Written guide to clock-day coverage in Biosignal Processing’s Actigraphy app. The main example contains 47 half-hour activity-index values across a 23.5-hour local day. A separate older-snapshot example below explains a repeated clock label on a 24.5-hour local day. The saved native workflow demonstrates elapsed coverage, explicit unavailable results and a complete export package. It contains no person, wearable-device benchmark or sleep-stage reference.
1. Bind the count definition to the selected recording
The app is identified as biosignal_. The saved native workflow loads a CSV, selects its count protocol, runs the requested workflows and exports the result. It uses app snapshot bce2b431… with the existing pinned Biosignal runtime; that source check does not accept a newly packaged release.
The selected activity_ channel contains 47 values, all equal to one. The protocol defines them as an arbitrary activity index with a native epoch of 1,800 seconds, bound to the recording fingerprint and selected channel. They are not steps, acceleration in mg or measurements from a named device.
The protocol’s preserve_ declaration requires the requested epoch to match the native epoch in this example. Keep the protocol, input CSV, channel and parameter record together. A protocol marked reviewed by the fixture author describes this synthetic input; it is not an exact-result review or a clinical validation.
The selected wear policy is all_, with the nonwear method set to none. Wear remains unclassified. Finite samples can support this declared coverage calculation without establishing that a device was worn. The analysis-settings lesson explains why a policy choice must remain attached to the reported result.
2. Keep timezone-aware time and elapsed time together
The fixture declares Australia/Lord_ and starts at 2026-10-04 00:00:00 +10:30. The following local midnight has offset +11:00. The NSW Government daylight-saving guidance identifies Lord Howe Island’s half-hour clock change; it must not be replaced with the mainland’s one-hour change.
The actual elapsed duration between these local midnights is 84,600 seconds, or 23.5 hours. Forty-seven complete 1,800-second epochs cover that duration. Their start coordinates run from 0 through 82,800 seconds; the final epoch’s support extends to the next midnight. A last sample timestamp alone does not describe the full covered interval.
Retain each timestamp’s UTC offset when reading the saved clock-interval table. On the same local date, three consecutive intervals are:
| Local start → stop | Elapsed seconds |
|---|---|
| 00:00 +10:30 → 01:00 +10:30 | 3,600 |
| 01:00 +10:30 → 02:30 +11:00 | 3,600 |
| 02:30 +11:00 → 03:00 +11:00 | 1,800 |
The middle row spans one elapsed hour despite its different wall-clock labels. The third row is a real half-hour interval. Removing offsets or forcing every row to 3,600 seconds would change the meaning. The complete saved table contains 24 intervals: 23 full hours and one half hour.
3. Read coverage as support, not an activity estimate
The daily row retains 84,600 expected seconds, 84,600 covered seconds and a coverage fraction of 1.0. Every saved interval also has coverage fraction 1.0. The recorded check verified the interval durations and nonoverlap; this guide reopens the tables without generating the clock intervals again.
The declared minimum daily coverage fraction is 0.8, so this row is marked usable. That threshold is a parameter of this example, not a universal research or clinical recommendation. Here complete_ is true while eligible_ is false.
The field hourly_ is true because the app publishes clock-interval support. The separate field hourly_ is false. The support table contains start, stop, expected seconds, covered seconds and coverage fraction; it does not contain an hourly activity mean or a 24-hour activity profile.
Wear-adjusted activity remains unavailable because the recording has no wear classification. Do not turn full finite-data support into a claim of full wear time. The saved epoch table retains unclassified wear and blank wear-adjusted quantities alongside the finite activity-index values.
4. Preserve the unavailable rhythm results and review state
Five requested rhythm outputs remain explicitly unavailable: L5, M10, relative amplitude, interdaily stability and intradaily variability. Their exported tables retain the status, reason and zero result-row count. The 23.5-hour local day is ineligible for the reviewed 24-hour rhythm calculation; interval support does not supply the missing hourly activity.
The interdaily-stability reason additionally requires at least two complete eligible 24-hour local days. That prerequisite is not evidence that this one-day recording produced an estimate. Keep an unavailable table with its reason instead of treating its absence of a coefficient as zero.
The selected output disposition remains proposal, and the exact result-review document says unreviewed with no review supplied. The bounded technical review of this synthetic workflow is separate from the application’s exact-result review state. Neither creates device, sleep-stage or clinical validation.
