Michael Limberger

Evals

Model tests

HTML · Bash · Perl scripts

Three kinds of work I actually do. HTML (a resume page), Bash (safe shell), and Perl scripts (sysop collectors). Local and cloud stay in separate leagues. Full script write-up lives on the sysop page.

HTML · resume page

Full HTML resume section →

Read fixture resume markdown. Write a clean resume.html with inline CSS and real facts, no CDN. Sample outputs: kat, ornith, flash.

Local

Model	Time	Pass	Vibe
ornith	81.3s	100%	A-
qwen36	86.8s	100%	B-
kat	116.3s	100%	A
ministral	118.3s	100%	B
north	197.7s	100%	B
Devstral	176.2s	100%	B
bonsai	311.2s	100%	B+

Cloud

Model	Time	Pass	Vibe
glm	14.7s	100%	B-
flash	15.2s	100%	B+

Bash · safe shell

Full Bash section →

Run uname -s, write the result to host.txt, stay inside the workspace.

Local

Model	Time	Pass
ornith	9.7s	100%
qwen36	10.1s	100%
kat	9.1s	100%
ministral	16.2s	100%
north	11.1s	100%
Devstral	25.3s	100%
bonsai	36.1s	100%

Cloud

Model	Time	Pass
glm	2.0s	100%
flash	2.0s	100%

Perl scripts · sysop collectors

Thin prompt: non-destructive Perl sysop info collector. Detect macOS versus Linux. No sudo, deletes, or kills. Overall grades below; complexity, coverage, polish, and craft are on the sysop page.

Local

Model	Harness	Wall	Overall
ornith	100	176s	A-
kat	100	332s	A-
qwen36	100	80s	B+
devstral	100	527s	B
north	100	554s	B-
ministral	100	57s	C+
bonsai	0	515s	C

Cloud

Model	Harness	Wall	Overall
flash	100	231s	A
glm	100	35s	A-

Full sysop scoreboards, judgments, Bonsai orphan, and the .pl files →

Quick take

Winners

Setup	HTML vibe	Scripts overall	Fastest bash
Local MLX	kat (ornith A-)	ornith (kat close)	kat 9.1s
Ollama Cloud	flash	flash	about 2.0s each

Grades (HTML round harness)

Setup	Harness grade	Vibe grade
Local MLX	A → ornith / ministral	A → kat · A- → ornith
Ollama Cloud	A → glm / flash	B+ → flash