IBL new-cohort channel provenance v0.1

Status: paired-reference route closed; signal-only design required (26 July 2026)

Question

The prospective IBL v0.2 query found 407 held-out sessions with alternating 470-nm rows and wavelength-0 rows named None, but no labelled 415-nm reference. This audit asked whether the zero-labelled rows could be recovered as a genuine isosbestic channel from primary acquisition provenance.

No fluorescence outcome, feedback contrast, or condition-specific result from the held-out cohort was inspected during this audit.

Evidence

Acquisition methods

The dataset tag 2024_Q3_Pan_Vazquez_et_al corresponds to Pan-Vazquez et al., Pre-existing visual responses in a projection-defined dopamine population explain individual learning trajectories. Its methods state that GCaMP6f was recorded with a Neurophotometrics FP3002 using 470-nm excitation: 464 of 476 sessions at 50 Hz and 12 at 20 Hz. The published processing used a ±30 s rolling average as F0 and calculated (F - F0) / F0. No second optical reference channel is described.

• Open-access article and methods
• PubMed record

This is direct study-level evidence that the recordings are signal-only.

Historical extraction semantics

The public dataset records identify ibllib 2.38.0 as the producing version. In that version's extractor, wavelength 0 is explicitly None / “No LED ON”, while 415 nm is separately Isosbestic and 470 nm is GCaMP. The extractor reads LED states through the acquisition channel map and writes the corresponding wavelength and name into the processed table. Exact zero/None values therefore represent a matched no-LED state under the producing code, not an unlabelled 415-nm state.

• ibllib 2.38.0 fibre-photometry extractor
• IBL photometry loading guide

Public asset boundary

Representative held-out and development sessions expose photometry.signal.pqt and photometryROI.locations.pqt, but not the raw Neurophotometrics table or per-session channels CSV expected by the historical extractor. Consequently, the public archive does not permit an independent frame-by-frame reconstruction from raw LED states. That limitation does not create positive evidence for a hidden reference channel; it bounds the audit.

Conclusion

The convergent acquisition paper and historical extractor establish the most plausible and documented interpretation: these are 470-nm signal-only recordings interleaved with no-LED camera frames. The wavelength-0 rows must not be used as an isosbestic reference. The paired-reference v0.2 readiness failure stands, and the 407 sessions may be reconsidered only under a separately frozen signal-only protocol.

This conclusion should be revisited only if raw acquisition files or corrected primary metadata are published.