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
Add Elo distribution charts and stats to ladder arena pages
- Simplify each CC:Ladder intro to a blockquote linking the announcement
post (drop the repeated "X has a CC:Ladder — ..." sentence).
- Add a summary-stat tile row (bots, Elo range, median, spread) and an
arena-specific sentence describing the distribution.
- Add a theme-aware inline SVG histogram of the Elo spread per arena
(accent-colored rounded bars, recessive grid, dashed median marker,
hover tooltips), driven off the same ranking data.
- CSS for .ladder-stats / .ladder-figure in layout.css.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
BattleSnake has a **CC:Ladder** — a ranked ladder of open-source human snakes that turns the arena into a hill-climbable evaluation. A model starts against the weakest human snake and must win a majority of `n` rounds (and the final round) to advance to a stronger opponent; its score is the highest-ranked human it defeats. See [Introducing CC:Ladder](/insights/20260115_human_ai_ladder/) for the full format and motivation.
27
+
>See [Introducing CC:Ladder](/insights/20260115_human_ai_ladder/) for the full format and motivation.
28
28
29
29
The ladder is built from **50 open-source human BattleSnakes** (hosted as `human/*` branches on [CodeClash-ai/BattleSnake](https://github.com/CodeClash-ai/BattleSnake)). To rank them, we run a round-robin over all `50 * 49 / 2 = 1,225` unique pairs at 250 simulations each, then fit a Bradley-Terry model to the pairwise win matrix via maximum likelihood with L2 regularization (strength 0.01, base Elo 1200, slope 400).
The 50 snakes span **379–1,883 Elo**— a 1,504-point spread — but the climb is gradual: the top snake (`robosnake`) leads the runner-up by just 68 Elo, and half the field sits between 968 and 1,433. With no runaway favorite, every rung is a meaningful step up.
34
+
35
+
<figureclass="ladder-figure">
36
+
<svgviewBox="0 0 680 250"role="img"aria-label="Histogram of snakes Elo ratings"preserveAspectRatio="xMidYMid meet">
Core War has a **CC:Ladder** — a ranked ladder of open-source human warriors that turns the arena into a hill-climbable evaluation. A model starts against the weakest human warrior and must win a majority of `n` rounds (and the final round) to advance to a stronger opponent; its score is the highest-ranked human it defeats. See [Introducing CC:Ladder](/insights/20260115_human_ai_ladder/) for the full format and motivation.
28
+
>See [Introducing CC:Ladder](/insights/20260115_human_ai_ladder/) for the full format and motivation.
29
29
30
30
The ladder is built from **264 open-source human warriors**, collected by crawling the Core War online [directory](http://www.koth.org/planar/by-name/complete.htm) and hosted as `human/*` branches on [CodeClash-ai/CoreWar](https://github.com/CodeClash-ai/CoreWar/branches). To rank them, we run a round-robin over all unique pairs at 4,000 simulations each, then fit a Bradley-Terry model to the pairwise win matrix via maximum likelihood with L2 regularization (strength 0.01, base Elo 1200, slope 400).
Core War's 264 warriors form the tightest field of any ladder: the strongest warrior edges the runner-up by under **7 Elo**, and 91% of the field falls within one standard deviation (±168) of the 1200 mean. Beneath that dense competitive pack sits a long tail of broken or trivial warriors trailing down to -889 — so the early rungs fall quickly before the real wall near the top.
35
+
36
+
<figureclass="ladder-figure">
37
+
<svgviewBox="0 0 680 250"role="img"aria-label="Histogram of warriors Elo ratings"preserveAspectRatio="xMidYMid meet">
<figcaption>Distribution of Elo across the 264 human Core War warriors. The competitive pack is packed tightly near 1200, with a long lower tail of broken warriors.</figcaption>
Copy file name to clipboardExpand all lines: pages/arenas/robotrumble.md
+44-1Lines changed: 44 additions & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -24,10 +24,53 @@ Success requires crafting logic that positions robots smartly, times attacks eff
24
24
25
25
<h2id="cc-ladder">🪜 CC:Ladder</h2>
26
26
27
-
RobotRumble has a **CC:Ladder** — a ranked ladder of open-source human robots that turns the arena into a hill-climbable evaluation. A model starts against the weakest human robot and must win a majority of `n` rounds (and the final round) to advance to a stronger opponent; its score is the highest-ranked human it defeats. See [Introducing CC:Ladder](/insights/20260115_human_ai_ladder/) for the full format and motivation.
27
+
>See [Introducing CC:Ladder](/insights/20260115_human_ai_ladder/) for the full format and motivation.
28
28
29
29
The ladder is built from **58 open-source human robots**, collected from the public [leaderboard](https://robotrumble.org/boards/2/robots) and hosted as `human/*` branches on [CodeClash-ai/RobotRumble](https://github.com/CodeClash-ai/RobotRumble/branches). To rank them, we run a round-robin over all unique pairs at 250 simulations each, then fit a Bradley-Terry model to the pairwise win matrix via maximum likelihood with L2 regularization (strength 0.01, base Elo 1200, slope 400).
RobotRumble has by far the widest spread — nearly **5,000 Elo** end to end — and one dominant outlier: `gigachad` sits **592 Elo** clear of every other robot. The rest split into a strong pack above ~1,500 and a long weak tail reaching -1,737, making the final rung an enormous leap.
34
+
35
+
<figureclass="ladder-figure">
36
+
<svgviewBox="0 0 680 250"role="img"aria-label="Histogram of robots Elo ratings"preserveAspectRatio="xMidYMid meet">
<figcaption>Distribution of Elo across the 58 human RobotRumble robots. Note the lone outlier far to the right (gigachad) and the wide tails on both ends.</figcaption>
0 commit comments