR. That is a stack decision: which code is allowed in the build. It is not a claim that the same page was read three ways and scored. On-device inference still keeps whatever the device actually ran. The document named in the sources is the record this page will keep.The device keeps the run.
Tesseract is a project, not a slur
The Tesseract repository is a maintained OCR engine with its own license and its own release history. Naming it as a default an agent might grab is a statement about agent habit. It is not a finding that the engine fails. A reader who needs Tesseract's accuracy needs Tesseract's own tests. Asked directly, What does on-device inference still require? A model that runs on the device, and a test of the text it returns. A crate pin does not supply that test. The document named in the sources is the record this page will keep.
PaddleOCR is a Python project
PaddleOCR's repository is the second default the note names. Pulling it pulls that project's language and its dependencies. A pin that refuses it is a boundary on the agent. The boundary is not a score. Asked directly, Are Tesseract and PaddleOCR fictional defaults? No. Both are public projects. Tesseract is the long-running OCR engine. PaddleOCR is a Python project. An agent that imports either is importing that project. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.
What this page will not copy
The note's install lines, skill-folder layout, and prompt text are the note's. This brief stops at the question the note answers: which OCR stack enters the repository. Weights a reader can fetch are a different artifact, and a fetch is not an accuracy trial. Asked directly, Is a house note an independent benchmark? No. It is one source. The other two sources are the projects the note wants an agent to stop defaulting to. The document named in the sources is the record this page will keep.A download is not a trial.
One house link is the source
The learn article is a source on this page. It is not a second link in the body, and it is not a benchmark bureau. A context window's bound is a third kind of limit and does not rank OCR engines. Asked directly, Does this page repeat the note's commands? No. The pin is the decision. The commands that carry it out stay in the note. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.A window does not rank OCR.
Who the pin is for
A developer who lets an agent edit a build file is the person the note addresses. A reader who already chose Tesseract on purpose is outside the default the note is worried about. Asked directly, What is the note asking a reader to decide? Which OCR dependency a coding agent is allowed to add. The note's answer is a pure-Rust crate, named so the agent does not reach for Tesseract or PaddleOCR first. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.
Accuracy still needs a page and a score
Two OCR engines can both be maintained and still disagree on a scan. The way to know is a shared set of pages and a published error count. Neither the house note nor the two project homepages, as cited here, are that exam. A paper's own timing does not become an OCR score either. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.A clock is not an OCR score.
A pin can be reversed
Pinning a crate records a choice in the build. A later commit can choose a different engine. The pin is evidence of intent at the moment it is written, not a permanent ranking of the field. A vendor advisory is a different kind of record and does not score an OCR engine. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep. The document named in the sources is the record this page will keep.An advisory is a vendor record.
What would make an accuracy claim
A published set of pages, the engine versions, and the error rates. Until that exists, the honest sentence is that the note chooses a stack. The short answer this page will stand on is this: No. The October 24, 2026 house note argues for pinning a pure-Rust OCR crate so a coding agent does not default to Tesseract or to Python PaddleOCR. Tesseract and PaddleOCR are real projects with their own licenses and their own results. The note does not publish a shared exam that ranks them.
Sources
The reports this brief is filing. Futureweb did not republish them.
Questions
What is the note asking a reader to decide?
Which OCR dependency a coding agent is allowed to add. The note's answer is a pure-Rust crate, named so the agent does not reach for Tesseract or PaddleOCR first.
Are Tesseract and PaddleOCR fictional defaults?
No. Both are public projects. Tesseract is the long-running OCR engine. PaddleOCR is a Python project. An agent that imports either is importing that project.
Does this page repeat the note's commands?
No. The pin is the decision. The commands that carry it out stay in the note.
Is a house note an independent benchmark?
No. It is one source. The other two sources are the projects the note wants an agent to stop defaulting to.
What does on-device inference still require?
A model that runs on the device, and a test of the text it returns. A crate pin does not supply that test.