Add 6 public-baseline forecasting bots#293
Open
CodexVeritas wants to merge 3 commits into
Open
Conversation
Adds agentic PydanticAI bots that estimate what a sampled group of people would forecast (public sentiment, expert opinion, credible news outlets, left, center, and right), as public baselines to compare against the Community Prediction. Each bot is wrapped in the ForecastBot interface, searches for evidence of its group's views, and reports the sampled sources and their implied forecasts in the final comment. Wires the bots into run_bots.py, bot_lists.py, and the AIB tournament workflow, and adds pydantic-ai-slim as a dependency plus unit tests. Co-authored-by: Cursor <cursoragent@cursor.com>
…ratchpad - Add ExaQuoteSearcher tool returning article summaries + verbatim highlight quotes (with AskNews fallback) so each bot can cite the sources it sampled. - Extend ExaSearcher/ExaSource with optional summary support. - Forecast each question with 3 independent model branches (Claude Sonnet 4.5, Grok 4.20, GLM 5.1 via OpenRouter), round-robined per prediction and aggregated by the framework; add a reasoning scratchpad per branch. - Surface per-model source breakdowns and implied forecasts in the comment. - Wire branch LLMs through run_bots.py and pass EXA_API_KEY to the 6 baseline workflow jobs. Co-authored-by: Cursor <cursoragent@cursor.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds 6 cheap, agentic public-baseline bots so the Metaculus Community Prediction can be benchmarked against "what various groups of people would forecast". Each bot estimates the forecast a randomized, representative sample of a target group would give — it is explicitly a group-belief proxy, not a best-guess of the true outcome.
The 6 bots (in
forecasting_tools/forecast_bots/public_baselines/):How they work
ForecastBot(SummerTemplateBot2026) interface so they run identically to the other open-source bots. Research is skipped (run_researchreturns""); the agent does its own evidence gathering.ExaQuoteSearchertool gives each agent article summaries + verbatim highlight quotes (with an AskNews fallback) so it can find and cite what the target group actually says.ExaSearcher/ExaSourceextended with optional summary support.PopulationBaselineBotsubclass only defines aPopulationSpec(who is sampled + how to find/interpret their views). All agentic machinery, prediction conversion, and comment formatting lives in the base class.Wiring
run_bots.py: registers the 6 bots and passes the 3 branch LLMs..github/workflows/run-bot-aib-tournament.yaml: 6 baseline jobs, each now also receivingEXA_API_KEY.Test plan
pytest code_tests/unit_tests/test_forecast_bots/test_population_baseline_bot.py(6 passed) — covers binary/MC/numeric conversion + clamping, option normalization, round-robin branch-model dispatch, and Exa quote formatting.anthropic/claude-sonnet-4.5,x-ai/grok-4.20,z-ai/glm-5.1).