diff --git a/docs/pdf-upload-validation-plan.md b/docs/pdf-upload-validation-plan.md
new file mode 100644
index 0000000..6b214dc
--- /dev/null
+++ b/docs/pdf-upload-validation-plan.md
@@ -0,0 +1,64 @@
+# PDF Upload Validation Plan
+
+This plan defines the validation rules Study Copilot should apply to PDF uploads before parsing them into study cards. The current backend already rejects non-`.pdf` filenames, limits PDF payloads to 10 MB, and returns clear errors when PyPDF2 cannot read a file or extracts no text.
+
+## Goals
+
+- Keep PDF uploads predictable for learners.
+- Protect backend memory, CPU, and AI generation limits.
+- Avoid leaking parser internals or uploaded document content through error messages.
+- Make future PDF handling changes easy to test.
+
+## Accepted File Rules
+
+- Accept files only when the submitted filename ends with `.pdf`, case-insensitively.
+- Require a non-empty filename so anonymous multipart fields are rejected.
+- Treat filename checks as a first-pass guard, not a full security boundary.
+- In a follow-up implementation, inspect the file signature and reject content that does not start with the PDF header `%PDF-`.
+- Keep generated study cards and API responses free of the original binary content.
+
+## Size Limits
+
+- Keep the current maximum upload size at `MAX_PDF_BYTES` (`10 * 1024 * 1024`, or 10 MB).
+- Return HTTP 413 when the raw file content exceeds that size.
+- Truncate extracted text to `MAX_NOTES_CHARS` before card generation so AI requests stay bounded.
+- Consider adding a lower configurable limit for hosted deployments with small memory budgets.
+- Document any deployment-specific limit in environment or hosting notes if it differs from the default.
+
+## Invalid PDF Handling
+
+- Return HTTP 400 when the parser cannot read the PDF.
+- Return HTTP 400 when extraction succeeds but no readable text is found.
+- Keep parser exception details out of public API responses in a future hardening pass; use stable messages such as `Could not read PDF.` and log details separately.
+- Avoid sending unreadable, encrypted, or image-only PDF content to AI generation.
+- Show frontend copy that suggests trying a text-based PDF or using the notes upload path.
+
+## Security And Performance Notes
+
+- Parse uploaded PDFs from memory only after enforcing the raw byte size limit.
+- Never store uploaded PDFs unless a future privacy-reviewed storage design is added.
+- Avoid logging full extracted text, raw bytes, filenames that may contain sensitive data, or generated prompts.
+- Add rate limiting before exposing upload-heavy endpoints publicly.
+- Consider malware scanning only if uploaded files are stored or shared with other users.
+- Revisit parser choice if PyPDF2 fails common learner documents or shows security advisories.
+
+## Backend Test Plan
+
+Add focused tests for:
+
+- Missing multipart file returns a validation error.
+- Non-PDF filename returns HTTP 400.
+- Oversized content returns HTTP 413.
+- Invalid PDF bytes return HTTP 400.
+- Empty or image-only PDFs return HTTP 400 when no text is extracted.
+- Valid text PDF calls card generation without requiring a real AI API key.
+
+These tests should force offline mode or mock card generation so they do not require `GOOGLE_API_KEY`.
+
+## Future Implementation Checklist
+
+- Add a PDF magic-header check before invoking PyPDF2.
+- Replace raw parser exception text in API responses with a stable user-safe error.
+- Add PDF upload error tests under the backend test suite.
+- Add frontend messaging for invalid, unreadable, oversized, and image-only PDFs.
+- Add deployment notes for upload size limits and privacy expectations.