# Optotagging protocol amendment 01

Date: 2026-10-09. Applies to optotagging-protocol-and-extract-spec.md (SHA-256 524a7f62ec1cb89fd677c8b464c5da638252d5f8be288f32fc13fc49217d8fb6), DANDI 000022 release 0.251116.2247, SST session 794812542 and wild-type session 767871931.

## Status and basis

Saved before any recording spike-response analysis. The user reports 25 trials per level for SST, 15 per level for wild-type, and no per-unit observation intervals. These statements are the basis for this amendment and have NOT yet been independently verified against the source metadata. The named optotagging-metadata.zip was not found by exact managed-file lookup, a broader optotagging managed-file lookup, or an exact relative-path check in the current session directory. No recording dates, genotype strings, pulse conditions, trial counts or coverage bounds have been newly verified. Source metadata review remains mandatory before response analysis. This amendment fixes handling of the reported limitations; it is not a completed metadata review.

## Trial eligibility and descriptive wild-type control

Keep the original minimum of 20 usable trials per unit and stimulus level for inferential response classification. Do not pool levels, pulse conditions or animals to achieve it. Keep the highest recorded level as the primary level and lower levels as separately reported secondary conditions, subject to confirming isolated 10 ms single pulses in the source tables. Trial counts after exclusions, rather than scheduled counts, govern eligibility.

SST's reported 25 trials per level can support the original analysis only if at least 20 remain usable after exclusions for that unit and level. Do not replace the highest level with a lower level if it fails the threshold. Mark the primary result unavailable and retain descriptive results as appropriate.

Wild-type's reported 15 trials per level are below the original threshold. Retain its quality-passing units as a descriptive artifact control at each level. Show individual-trial rasters, unsmoothed 1 ms peri-event rates, baseline and response rates, paired rate differences, and fractions of trials containing a spike. Report the distribution, median and interquartile range of first spikes in [2,8) ms conditional on a spike, with the numerator and trial denominator. Do not assign wild-type inferential responder/nonresponder labels, p-values, q-values, significant onset latencies, or a responder percentage. In figures/tables use 'descriptive only: fewer than 20 usable trials', never 'no response'. Do not interpret a weak or absent wild-type transient as evidence that artifacts are absent in SST.

The primary Benjamini-Hochberg family comprises all eligible primary-level tests; with wild-type ineligible this is SST only. Do not insert wild-type tests with artificial p=1. Lower-level inferential tests retain their separate prespecified joint correction across eligible unit-by-level tests. There is no between-genotype significance test; one mouse per group does not establish a population genotype effect. Label differences in available trial counts and nominal stimulus levels, and do not assume equal optical power.

## Coverage assumption replacing mandatory per-unit intervals

The original protocol required demonstrated per-unit observation coverage. In the absence of per-unit observation intervals, the amended analysis may proceed conditionally under the following explicit assumption, only after metadata review:

For a unit passing the fixed quality filters, its recording and spike detection are assumed to remain active throughout each selected event's [onset-0.100 s, onset+0.100 s) window, provided that window is within documented acquisition coverage for the associated probe/session and does not overlap any known invalid interval or acquisition interruption. This is assumed per-unit coverage, not measured per-unit coverage.

Record coverage_status='assumed_from_acquisition_metadata' for such unit/trial pairs and retain the exact source of acquisition bounds. A session_start_time alone is not an acquisition duration or proof of coverage; stimulus timestamps establish event times, not electrode availability. A presence_ratio >0.95 is a unit-quality check, not proof that the optotagging interval was recorded continuously. Do not reconstruct coverage from first/last spike times or treat a silent trial as missing solely because it has no spikes.

Use any available probe-specific recording start/stop, acquisition timestamps, recording interruptions, and source invalid intervals to reject affected events. Missing per-unit intervals do not by themselves exclude all units. Do not invent intervals or convert missing invalid-interval metadata into 'verified no invalid intervals'. If the source explicitly contains an empty invalid-interval table, record 'no intervals annotated', which also does not prove uninterrupted recording.

If no acquisition coverage evidence is available at either probe or session level, label coverage 'unverified'. Before inferential analysis, obtain exporter confirmation that spike extraction covers all requested event windows for every supplied unit and that known recording gaps are reported. If that confirmation and supporting acquisition bounds remain unavailable, retain only descriptive outputs with an unverified-coverage limitation; do not silently present them as satisfying the primary protocol. Never fill an unverified missing observation with a zero count. A zero count is valid only for a delivered complete event window under the stated coverage assumption.

All qualifying results based on this assumption must be described as conditional on uninterrupted unit observation during optotagging. A full-session quality score cannot rule out short local dropouts or sorting instability.

## Unchanged analysis choices

All original quality cutoffs, pulse duration tolerance, event-isolation requirements, artifact masks, primary [-8,-2) versus [2,8) ms windows, effect-size thresholds, trial pairing, randomization procedure, SST latency definition, seed, and stricter [3,8) ms sensitivity analysis remain unchanged. Keep all levels separate. Retain spikes in artifact windows in the source extract and show them as excluded context. Wild-type's descriptive status is the explicit exception to the original inferential latency/reporting workflow.

## Metadata review required before spike analysis

Verify the exact source assets/release, genotype and recording dates from session_start_time; distinguish acquisition dates from release/upload dates. Inspect complete stimulus tables for 10 ms single pulses, condition/level units, counts, chronological spacing and usable events at each level. Inspect the full unit-table schema, quality metric definitions, missing values, electrode/probe mapping and quality-filter exclusions without opening spike responses. Inspect acquisition bounds and invalid intervals, then document which coverage claims are observed, assumed or unavailable. Confirm that the extract retains every requested window and units with zero spikes. Record discrepancies before response analysis; do not resolve them by inspecting which choices yield responses.

The pending spike extract remains unchanged: all-unit spike timestamps in the union of [-100,+100) ms windows around every candidate single 10 ms pulse at every level, complete stimulus and unit tables, with no response filtering or artifact blanking. Add acquisition bounds/gaps and an extraction-completeness statement; do not fabricate per-unit observation intervals.

No spike analysis has been run, and no response results or figures are available. The next dependency is making the metadata archive accessible for the required review.
