Issues

• Eleven smaller review observations (M7-M11, M14, L1-L7): reproducibility, validation gaps, docs-vs-code

  cmu-delphi/exploration-tooling#230 · Filter by author docxology opened on Sep 16, 2026

• prod revision_aware grid row uses g_quantreg (quantile_reg default method), but the explore config that validated the family uses quantreg_fn (method = "fn") — explore-vs-prod divergence

  cmu-delphi/exploration-tooling#228 · Filter by author docxology opened on Sep 16, 2026

• covid explore's nhsn_archive keeps the nation as a geo_type = state row ("usa" vs the flu twin's usa → us rename)

  cmu-delphi/exploration-tooling#227 · Filter by author docxology opened on Sep 16, 2026

• nwss_hhs_region aggregation computes region_value = mean(region_value * population) / sum(population) — off by a factor of the region's state count, and NA-prone

  cmu-delphi/exploration-tooling#226 · Filter by author docxology opened on Sep 16, 2026

• data_substitutions drops unmatched correction rows without any warning

  cmu-delphi/exploration-tooling#225 · Filter by author docxology opened on Sep 16, 2026

• parse_prod_weights silently reuses the oldest ("defaults") block when the requested date has no block, defeating the downstream empty-table abort

  cmu-delphi/exploration-tooling#224 · Filter by author docxology opened on Sep 16, 2026

• climatological_model's week-window filter uses plain abs() on epiweeks 1–53 — no wraparound around the year boundary (the sibling climate_median path already uses mod_dist(..., 53))

  cmu-delphi/exploration-tooling#223 · Filter by author docxology opened on Sep 16, 2026

• run_prod_if_fresh.R completes green and creates the done-marker even when g_submission_directory silently defaulted to "cache" (no submission files written)

  cmu-delphi/exploration-tooling#222 · Filter by author docxology opened on Sep 16, 2026

• S3 targets-store sync (make pull) is followed by unverified RDS/qs/parquet deserialization — treat the stores as an untrusted input boundary

  cmu-delphi/exploration-tooling#221 · Filter by author docxology opened on Sep 16, 2026

• population/state-code/HHS crosswalk CSVs are fetched from refs/heads/main at run time on the submission path, and location typing relies on readr's implicit column guessing

  cmu-delphi/exploration-tooling#220 · Filter by author docxology opened on Sep 16, 2026

• score_forecasts rescales wk inc covid prop ed visits by ×100 but not wk inc flu prop ed visits, so flu NSSP scores compare proportion-scale forecasts against percent-scale truth

  cmu-delphi/exploration-tooling#219 · Filter by author docxology opened on Sep 16, 2026

• scripts/oracle/compare.R: NA/NaN differences are dropped by na.rm = TRUE, so NA-introducing (or NA-repairing) refactors pass the golden gate as "match"

  cmu-delphi/exploration-tooling#218 · Filter by author docxology opened on Sep 16, 2026

• build_nhsn_archive.R / build_nssp_archive.R: no locking around read-modify-write, per-minute cron, and a partial dataset silently replaces the entire previous version-date block

  cmu-delphi/exploration-tooling#217 · Filter by author docxology opened on Sep 16, 2026

• ensemble weighting silently drops unweighted forecasters and collapses zero-mass groups to 0 (no coverage/zero-mass validation anywhere)

  cmu-delphi/exploration-tooling#216 · Filter by author docxology opened on Sep 16, 2026

• in evaluation mode the submission gate still writes when the replayed date is the schedule maximum — with FORECAST_REFERENCE_DATE unset that is the current week, i.e. the live hub directory

  cmu-delphi/exploration-tooling#215 · Filter by author docxology opened on Sep 16, 2026

• five rows in the 2025-11-19 block of flu_geo_exclusions.csv are dated 2024-11-19 (year typo) — linear, linearlog, climate_base, climate_geo_agged, climate_quantile_extrapolated receive no weights for that date

  cmu-delphi/exploration-tooling#214 · Filter by author docxology opened on Sep 16, 2026
  1

• covid NHSN ensemble declares revision_aware as a component, but neither covid weights CSV contains a revision_aware row (silent zero contribution; horizon −1 rows emitted as confident zeros)

  cmu-delphi/exploration-tooling#213 · Filter by author docxology opened on Sep 16, 2026

• commit-script.sh: no set -e, stale work-backup branch resurrects last run's files, reset --hard destroys unpushed hub commits, exits 0 on failed push

  cmu-delphi/exploration-tooling#212 · Filter by author docxology opened on Sep 16, 2026

• validate_submission() results are discarded — hub-invalid files don't fail validate_result/validate_climate_result, and hubValidations` is only loaded when the COVID env var is set

  cmu-delphi/exploration-tooling#211 · Filter by author docxology opened on Sep 16, 2026

• Re-organizing and cleaning up this codebase

  cmu-delphi/exploration-tooling#209 · Filter by author dshemetov opened on Jul 27, 2026

• enh: more in-depth views

  cmu-delphi/exploration-tooling#193 · Filter by author dshemetov opened on Apr 9, 2025
  1

  

• compare us as a geo value in training set vs us as an aggregate in post

  cmu-delphi/exploration-tooling#191 · Filter by author dshemetov opened on Apr 4, 2025
  

• Consider branching targets over forecast_date

  cmu-delphi/exploration-tooling#182 · Filter by author dshemetov opened on Mar 17, 2025
  

• Consider bundling code into a package (again)

  cmu-delphi/exploration-tooling#181 · Filter by author dshemetov opened on Mar 14, 2025

• [meta] Porting code elsewhere

  cmu-delphi/exploration-tooling#176 · Filter by author dshemetov opened on Mar 5, 2025