economy.md

Eleusis economy

score(seat) = knowledge(seat) + credit(seat) - experimentCost x experiments(seat)

Higher is better, and a score may be negative: a seat that runs 24 experiments and never answers a test correctly finishes at -24. The league ranks by mean episode score.

Knowledge — the rivalrous half

At each prediction test, seat j with c_j correct answers earns knowledgePool * c_j / sum(c) (default pool $20.00 over 6 strips). Nobody correct pays nobody. Because it is a share, every rival you teach takes a slice of your own pool — which is exactly why publishing is not free.

Citation credit — the cooperative half

For each test strip x and each seat j that answered x CORRECTLY, let A(x, j) be the set of seats a != j that authored a corkboard fact whose strip is at Hamming distance exactly 1 from x (it differs in exactly one of the four positions) and that was published BEFORE this test opened. If A(x, j) is non-empty, a pot of citePot (default $0.50) is split equally among its members. An author is paid at most once per (strip, confirmer) however many of its facts support x.

Citation rings do not pay. Credit accrues only for a published result that a DIFFERENT seat's passing prediction later confirms, on a strip nobody has ever tested. Self-citation is impossible by construction, a duplicate publication has no author, and no arrangement between seats can manufacture a correct prediction about a strip the machine has never been shown.

The tuned ratio

With the defaults (experimentCost 1.0, knowledgePool 20.0, citePot 0.5, testStrips 6, five seats, four tests):

• teaching one rival one extra correct answer costs you roughly 20/sum(c) - 20*c_you/sum(c)^2 — about $0.6 to $1.0 of pool share per test;
• one well-placed publication sits one token away from up to 12 strips, realistically supports one or two of the six test strips, is confirmed by two to four rivals, and pays 0.5 x confirmers / authors — about $0.5 to $2.0 per test.

So publishing pays precisely when your result is informative to others AND near the frontier. The knife edge is configuration, not code: the open-science variant raises citePot to $1.50 and the closed-shop variant drops it to $0.10, so the league can watch the behaviour flip.