CORA turns that document into a short course the worker actually finishes, built from what the clinician wrote, not from a general topic near it.
Upload one document. CORA builds the course from it.
A behaviour support plan, a mealtime management plan, a participant support plan, or any provider policy or procedure: upload the one document and CORA builds a short course from it, styled the same as the rest of the library. No topic prompt, no extra context added on top, no augmentation from anywhere else. The course is built from what's on the page.
CORA accepts PDF, Word, OpenDocument, rich text, plain text, Markdown, and a photo or scan in JPEG, PNG or WebP.
Word, OpenDocument, ordinary PDFs and plain text: read on the spot
Most documents never leave CORA's own infrastructure to be read. Word files, OpenDocument files, plain text, and a PDF that was exported from Word, Google Docs or a plan template, all of these are decoded locally, on the spot, with no model reading them at all.
A scan or a photograph gets a closer look
Only two situations reach a model: a PDF with no readable text layer, usually a scanned page, and a photograph of a printed page. Both have no text to decode locally, because a scan or a photo never had one to begin with. That's the only case where CORA reads a document by looking at it rather than decoding it directly.
That reading is transcription, not writing
There's a real difference between those two words, and it's the one this whole feature is built around. Transcription copies what's on the page. Writing decides what to say. When CORA reads a scan, it's doing the first one only: copying every word, in the order it appears, exactly as printed, never rephrasing, never correcting a typo, never expanding an abbreviation to make it clearer.
What it does when it can't read something
If a word, a number or a handwritten note on a scan genuinely can't be made out, CORA marks it [could not read] instead of guessing. If one digit in a dose is unclear, the whole value gets marked unclear rather than printed as a best guess dressed up as certainty.
Doses and medication names, food texture and fluid levels, times, dates, frequencies, measurements and phone numbers all get this treatment by name, because these are exactly the details a support worker will act on for a real person. A gap you can see and fill in yourself is safe. A confident wrong number, printed as though it came straight off the page, is not.
You check it before a single screen gets built
Before CORA turns any of this into a course, you see the full text it read, in full, and you can correct it. CORA records that you reviewed and edited it before generation, as its own dated step, so there's a record of the check happening, not just an assumption that it did.
What the course can do with your document, and what it won't add
From there, CORA can reorganise the material, condense it, and turn a statement into a scenario decision or a knowledge check a worker actually engages with rather than skims. What it won't do is introduce a fact, a number, a dosage, an escalation pathway, or an instruction that isn't already in your document. Not from general knowledge, not from the Practice Standards, not from a model's own sense of good practice. If a document is too thin to make a proper course from, CORA says so rather than padding it out.
Clinical instructions, copied character for character
A texture level, a positioning instruction, a supervision requirement, a medication step, an escalation pathway: these are quoted word for word from your document, never paraphrased into different wording that happens to mean roughly the same thing.
Every generated screen is checked back against the source document before it's shown to you. What it names as the passage it came from has to actually be there. Every instruction it marks as a direct quote is checked to appear, word for word, in both your document and in what the worker will read on screen. A number the worker will see has to appear in your source document too. None of that is a promise taken on trust. It's a check that runs against your own file every time.
The plan stays the plan
The clinician wrote the plan. The provider owns it and approved it. What CORA changes is the delivery: turning a document sitting in a shared drive into three short screens a worker actually completes, with a real completion record behind it, instead of a signature that only ever proved someone opened a file.
Where this lands as evidence
A course built from your own document feeds the same training register and completion summary as every other course in the library. It's also the direct answer to a specific gap: Emergency and Disaster Management asks that workers be trained in the provider's own plans, not a general script about emergencies, and a course built from your own disruption plan or evacuation procedure is exactly that.
Every claim on this page is written the same way every claim on the site is written: about a named artefact, checkable in a demo. See the standard CORA holds every claim to.
Common questions
What kinds of documents can CORA turn into a course?
A behaviour support plan, a mealtime management plan, a participant support plan, or any provider policy or procedure. Accepted formats are PDF, Word, OpenDocument, rich text, plain text, Markdown, and a photo or scan in JPEG, PNG or WebP.
Does a photo or scan get treated differently from a Word document?
Yes. A Word file, an OpenDocument file, plain text, or a PDF exported from a word processor is decoded locally with no model involved. Only a scan or a photograph, where there's no text layer to decode, is read by a model, and that reading is transcription: copying what's on the page rather than writing anything new.
What happens if CORA can't read part of a scanned document?
It marks that part [could not read] rather than guessing. Doses, food textures, fluid levels, times, dates and frequencies get this treatment specifically, because these are the details a support worker will act on for a real person.
Do I have to check the course before it goes out to workers?
Yes. You see the full extracted text before anything is generated from it, and you can edit it. Nothing publishes automatically, and CORA records that you reviewed it.