Real numbers from this tool, line by line.

7 errors and 11 warnings is what a 14-line Intel HEX image returned for an embedded firmware engineer merging a bootloader patch before the flashing station, checked against 17 rules.
| line | what the record does | finding |
|---|
| 3 | ends in checksum 0x2A | recomputed checksum is 0xE8 |
| 4 | byte count says 16 | the record carries 15 data bytes |
| 5 | # regenerated by patch_tool.py | not a record at all |
| 9 | writes 0x08000108 | line 7 already wrote that address |
| 10 | starts at 0xFFF8 | runs 16 bytes past 0xFFFF with no type 04 record |
| 11 | record type 0x07 | only types 00-05 exist |
| 14 | a data record | sits after the end-of-file record on line 13 |
The 11 warnings are the map around those errors: five blocks that do not fill whole 256-byte flash pages, four unwritten gaps (one of them 65,000 bytes), an entry point at 0x00000200 that no record writes, and a switch from upper- to lower-case hex on line 6 that says two different tools wrote this file.
Why each of those breaks a board
An Intel HEX checksum is the two's complement of every byte before it. A hand edit that changes one data byte and leaves the old checksum is the most common reason a strict programmer aborts — and a lax one flashes the record anyway. A byte count that no longer matches the data makes a loader read past the record or cut it short. The comment line on line 5 is exactly what a generator script leaves behind; many programmers stop at it.
The overlap on line 9 is the quiet one. Two records write the same absolute byte, so the value that ends up in flash depends on the load order of the tool. The record on line 14 is the classic appended fix: it sits after :00000001FF, so conforming loaders never read it and the patch never reaches the device.
Three images, three different answers
The same 17 checks give different answers to different files. A clean 19-line image — one type 04 record, 16 data records filling one 256-byte page, a start record and the EOF — returns 0 findings. A 6-line Motorola S19 image with one edited byte returns 3: the S-record checksum on line 3 (0x00 where 0x64 is correct), a block that stops 192 bytes short of a page, and no S0 header. An audit that reports nothing on a good image is the half nobody tests.
Why the archive copy matters now
Since 11 September 2026 the EU Cyber Resilience Act's reporting obligations apply to products with digital elements, and the technical documentation is kept for 10 years after the product is placed on the market or for the support period, whichever is longer. A firmware image kept as evidence is only useful if it is the image that shipped and still says which release it is. An S-record image without an S0 header names no module or version; an Intel HEX image has no header record at all.
What is free and what is not
Opening one image is free and whole: every check, every finding, every line, in VS Code or in the browser, no key. The licence covers a different job — sweeping every .hex, .s19, .srec and .mot file in a build output folder in one pass and writing a dated Markdown audit report you keep with the release. A freelance embedded software engineer averages about $103 an hour on crowdsourced rates; re-reading one merged image byte by byte takes longer than that hour.
Paste the image you are about to flash and read its lines first.