5. Retain the 39-file package and complete tables
The saved export contains 38 listed members plus manifest.json, giving 39 files in total. The listed members comprise:
| Members | Purpose |
|---|---|
| 33 CSV files | Metrics; input, protocol and timebase evidence; epoch and wear tables; daily and interval coverage; unavailable rhythm results; route statuses and export-accounting tables. |
| 3 JSON files | The exact result-review state, input receipt and parameter manifest. |
| 2 PNG files | The epoch-activity and daily-coverage views. These images do not replace the complete numeric tables. |
| 1 additional manifest | Identities of the 38 members, parameters, selected channels, environment and result metadata. |
For the clock example, retain series_, series_ and series_ with the rest of the package. The support CSV has all 24 rows and full offset-bearing start/stop timestamps. The daily CSV retains all eleven fields, including both availability flags.
This guide checks all 39 file identities and the structures of all 33 CSVs. It matches 166 saved fields to the native receipt: eleven daily fields, 120 interval fields, twenty unavailable-rhythm fields and fifteen daily-route-status fields. These are export-consistency checks, not a repeat of the scientific analysis or independent clock calculation.
The native viewer’s default pane clipped columns. The recorded visual review accepted the support table only after one native pane drag at a 1600 × 1000 viewport. Use the complete CSV when a visible table truncates timestamps or fields. That specific adjusted view does not establish general mobile, screen-reader or native-viewer accessibility.
6. Read repeated clock labels in an earlier snapshot
A separate retained native workflow used 49 half-hour values, all equal to one, across 5 April 2026 in Australia/Lord_. Its declared day begins at 00:00:00+11:00; the retained reference gives 88,200 elapsed seconds, or 24.5 hours, between local midnights. This is a different input and an older app snapshot, 50a4802a…, from the bce2b431… spring example above.
The saved epoch table contains two 01:30 labels with different offsets and different elapsed positions. Keep the offset, epoch index and elapsed coordinate together. Do not merge these rows by wall-clock text alone. The following four consecutive rows are transcribed from the complete 49-row CSV; no clock intervals were recalculated for this lesson.
| Epoch index | Onset (s) | Saved clock timestamp | Duration (s) |
|---|---|---|---|
2 | 3600.0 | 2026-04-05 | 1800.0 |
3 | 5400.0 | 2026-04-05 | 1800.0 |
4 | 7200.0 | 2026-04-05 | 1800.0 |
5 | 9000.0 | 2026-04-05 | 1800.0 |
Epochs 3 and 4 are distinct saved observations: their onsets are 5,400 s and 7,200 s. Their identical 01:30 wall-clock labels do not make them duplicate samples. On a narrow screen, focus a table and use the arrow keys to reveal its remaining columns.
Read all ten older daily fields
The older native bundle preserves the complete daily row below. Its hourly_ value is false. Keep it with that older result; it differs from the interval-support availability in the spring example above. The older row has no separate hourly_ column. Missing columns must not be filled from a different run.
| Saved field | Saved value |
|---|---|
day_ | 0 |
clock_ | 2026-04-05 |
expected_ | 88200.0 |
covered_ | 88200.0 |
daily_ | 1.0 |
complete_ | True |
usable_ | True |
eligible_ | False |
hourly_ | False |
coverage_ | qualified_ |
The count protocol defines an arbitrary activity index with preserve_ epochs, and the coverage policy is all_. All 49 saved epochs have unclassified wear, blank wear fractions and blank wear-adjusted activity. Complete finite support does not establish that a device was worn or that a rhythm estimate exists. The older result remains proposal with an unreviewed exact-result review.
Distinguish a status row from a computed result
Each unavailable rhythm output publishes one explanatory row. The separate row_ inside that status table is zero, meaning no computed result rows. The route ledger's published_ is one because the explanation itself was saved. Neither count is a zero-valued rhythm coefficient.
| Route | Published status rows | Computed result rows | Status |
|---|---|---|---|
l5 | 1 | 0 | unavailable |
m10 | 1 | 0 | unavailable |
relative_ | 1 | 0 | unavailable |
interdaily_ | 1 | 0 | unavailable |
intradaily_ | 1 | 0 | unavailable |
For this older workflow, the recorded reason keeps hourly and rhythm results unavailable when epochs cross a clock-hour boundary or include a partial clock hour. Interdaily stability also requires at least two complete eligible 24-hour days. The full reasons, 49-row epoch table and ten-field daily row are retained in the evidence download.
The older fall export has 32 CSVs, three JSON files and two PNGs, plus one manifest: 38 files total. Its payload has no clock-interval-support table. Keep it separate from the spring bundle's 39 files. The native fall table clipped long field names and the coverage-basis text; its nominal left/right screenshots are byte-identical, so they do not establish horizontal scrolling or complete text visibility. Read the complete CSV values.
