Put the interactive headline panels back under the teaser
Browse filesPreferred placement: demo, then the selection schematic, then the interactive
four-benchmark panels, all above the abstract. Results at a glance keeps the
headline number cards.
- README.md +3 -3
- index.html +15 -15
README.md
CHANGED
|
@@ -82,9 +82,9 @@ rounds them from unrounded rates.
|
|
| 82 |
**3. WebArena-Infinity evolution curve.** All 37 evaluated variants with a best-so-far envelope,
|
| 83 |
from `notes/tw_dynamics.json` → `wai_adaptive_scores`. Toggle either series.
|
| 84 |
|
| 85 |
-
**4. Headline four-benchmark panels.** Replaces the right half of the teaser, and
|
| 86 |
-
|
| 87 |
-
|
| 88 |
`scripts/gen_summary_figure.py`. The reason to make this one interactive is that the paper's figure
|
| 89 |
has to *state* its truncation in the caption: on a shared axis WebArena-Infinity's +49.5 flattens the
|
| 90 |
two terminal benchmarks into slivers, so each panel is scaled to its own range. Here the reader can
|
|
|
|
| 82 |
**3. WebArena-Infinity evolution curve.** All 37 evaluated variants with a best-so-far envelope,
|
| 83 |
from `notes/tw_dynamics.json` → `wai_adaptive_scores`. Toggle either series.
|
| 84 |
|
| 85 |
+
**4. Headline four-benchmark panels.** Replaces the right half of the teaser, and sits directly under
|
| 86 |
+
it. Clicking the schematic expands it in place to the paper's full figure, so while expanded the
|
| 87 |
+
paper's static bars and this interactive version of the same chart are both on screen. From
|
| 88 |
`scripts/gen_summary_figure.py`. The reason to make this one interactive is that the paper's figure
|
| 89 |
has to *state* its truncation in the caption: on a shared axis WebArena-Infinity's +49.5 flattens the
|
| 90 |
two terminal benchmarks into slivers, so each panel is scaled to its own range. Here the reader can
|
index.html
CHANGED
|
@@ -391,6 +391,21 @@
|
|
| 391 |
four benchmark panels alongside; click again to come back.
|
| 392 |
</figcaption>
|
| 393 |
</figure>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 394 |
</div>
|
| 395 |
</section>
|
| 396 |
|
|
@@ -505,21 +520,6 @@
|
|
| 505 |
(Terminal-Bench 2.1), held-out task split (TerminalWorld), synthetic → real intent
|
| 506 |
shift (WebArena-Infinity), and zero-shot cross-benchmark transfer (SWE-bench Verified).
|
| 507 |
</p>
|
| 508 |
-
|
| 509 |
-
<div class="block">
|
| 510 |
-
<div class="widget" id="hl-widget">
|
| 511 |
-
<div class="wtop">
|
| 512 |
-
<span class="wlabel">y-axis</span>
|
| 513 |
-
<div class="chips" id="hl-chips"></div>
|
| 514 |
-
<span style="flex:1"></span>
|
| 515 |
-
<label class="toggle"><input type="checkbox" id="hl-prior" checked /> best prior agent</label>
|
| 516 |
-
</div>
|
| 517 |
-
<div class="panels" id="hl-panels"></div>
|
| 518 |
-
<div class="wlegend" id="hl-legend"></div>
|
| 519 |
-
<p class="tnote" id="hl-note"></p>
|
| 520 |
-
</div>
|
| 521 |
-
<noscript class="nojs"><img src="assets/teaser.png" alt="Four-benchmark summary: base vs. evolved vs. best prior agent." /></noscript>
|
| 522 |
-
</div>
|
| 523 |
</div>
|
| 524 |
</section>
|
| 525 |
|
|
|
|
| 391 |
four benchmark panels alongside; click again to come back.
|
| 392 |
</figcaption>
|
| 393 |
</figure>
|
| 394 |
+
|
| 395 |
+
<div class="block">
|
| 396 |
+
<div class="widget" id="hl-widget">
|
| 397 |
+
<div class="wtop">
|
| 398 |
+
<span class="wlabel">y-axis</span>
|
| 399 |
+
<div class="chips" id="hl-chips"></div>
|
| 400 |
+
<span style="flex:1"></span>
|
| 401 |
+
<label class="toggle"><input type="checkbox" id="hl-prior" checked /> best prior agent</label>
|
| 402 |
+
</div>
|
| 403 |
+
<div class="panels" id="hl-panels"></div>
|
| 404 |
+
<div class="wlegend" id="hl-legend"></div>
|
| 405 |
+
<p class="tnote" id="hl-note"></p>
|
| 406 |
+
</div>
|
| 407 |
+
<noscript class="nojs"><img src="assets/teaser.png" alt="Four-benchmark summary: base vs. evolved vs. best prior agent." /></noscript>
|
| 408 |
+
</div>
|
| 409 |
</div>
|
| 410 |
</section>
|
| 411 |
|
|
|
|
| 520 |
(Terminal-Bench 2.1), held-out task split (TerminalWorld), synthetic → real intent
|
| 521 |
shift (WebArena-Infinity), and zero-shot cross-benchmark transfer (SWE-bench Verified).
|
| 522 |
</p>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 523 |
</div>
|
| 524 |
</section>
|
| 525 |
|