On 15 November 2026 we made three small sets of practice files for readers checking a document or video handoff. Each started with our own sample text, drawing or generated footage. We compared the exported files in software. You can open the same files on the computer or tablet you are considering.
Which copy can give you the spelling back?
Our four-line proof contains Şişli, Iğdır, İzmir and Çanakkale. We embedded a DejaVu Sans text font in a one-page PDF, then rendered that page at 144 dpi and placed the resulting image on a second PDF page of the same size. Poppler pdftotext recovered all four spellings from the text PDF. Its extraction from the image PDF contained zero non-whitespace characters. We left OCR off for both runs.
Open both copies in your intended app. Find Iğdır, select it, paste it into a blank note and compare each letter with the reference. If your app extracts the name from the image copy, check the spelling from that route separately, particularly the dotless ı.
Text proof (PDF) · Image proof (PDF) · Reference spellings (UTF-8 text)
When switching languages during a meeting, the bilingual-keyboard comparison uses these spellings to check the typing and PDF handoff together.
A mark can survive after its instruction disappears
On a practice drawing we placed one rectangular PDF annotation around the rear measurement, M2. Its comment says “M2: remeasure the rear opening at skirting height.” We made a second copy using PyMuPDF’s annotation-baking operation, which places the visible mark in the page content. Both marked copies rendered identically at 144 dpi in PyMuPDF. The live copy contained one annotation with the instruction. The baked copy contained zero annotations. M2 remained searchable as page text in both.
Open each marked drawing in the viewer your colleague uses. Find M2 and see whether you can recover the follow-up instruction by selecting the red rectangle or opening the comment list. A printout can look complete while the recipient still has no way to tell what needs measuring.
Original drawing (PDF) · Drawing with live comment (PDF) · Drawing with flattened mark (PDF)
The site-drawing tablet guide uses these files to check what the next person can read after you return a marked plan.
Can a smaller edit copy take you back to the right frame?
We generated a five-second silent clip with a visible frame number on every frame and a different color for each shot. FFmpeg encoded the 960 × 540 source to H.264, then resized it to 480 × 270 for a second encode. The source was 52,681 bytes and the proxy was 20,912 bytes. We decoded both video streams and read the shot-color block frame by frame. Each stream contained 120 frames at 24 fps. The shot starts were 0, 48 and 84 in both files, counting from zero.
In the side-by-side clip, frame 48 (2.0 seconds) changes from “Establish the room” to “Check the detail.” Frame 84 (3.5 seconds) changes to “Confirm the change.” In an editor, mark those cuts in the proxy, relink to the source and step one frame backward at each mark. You should see A then B around the first cut and B then C around the second. This generated clip has no audio and little motion. Use a representative camera file separately to judge playback, sound synchronization and editing speed.
Frame-numbered source (MP4) · Half-size edit proxy (MP4)
The laptop guide to editing rough cuts on the road explains when to make proxies and how to check the relink before handing over an edit.
Reproduce the files
The output log records the exact file sizes, cut positions, extraction results and tool versions. To rebuild the samples, download the Python script, install the dependencies listed at its top and run python reproduce.py ./file-samples. A different video encoder build can produce a different byte count. The script separately checks the decoded frame positions and PDF results.