forked from letta-ai/letta-evals
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathrubric.txt
More file actions
34 lines (26 loc) · 1.62 KB
/
Copy pathrubric.txt
File metadata and controls
34 lines (26 loc) · 1.62 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
Here is an example of a high-quality ASCII art for this task:
```
{reference_ascii}
```
Use this as a reference for quality, detail level, and craftsmanship when evaluating the submission. Below are the coring dimensions (sum → 1.0):
1. Instruction compliance (0.30)
0.30: Art wrapped in backticks as requested (```art```); no extra content outside backticks
0.15: Art present but formatting issues (missing/extra backticks, minor extra text)
0.00: No backticks, extra commentary, multiple tool calls, or mixed content
2. Requested item match (0.30)
0.30: Depicts the requested animal (or item) with unmistakable, item-specific features (e.g., trunk for elephant, long neck for giraffe, beak + wings for bird)
0.15: Vaguely animal-like or generic creature; could be the item but weakly indicated
0.00: Wrong subject or unrelated shapes
3. Recognizability and silhouette (0.20)
0.20: Clear silhouette; core anatomy reads at a glance (in monospaced font)
0.10: Recognizable with effort; proportions or anatomy are a bit off
0.00: Unrecognizable
4. Craftsmanship and layout (0.10)
0.10: Good alignment and spacing; consistent line lengths; no tabs or trailing escape codes; holds shape in monospaced render
0.05: Minor alignment issues; still coherent
0.00: Misaligned or broken on monospaced display
5. Effort / “best version” signal (0.10)
0.10: Shows detail, multi-line effort, and clear attempt at quality
0.05: Minimal or simplistic attempt, but functional
0.00: Trivial or careless (e.g., one character)
IMPORTANT: Make sure to clearly express your rationale and scoring.