Salience is Euclidean even when geometry is not

Eric's final layer on the torus lesson: humans see center tiles and
aim there, so the center-perceived seat pays an attention tax — and a
strong player arbitrages the gap between how the board looks and how
it routes. The bots, running true-graph BFS, have neither the bias
nor the ability to exploit it.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015RCWSTnb1KYTPyL4GmhGnF
This commit is contained in:
Eric Wagoner
2026-08-28 13:30:17 -04:00
co-authored by Claude Fable 5
parent b32db9878d
commit 5dde7b4c21
+8
View File
@@ -216,6 +216,14 @@ probing legality is free, so when unsure, try it and read the error.
threat is graph threat, and the router (BFS over edges + warps) is threat is graph threat, and the router (BFS over edges + warps) is
the only geometry that exists. Mentally re-unroll the map from the only geometry that exists. Mentally re-unroll the map from
wherever you stand — every wizard is the center of their own map. wherever you stand — every wizard is the center of their own map.
- BUT salience is Euclidean even when geometry is not (Eric): humans
see "center tiles" and aim their aggression there, so the
center-perceived seat pays an attention tax — and a strong player
turns it into arbitrage: absorb the squabble, footrace out and back
through wraps that make you secretly equidistant from everything.
Play the graph for movement, the salience for predicting humans;
expect neither bias nor exploitability from the bots — their BFS is
natively non-Euclidean.
## The board is the truth (I keep paying for this one) ## The board is the truth (I keep paying for this one)