Skip to content

Paper-figure correspondence

This register prevents a public-data analysis from being described as a paper reproduction merely because it uses the same recordings. It maps each worked example to an exact source panel, or states that no such claim is intended.

Current examples

Worked example Source target Current label What matches Material departures
DANDI 000971 reward analysis Seiler et al. (2022), Figures 3E–F and 4E–F Partial source-panel reproduction plus adapted robustness analysis the new panel matches rewarded/unrewarded PSTHs, DMS/DLS, phenotype, session z-score, and the 0–1.5 s maximum-minus-minimum statistic one checksum-pinned session from two animals per phenotype; no training-stage model; the separate robustness result intentionally uses fractional dF/F, IRLS, and a mean-window estimand
DANDI 000971 event kernels no source panel; methodological extension of the same event data Adapted reanalysis public cohort, events, regions joint regularized kernels and held-out prediction were not source-paper analyses
DANDI 000251 transient validation Kim et al. (2020), Figure 2E–H, especially teleport-aligned traces and normalized peaks Adapted construct-validation analysis public standard/three-teleport sessions and the prespecified 0.6–2.1 s response interval detects peaks across continuous traces; event enrichment tests detector behavior and does not reproduce the source RPE/value model or normalized-peak statistic
Public IBL feedback report none Product demonstration public observations and animal-level report no source-paper panel is claimed
IBL–Unspool longitudinal forecast none Cross-package reanalysis public sessions and prospective held-out comparison new coarse neural summary and forecasting question
Simulation tutorials none Method illustration known generating truth synthetic data are implementation evidence, not biological evidence

The Seiler source figures plot rewarded and unrewarded peri-stimulus time courses by phenotype and training stage, then quantify a maximum-rewarded minus minimum-unrewarded peak score. The Kim source figure plots condition-aligned calcium traces and normalized peak/residual summaries across teleport conditions. Those panel semantics—not superficial colors or layout—define the reproduction targets.

Priority reproductions

The bounded Seiler reproduction is now the first implemented source-aligned figure. The next priorities are:

  1. Kim Figure 2E–H, DANDI 000251. Reproduce three-teleport population traces and normalized peak summaries before layering detector sensitivity on top.
  2. Published peri-event bootstrap/permutation example. Choose a public animal-level dataset whose source statistic and resampling denominator can be matched exactly, then compare the package's simultaneous-band route.
  3. Multi-site or multi-color source panel. Select data with independent optical calibration so association is not mistaken for successful unmixing.

Acceptance criteria

A reproduction moves from the priority list into the examples only when:

  • source assets are immutable or checksum-pinned;
  • source methods and code have been read and cited;
  • the figure-generating script starts from those assets, not digitized pixels;
  • parity fixtures cover intermediate preprocessing and the final statistic;
  • the figure shows the correct independent unit and uncertainty denominator;
  • every departure is visible beside the figure; and
  • a failed or discordant reproduction is retained as evidence.

Primary sources