SDR-0005: Do not infer photometry channel identity from row position
- Status: Accepted
- Date: 2026-07-26
- Decision owners: project maintainers
- Related protocol/report: prospective IBL v0.2
Context
The prospective IBL query found 407 sessions from 18 new animals, but none had
labelled 415-nm isosbestic rows. They instead alternated labelled 470-nm rows with
wavelength-0 rows named None. ROI values were finite on both. Official IBL
extraction semantics distinguish wavelength 0 (“No additional signal”) from 415
nm (“Isosbestic”). Treating finite values or alternating position as sufficient
channel identity would pass a readiness gate only by changing the schema after it
failed.
Decision
Require explicit wavelength/name metadata or primary acquisition provenance before assigning signal or reference semantics. Do not infer an isosbestic channel from row alternation, finite values, column position, correlation, or analytical convenience. Unknown channels remain unknown and must fail adapters that require a reference.
Consequences
Some large public cohorts cannot enter paired-reference workflows despite having apparently regular tables. This reduces immediate sample size but prevents silent scientific relabelling. Signal-only analyses remain possible only under separately declared methods and protocols.
Alternatives considered
- Treat wavelength-0 rows as 415 nm because they alternate with 470 nm: rejected because official metadata assigns a different meaning and no primary provenance supports the reinterpretation.
- Infer identity from correlation with 470 nm: rejected because correlation cannot establish acquisition wavelength or biological inertness.
- Drop the reference requirement and run the frozen multiverse: rejected because all nine universes explicitly use reference correction.
Revisit trigger
Revisit for these assets if IBL publishes corrected metadata or acquisition-level documentation that identifies the wavelength-0 values as a specific optical channel.
Evidence added later
The producing ibllib version (2.38.0) explicitly maps wavelength 0 to “No LED
ON”, and the associated study methods describe 470-nm-only acquisition with no
isosbestic channel. The public archive does not expose the raw channel-map files,
but the two independent primary sources agree. The paired-reference route is
therefore closed unless new raw or corrected metadata appear. See the
channel-provenance audit.