80 planned · 80 attempted · 80 valid completions · 0 incomplete · 0 unplayed.
Four players, ten points, balanced seat rotation. Earlier smoke and failed runs
are kept in a separate audit trail.
The candidate did not earn promotion
95% interval
Liquidity · candidate25.0% · n=80
ETA · control32.5% · n=80
Fast · opponent A16.3% · n=80
Fast · opponent B26.3% · n=80
0%50%100%
Run de39917c-55da-40ad-a53a-5b8588898810. 80 completed of 80 planned games. 95% Wilson intervals; descriptive evidence only. Candidate 20 wins; ETA control 26. The two Fast rows are distinct competitor slots, not independent duplicate samples. Whiskers are per-slot uncertainty, not the uncertainty of their difference.
One change: count what surplus can buy
ETA estimates how long direct production takes to fund a build. Liquidity also
counts whole bank and owned-port exchange bundles, after reserving the target's
cards. The other strategic rules stay shared.
For example, twelve surplus lumber can become three ore at 4:1, if the bank can
supply them. Waiting for ore production is not the only route to a city.
Official trade rules.
01Count surplus
02Price conversion
03Bargain or bank
A port exchange sets the minimum value of a player trade.
This study tests the conversion part of that idea. It does not test port acquisition,
reputation, deception, or expectimax.
A different spending pattern did not produce more wins
Liquidity averaged 6.34 exchanges per game; ETA averaged 5.86.
It also discarded 10.16 cards versus 8.71. These descriptive differences
suggest a diagnosis to investigate, not a demonstrated cause of the result.
Did exchange planning change bank use?
Measured evidence
Scroll the chart horizontally to inspect all values.
LiquidityETA control
Fraction of 80 games with at most the indicated number of bank/port exchanges. Higher curves mean fewer exchanges. Values are exact empirical proportions at integer thresholds; connecting segments are guides. More trading alone is not evidence of stronger play.
Source: Run de39917c-55da-40ad-a53a-5b8588898810; verified inventory SHA-256 063a024a2f341520bdf1188904b7ddc66a23836087fb7a14abef06c976ab6046. analysis/liquidity.py SHA-256 f22fc5d15a3c27cd6e0de0e4c17ff51506159974bbfca5775061f25a95d47898.
View data table
Series
Bank/port exchanges per game
Fraction of completed games
Liquidity
0
0.0125
Liquidity
1
0.05
Liquidity
2
0.1625
Liquidity
3
0.2375
Liquidity
4
0.35
Liquidity
5
0.4625
Liquidity
6
0.525
Liquidity
7
0.625
Liquidity
8
0.75
Liquidity
9
0.8
Liquidity
10
0.875
Liquidity
11
0.925
Liquidity
12
0.9625
Liquidity
13
0.975
Liquidity
14
0.975
Liquidity
15
0.975
Liquidity
16
1
ETA control
0
0.0125
ETA control
1
0.0375
ETA control
2
0.1375
ETA control
3
0.2125
ETA control
4
0.35
ETA control
5
0.475
ETA control
6
0.625
ETA control
7
0.7
ETA control
8
0.8125
ETA control
9
0.9375
ETA control
10
0.95
ETA control
11
0.9625
ETA control
12
0.9625
ETA control
13
0.9875
ETA control
14
0.9875
ETA control
15
0.9875
ETA control
16
1
Final scores tell a similarly mixed story
Liquidity finished ahead of ETA in 35 games, behind in 39, and tied in
6, using exact final points. The grid groups those scores into readable bands.
How did the two policies finish in the same game?
Measured evidence
Scroll the chart horizontally to inspect all values.
Games in each score combination
Darker cells mean higher number of games: 0–12. Blank cells have no value.
80 complete games. Each cell counts games in the two final-score bands. Above the diagonal favors Liquidity; below favors ETA. Blank cells contain zero games. Score bands are descriptive; winning remains the primary outcome.
