IBL feedback signal-only results v0.3.2

Status

This is the completed report for the prospectively frozen IBL feedback multiverse, with the two documented amendments in v0.3.1 and v0.3.2. The machine-readable result is ibl-feedback-results-v0.3.2.json. Its result SHA-256 is 5c9b4e844380c8bd61ae0589f6c638920570f2c0e27c7100ea72b042c1caeef7.

The run loaded all 383 requested sessions from 18 animals and summarized 224,272 feedback events. Nine universes executed successfully, nine were retained as structurally incompatible, and none failed during execution.

Amendments and outcome access

The original v0.3 artifacts remain unchanged. Before fluorescence outcomes were accessed, v0.3.1 corrected the implementation to match the frozen prose: session contrasts are weighted equally within animal rather than pooling all events within animal.

During the first session smoke test, raw IBL timestamps were found to exceed the regularity required by the AsLS implementation. Aggregate effects were not inspected, but fluorescence for that session had been accessed. Version v0.3.2 therefore records a narrow compatibility amendment: all six AsLS universes are retained as incompatible with reason asls_requires_regular_sampling. No resampling rule was added after outcome access. Three rolling/subtractive universes remain incompatible because the rolling estimator does not define a baseline suitable for subtraction.

This is a product-compatibility finding, not evidence against AsLS and not a scientific null result. It motivates a future prospectively specified resampling feature.

Results

The reference universe—published rolling baseline, standard divide window— estimated correct minus incorrect feedback at 0.01582 fractional dF/F (95% CI 0.00647 to 0.02517; paired animal-level p = 0.00235).

Across the six divided dF/F universes, every estimate was positive and every 95% interval excluded zero. Estimates ranged from 0.01582 to 0.02210, with a median of 0.01873 fractional dF/F. Across the three subtractive universes, every estimate was positive and every interval excluded zero; estimates ranged from 0.0000911 to 0.0001225 acquired-fluorescence units, with a median of 0.0001019. These two lanes use different units and must not be numerically pooled.

Leaving out each animal in turn did not reverse the reference effect. All 18 leave-one-animal-out estimates were positive, ranging from 0.01292 to 0.01688 fractional dF/F.

Interpretation

The direction of the correct-minus-incorrect feedback contrast is robust to the executable baseline estimator and event-window choices in this frozen multiverse, and is not driven by any single animal. This supports a stable descriptive contrast in this public cohort. It does not establish that the same result will survive new preprocessing families, regions, cohorts, or a prospectively specified resampling implementation for AsLS.

Figures

The specification curves are separated by units:

• Divided fractional dF/F
• Subtractive acquired fluorescence

The figures are generated by plot_ibl_feedback_signal_only_v32.py.