You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* `includes/setup.html` - Agent info, metadata, and tournament log foldouts
36
+
* `includes/overview.html` - Results summary table with round navigation
37
+
* `includes/trajectory.html` - Player trajectory display with messages, diffs, and files
38
+
* `includes/rounds.html` - Main rounds section that includes trajectories and simulation logs
39
+
22
40
## Input folder
23
41
24
42
The input folders are all subfolders of `logs/`
@@ -314,18 +332,21 @@ Note that he last round only has the `round_n.log`, but there's no more trajecto
314
332
315
333
#### Trajectory view
316
334
317
-
This should be its own jinja template.
318
-
319
-
First show the following properties from the `info` section
335
+
This is implemented as its own jinja template (`includes/trajectory.html`) that is included by the rounds template.
320
336
321
-
* Number of LM calls (api calls)
322
-
* Cost (instance cost)
337
+
The trajectory template displays:
323
338
324
-
Then, show all the messages. Each message should be its own block together with the role.
325
-
Because the content can be long, make sure to have a foldout for the content.
339
+
* **Trajectory Stats**: API calls, cost, and exit status from the `info` section
340
+
* **Messages Foldout**: All trajectory messages with role badges and content preview/expand functionality
341
+
* **Diff Foldouts**: Full diff and incremental diff sections (if available)
342
+
* **Modified Files**: Changed files during the trajectory (if available)
343
+
* **Submission**: Final submission content (if available)
344
+
* **Memory**: Memory state content (if available)
326
345
327
346
**Message Role Badges**: Each message displays a small role badge (e.g., "User #1", "Assistant #2") in the top-right corner. These badges automatically hide on hover over the message block to improve readability and allow users to see text that might be obscured by the badge.
328
347
348
+
**Content Preview**: Long message content (>5 lines) shows a preview with expand/collapse functionality to keep the interface manageable.
349
+
329
350
### Important
330
351
331
352
* Always use the player name from the metadata.json file, do not assume players are called p1, p2, etc.
0 commit comments