Build tempo
Which next build converts your current resources into progress fastest?
01Inspect the hand
02Find the bottleneck
03Fund the next build
A first, narrower exchange-aware ETA experiment is now complete: 80 valid games, with no established improvement over ETA. The broader study below remains a proposal.
The decision at the table
A city can have a higher long-run return while a settlement can arrive sooner and secure scarce space. A road may spend the exact cards needed for that settlement. Counting resources and dividing by average income misses these competing uses.
The mechanism to test
Estimate a distribution of times to afford each useful build from the current hand, production, bank exchanges, and plausible trades. Compare short sequences, including what becomes possible after the first build. Expected time to afford is different from expected resources divided by production: arrivals are discrete, joint, and interrupted by robber events.
Proposed experiment
Compare a fixed build priority with the existing ETA baseline, then ETA plus an explicit opportunity cost for spending bottleneck cards. Freeze opening placement and trade behavior. Hypothesis: bottleneck-aware sequencing reduces time to meaningful builds and improves win rate.
Plot build-time distributions and the share of turns waiting for each missing resource. Retain unfinished games alongside the completion denominator. Before a build occurs, its waiting time is censored, not zero. Use event-aligned trajectories to show exactly where a plan stalled.
First ablation
Switching the acquisition-time plan term off changed the contrast by +0.062 wins per game (95% interval -0.039 to +0.164): once the search plans whole turns, the card and production terms carry most of the signal the term was added for. A confirmation on fresh seeds is registered. On 192 fresh seeds the contrast was +0.039 (95% interval -0.021 to +0.099): the term is redundant at depth 2. Details and every run ID are in the ablation report; this is a development-tier result on one lineup.
What could disprove it
Faster first builds may mortgage the endgame. Test wins as the primary endpoint; turns to first city is a mechanism diagnostic. Cheap roads that lead nowhere should not count as equivalent progress.
Agent notes
Use the shared experiment design to freeze candidate versions, full lineups, budgets, sample size, primary contrast, and stopping rules before collecting evidence. This is a draft study brief, not a preregistration. This full proposal has no run; the linked narrower ablation has its own retained run.
The deployable policy reads only its own observation and recipient-visible events. The diagram is a conceptual schematic. Build new measured exhibits from retained artifacts using the visual publishing guide.
The existing eta opponent is runnable; this extension is a proposal. See protocol inputs.
All approaches · Player’s guide · Experiment program
Tracked investigations
| Updated | Investigation | Status | Finding and next step |
|---|---|---|---|
| 2026-09-09 | The acquisition-time plan term | Complete | Screen +0.062 then confirmation +0.039 (95% interval -0.021 to +0.099) on 192 fresh seeds: removing the plan term does not hurt at depth 2; the term is redundant once turns are planned. Next: Retest under the depth-3 candidate before removing the term from the leaf; keep it as a documented setting. Log 2026-09-09 |