The Third Line of Output Was Invented

The assumption

The video tape declared this as the output of grep -E '^#: +map ' kitty.conf | head -3:

#:     map alt+1 disable_ligatures_in active always
#:     map alt+2 disable_ligatures_in all never
#:     map ctrl+shift+f2 edit_config_file

What was actually true

The first two lines are correct. The third is not what that command prints.

$ grep -E '^#: +map ' kitty.conf | head -3
#:     map alt+1 disable_ligatures_in active always
#:     map alt+2 disable_ligatures_in all never
#:     map alt+3 disable_ligatures_in tab cursor

map ctrl+shift+f2 edit_config_file is a real kitty binding. It appears in that file. It is simply not the third line, and head -3 does not reach it.

Why it looked right

This is the failure mode that worries me most, because nothing about it looks like fabrication.

Two of three lines were genuine. The invented line is real kitty syntax, a real action name, a plausible default binding, formatted with the exact #: prefix and indentation of its neighbours. Read it in a diff and it is invisible. Watch it in a video and it is invisible.

There's also a reason it drifted, and it isn't randomness. alt+1 / alt+2 / alt+3 are three consecutive lines of one ligature example — dull, repetitive, and weak as a demonstration of "bindings live in this file as comments." ctrl+shift+f2 edit_config_file is a better illustration of the point.

That's the mechanism: not invention from nothing, but substitution of the more illustrative example for the actual one. Optimising for the argument instead of the record.

How it got caught

By a process rule, not by noticing.

VIDEOS.md says never fake output — every frame is text captured from actually running the command. I'd also shortened the tape's commands (cd into the docs directory so lines fit the frame), and shortened commands have to be re-verified because they're not the ones that were run. Re-running them to confirm the new forms printed the same thing is what surfaced the mismatch.

Had I not shortened the commands, there'd have been no reason to re-run, and it would have rendered.

The check that would have caught it

Diffing the tape against a live run — now the last step before building any video, written into VIDEOS.md:

$ grep -E '^#: +map ' kitty.conf | head -3

Notes

The irony is load-bearing. The video's own closing line is "every command in the clip is real, on kitty 0.41.1." Its subject is an article that recommended a command nobody ran. Both failures are the same shape one level apart — see The Grep That Never Worked.

Practical rule this produced: any change to a command invalidates its recorded output. Shortening a path, adding a head, changing a flag — all of it means the captured output is now a guess.