There are some projects for which the dataset used is not clear #176

Activity

1. claude commented on Sep 23, 2026

   

   claudeBot

   on Sep 23, 2026 – with Claude

   Last edited by claude

   More actions

   Claude finished @sgbaird's task in 26m 22s —— View job • claude/issue-176-20260923-2102 • Create PR ➔

   ---

   Summary

   You were right to double check: in #171, Referee 4's point 3 was answered with just "Noted." and no manuscript change was made for it. This branch addresses it for real. Since the revised manuscript lives on the #171 branch, I merged copilot/address-dd-reviewer-feedback into this branch and built on top of it, so the PR below targets that branch rather than main.

   What each flagged project actually used

   I verified each one against its repository, project page, and (where repos are gone) the team's publication:

   Project	Data source
   6	No dataset; objectives computed on demand with the team's LayerLumos transfer-matrix simulator
   15	1,728-reaction direct arylation benchmark of Shields et al. (Nature 2021); the code fetches it at run time from the Doyle lab EDBO+ repo, which is why it looked buried
   19	Proposal only, never carried out; no dataset
   20	Proposal; example script fills a BayBE campaign with mock results, no experimental data
   21	MoleculeNet Lipophilicity (4,200), ZINC 250k log P benchmark (Kusner et al.), Harvard Clean Energy Project (30,000)
   29	Proposal only; intended benchmark was the Sanchez-Lengeling et al. solubility-parameter compilation
   30, 31	In-house in vitro rapid pulse voltammetry measurements (serotonin prediction error per waveform), hard-coded in the project notebook
   36	Synthetic test functions plus a protein oracle fitted to the ProteinEA fluorescence dataset (per the team's ICLR 2025 workshop paper, arXiv:2601.06505)
   37	Simulated SnAr plug flow reactor from the Summit package, kinetics from Hone et al. 2017; no external data
   40	Suzuki-Miyaura screen of Perera et al. (Science 2018), 308 ligand/base/solvent combinations, shipped in the repo as experiment_index.csv
   42	Proposal only; no dataset
   45	Synthetic only: Dixon-Price, Branin, and Beale surfaces adapted into multi-task generality benchmarks

   Changes made (commit 0258d7a)

   • Each of the 13 flagged summaries in _projects/json_summaries/ now states its data source with a citation, or says explicitly that objectives came from simulation, synthetic functions, or in-house measurements, or that no data were used. This also fixes two transcription errors from the video pipeline: "Hornel" is now Hone et al. (with the E-factor direction corrected to minimize), and "MOLDES" is now MolDAIS per the project title.
   • Added a data-provenance paragraph to Cross-Project Synthesis in main.tex (fifth paragraph, page 19) summarizing the patterns and recommending teams archive queried datasets in-repo.
   • Added 7 supporting references to latex/summaries-ref.bib (Shields 2021, Perera 2018, MoleculeNet, Kusner 2017, Harvard CEP, Movassaghi 2021, Truong 2026).
   • Replaced "Noted." with a substantive reply in RESPONSE_TO_REVIEWERS.md.
   • Recompiled copilot-main-fixed.pdf and regenerated copilot-main-diff.pdf (0 unresolved citations; MiKTeX's server was unreachable so I used TeXLive with the same latexmk -f -shell-escape invocation). The additions pushed some content by a page (25 pages now), so I verified every page reference in the response letter against the new PDF and updated the 15 that shifted, including two that were already off (Project 34's summary and the Acknowledgements).

   Notes: repos for projects 21, 36, and 45 are no longer public (matching the 4 "inaccessible" repos in the license tally), so those entries were verified via the video summaries and the teams' publications. Project 15's CSV row in AC-bo-hackathon-2024.csv lists an unrelated repo (jakobzeitler/interpretable-BO); the project page's janweinreich/best_batchers is the real one. I left the CSV untouched.

   | Branch