Pre-registration and result (17 Nov 2026): a high UCB exploration constant does not reduce premature convergence in a 14-bus generator-bidding game, under a criterion that saturates

Version 2. Version 1 held the protocol and code alone and was published before the confirmatory run returned; this version adds the result, so the deposit order is part of the record. CLAUSE: UCB with kappa = 1.0 has a nonzero premature-convergence rate at least as high as low-exploration UCB with kappa = 0.1, under identical 200-round budgets (IEEE 14-bus, DC clearing, finite bid grids, 100 seeded runs). VERDICT: FALSIFIED by the registered rule. Premature-convergence rate was 0.98 at kappa = 1.0 and 0.98 at kappa = 0.1 (kappa = 0 control: 0.99); discordant pairs 1 versus 1; one-sided exact McNemar p = 0.75. IMPORTANT LIMIT: the row's own criterion is saturated in this market. It fires on 98-99 percent of runs whatever kappa is, and the detector control passed by a single run (0.99 versus 0.98). The result therefore shows that the stated benefit cannot be shown under the stated criterion here, not that exploration is irrelevant to convergence quality. The criterion was not altered. DESCRIPTIVE ONLY, NOT TESTED: mean best-response regret falls from 132.7 (kappa = 0) to 111.1 (kappa = 0.1) to 86.2 (kappa = 1.0); the bus-2 generator is the one agent whose rate drops clearly at kappa = 1.0 (0.34 to 0.21). Both are single-run readings and may be noise. CONTROLS: clearing, exhaustive regret instrument and seed pairing passed; the detector control passed narrowly. SCOPE: canonical MATPOWER case14 run as published, which is UNCONGESTED (every branch rateA = 0), five generators, 21-point bid grid, grid-aggregated GP-UCB, 200 rounds, two kappa values. Nothing here transfers to congested networks, other learners or real markets.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24293494
Primary Topic
Electric Power System Optimization
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Pre-registration and result (17 Nov 2026): a high UCB exploration constant does not reduce premature convergence in a 14-bus generator-bidding game, under a criterion that saturates

John Goodman
Zenodo (CERN European Organization for Nuclear Research)
Electric Power System Optimization
article

Pre-registration and result (17 Nov 2026): a high UCB exploration constant does not reduce premature convergence in a 14-bus generator-bidding game, under a criterion that saturates

John Goodman
article en

Abstract

Version 2. Version 1 held the protocol and code alone and was published before the confirmatory run returned; this version adds the result, so the deposit order is part of the record. CLAUSE: UCB with kappa = 1.0 has a nonzero premature-convergence rate at least as high as low-exploration UCB with kappa = 0.1, under identical 200-round budgets (IEEE 14-bus, DC clearing, finite bid grids, 100 seeded runs). VERDICT: FALSIFIED by the registered rule. Premature-convergence rate was 0.98 at kappa = 1.0 and 0.98 at kappa = 0.1 (kappa = 0 control: 0.99); discordant pairs 1 versus 1; one-sided exact McNemar p = 0.75. IMPORTANT LIMIT: the row's own criterion is saturated in this market. It fires on 98-99 percent of runs whatever kappa is, and the detector control passed by a single run (0.99 versus 0.98). The result therefore shows that the stated benefit cannot be shown under the stated criterion here, not that exploration is irrelevant to convergence quality. The criterion was not altered. DESCRIPTIVE ONLY, NOT TESTED: mean best-response regret falls from 132.7 (kappa = 0) to 111.1 (kappa = 0.1) to 86.2 (kappa = 1.0); the bus-2 generator is the one agent whose rate drops clearly at kappa = 1.0 (0.34 to 0.21). Both are single-run readings and may be noise. CONTROLS: clearing, exhaustive regret instrument and seed pairing passed; the detector control passed narrowly. SCOPE: canonical MATPOWER case14 run as published, which is UNCONGESTED (every branch rateA = 0), five generators, 21-point bid grid, grid-aggregated GP-UCB, 200 rounds, two kappa values. Nothing here transfers to congested networks, other learners or real markets.

Zenodo (CERN European Organization for Nuclear Research)
Electric Power System Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon