Tonic's picture
Protostar Climate submission: v2 forecast + v3/DCPP amendment audit (29 gated candidates)
49f3de8 verified
|
Raw History Blame Contribute Delete
2.12 kB

Instructions for AI agents

This repository is a submission to the Protostar Challenge: Climate Risk Intelligence. Follow these rules in every task.

Data rule (never break it)

  • Forecasts are made as of 2018-12-31. Weather data used as a forecast input must be dated on or before 2018-12-31.
  • Do not download, query or use weather observations dated 2019-01-01 or later, from any source (e.g., NOAA, Meteostat, Open-Meteo, ERA5, weather websites), nor 2019-2025 data driven by the weather (e.g., streamflow, air quality, energy demand).
  • Set an end date of 2018-12-31 or earlier on every weather download.
  • Never write predicted_value from memory or from a web search: every value must come from the code's model.
  • If a task seems to need weather data after 2018, stop and ask a team member.
  • Data that does not respond to the weather (e.g., census, land cover, elevation) may come from any year; record its source in data/README.md.

OlmoEarth-specific rules (from OLMOEARTH_PLAN.md)

  • NEVER load released OlmoEarth weights (allenai/OlmoEarth-v1*, any ModelID, load_model_from_id, rslearn model_id: OLMOEARTH_V1*), the OlmoEarth Platform, or precomputed OlmoEarth embeddings (incl. v1.3 2017-2025) as forecast inputs. All were trained on 2016-2024 data.
  • OlmoEarth architecture/code may be used only with random (seeded, logged) initialisation trained exclusively on data dated <= 2018-12-31.
  • Do not use Ai2's shipped normalisation statistics (computed over 2016-2024); recompute on the <=2018 subset.
  • Released OlmoEarth artifacts may be used only inside analysis_policy/ (analysis/policy, never forecasts).

Working rules

  • Training data: CMU-OSPO/protostar-climate-train-to-2018 (Hugging Face), pinned revision.
  • Code goes in code/, curated data in data/ (described in data/README.md).
  • After changing submission.csv, run python code/check_submission.py.
  • The code must reproduce submission.csv without an agent.
  • Never write tokens or credentials into any file.
  • Summarise what you did in the README's "AI Tools and Agents" section.