Betomcat
Forecasting. — 2026-09-21. — 2 tracings.
A Metaculus forecasting bot, forked from the Metaculus bot template, that researches each question through the Intelligence Workbench and forecasts it with two models drawn from a pool.
Betomcat is a bot that forecasts on Metaculus. It is a fork of the Metaculus bot template, entered in the FutureEval Fall 2026 AI forecasting tournament, including its MiniBench rounds.
For each open question it classifies the question into a family of recurring questions and researches it through the Intelligence Workbench, which returns sourced, time-stamped claims. It then draws two models at random from a pool listed in a YAML file, asks each for a forecast, and submits the two combined as a weighted average, with a private comment on the question.
The bot runs unattended as a chain of GitHub Actions jobs of about five and a half hours each, with its state carried from one job to the next. It never submits a made-up default: if no model forecast exists by the final deadline before a question closes, the question is recorded as missed.
An unmodified copy of the Metaculus template bot runs alongside it as a control, to compare the two on the same questions.
Built with: Python, OpenRouter, AskNews, SQLite, GitHub Actions.
Tracings
Drawn from this record.
- F103Intelligence WorkbenchForecasting
The workbench is built to feed this bot: it gathers and grades the evidence, the bot makes the forecast.
- F101Ensemble AI forecasting botForecasting
Traced from
Records that point here.
None recorded.