Source: Run de39917c-55da-40ad-a53a-5b8588898810; verified inventory SHA-256 063a024a2f341520bdf1188904b7ddc66a23836087fb7a14abef06c976ab6046. analysis/liquidity.py SHA-256 f22fc5d15a3c27cd6e0de0e4c17ff51506159974bbfca5775061f25a95d47898.
View data table
Series
ETA control · final points
Liquidity · final points
number of games
Games in each score combination
2–4
5–6
3
Games in each score combination
2–4
7–8
1
Games in each score combination
2–4
10+
1
Games in each score combination
5–6
2–4
4
Games in each score combination
5–6
5–6
6
Games in each score combination
5–6
7–8
4
Games in each score combination
5–6
9
1
Games in each score combination
5–6
10+
5
Games in each score combination
7–8
5–6
4
Games in each score combination
7–8
7–8
6
Games in each score combination
7–8
9
2
Games in each score combination
7–8
10+
10
Games in each score combination
9
5–6
2
Games in each score combination
9
7–8
1
Games in each score combination
9
10+
4
Games in each score combination
10+
2–4
3
Games in each score combination
10+
5–6
8
Games in each score combination
10+
7–8
12
Games in each score combination
10+
9
3
Watch the preselected example
The first scheduled game ended on turn 68. Fast opponent B won; Liquidity
finished with 7 points and ETA with 8. The replay was selected by schedule,
not because it supports the conclusion.
liquidity slot 0’s turn
Turn 0 · Place a settlement · 10 points to win
Bank19Brick19Lumber19Wool19Grain19Ore25 development cards left
L
liquidity slot 0
Playing
0 VP
0 resource cards0 development cards0 settlements0 cities0 roads0 knights played
Longest route: 0
E
eta-control slot 1
0 VP
0 resource cards0 development cards0 settlements0 cities0 roads0 knights played
Longest route: 0
F
fast-control slot 2
0 VP
0 resource cards0 development cards0 settlements0 cities0 roads0 knights played
Longest route: 0
F
fast-control slot 3
0 VP
0 resource cards0 development cards0 settlements0 cities0 roads0 knights played
Longest route: 0
Actions & conversation
1 events · since the previous position
Table
The game started.
Exchange-aware ETA against ETA and fast policies in the first scheduled game.
41 recorded positions. Board changes between positions are grouped.
Replay sources
Game 0eac44d7-babd-4016-ab4a-9b3bbd1e8ce2 · version 14
Public events: 93c3a8edab112ac42d1baee823a95221e8c8d8ad9263eadfdd9c4f3410bb670e
Next test: record the target and alternatives at each decision, then compare
mean-income planning with a discard-aware estimate. The present data cannot tell
us which particular target changes helped or hurt.
The promotion rule required at least +10 percentage points and a one-sided exact
conditional test at p ≤ 0.05. Observed p = 0.849. An exploratory whole-game
bootstrap gives a 95% interval of −23.75 to +8.75 percentage points; it was not the
preregistered decision test. Per-policy Wilson intervals appear in the chart.
The lineup was one Liquidity, one ETA control and two Fast controls. All used the
same Rust runner and a shared limit of 16 trade-offer attempts per game; acceptance
remained available. Each received only its own participant view. No language model
made game decisions. This is a heuristic ablation on one opponent population.
The estimator uses deterministic mean production plus the baseline's 0.025
smoothing. It ignores robber blocking, production variance, discards and future
bank shortages. The server alone checks actual legality and determines outcomes.
Native decision latency is unavailable in harness measurements; zero in the
legacy counter is not zero computation. Timing is not a throughput result.
The frozen evaluation is d9cbd52e-0a90-4c4b-8fae-23c0372f1d6e and the run is
de39917c-55da-40ad-a53a-5b8588898810. Sources and binary hashes match the final
four-game smoke. Verified archives are local; no external publication or backup
is claimed. Attribution: Codex; exact model identifier unknown.