A saved-only source review on 11 November 2026 finds seven selected helper functions unchanged between 50a4802a… and bce2b431…. The main analysis and view functions differ, and the later snapshot adds an interval-support helper. Those matches do not turn the older fall workflow into a run of the later app. The retained 25 native checks reuse one synthetic fixture; they are not 25 new scientific cases, and this lesson repeats no native Run or export.
7. Keep each example’s eligibility boundary
The bce2b431… support path for the 23.5-hour example has explicit conditions for the bound count protocol, declared timezone, finite nonoverlapping epochs and this form of half-hour clock advance. It does not split arbitrary aggregate epochs across clock boundaries or infer missing activity. The separate historical reversal example does not extend that current-source walkthrough to other dates, timezones, incomplete epochs, cancellation or different wear policies.
The saved claim-boundary table explicitly keeps sleep staging, endogenous circadian phase, disorder diagnosis and clinical interpretation outside the result. The synthetic all-one signal cannot answer those questions. Keep the source data and protocol separately from the result package; an input receipt is not a replacement for the original recording.
8. Explain two cases with daily coverage but no interval support
Two separate saved checks reviewed on 10 November 2026 call the public analysis function directly on synthetic inputs. They are not native GUI workflows or native exports; the CSVs were written by the test harness. The original positive native example and its package above remain separate. These checks help explain why a daily coverage number alone does not promise an interval-support table.
Distinguish a retained missing value from shifted finite epochs
| Saved setting or observation | Missing-value case | Clock-boundary case |
|---|---|---|
| Retained epoch rows | 47 | 46 |
| Declared local start | 2026-10-04T00:00:00+10:30 | 2026-10-04T00:15:00+10:30 |
| Epoch duration (s) | 1800.0 | 1800.0 |
| Activity input | Index 4 is NaN; other 46 values are 1 | All 46 values are 1 |
| Finite support (s) | 82800.0 | 82800.0 |
| Minimum daily coverage fraction | 1.0 | 1.0 |
| Interval helper status | unallocatable_support | unallocatable_support |
Both protocols retain arbitrary activity-index values and preserve_exact epochs. The missing-value case keeps row index 4 rather than deleting it. Its CSV has the literal NaN in the activity field and zero finite support. That is missing activity, not a measured activity value of zero. It does not test a generally omitted-row input.
The second case begins fifteen minutes after local midnight and retains 46 complete finite epochs. Its first two saved rows are shown alongside the missing row from the first case. CSV timestamps use a space between the date and time; the timezone offsets remain part of each value.
| Case | Epoch index | Exact CSV timestamp | Activity | Finite support (s) |
|---|---|---|---|---|
| Missing value | 4 | 2026-10-04 02:30:00+11:00 | NaN | 0.0 |
| Clock boundary | 0 | 2026-10-04 00:15:00+10:30 | 1.0 | 1800.0 |
| Clock boundary | 1 | 2026-10-04 00:45:00+10:30 | 1.0 | 1800.0 |
In the saved observer record, the clock-boundary case’s second epoch runs from 14:15 to 14:45 UTC on 3 October. It crosses the interval edge at 14:30 UTC. The first epoch fits its interval, but the next does not; the helper returns an empty support table instead of dividing the aggregate activity across intervals. No timezone intervals or source activity were recalculated for this lesson.
Read the identical daily rows with their different inputs
The two daily CSVs are byte-identical despite distinct input identities and refusal causes. Each has exactly these ten fields. The saved fraction is the retained decimal representation of 46/47. These negative cases require minimum daily coverage 1.0, whereas the positive native example above uses 0.8; do not combine their parameters.
| Saved daily field | Exact text in both CSVs |
|---|---|
day_index | 0 |
clock_date | 2026-10-04 |
expected_day_seconds | 84600.0 |
covered_seconds | 82800.0 |
daily_coverage_fraction | 0.9787234042553191 |
complete_clock_day | False |
usable_day | False |
eligible_for_24h_rhythm | False |
hourly_coverage_available | False |
coverage_basis | qualified_epoch_elapsed_support |
Completeness, usability, 24-hour eligibility and hourly coverage are all false here. The hourly_activity_available column is absent from these ten-field daily tables. It must not be invented or copied from the positive native example’s eleven-field table. No public interval-support table, hourly activity table or mean 24-hour profile was published by either saved call.
Keep the unavailable result rows and reasons
Both cases preserve the same five unavailable rhythm results. Each saved row retains result-row count 0. That count describes the absence of an estimate; it is not an estimate of zero.
| Result | Saved status | Exact saved reason |
|---|---|---|
| l5 | unavailable | the epochs cross a clock-hour boundary or include a partial clock hour; hourly and rhythm outputs remain unavailable |
| m10 | unavailable | the epochs cross a clock-hour boundary or include a partial clock hour; hourly and rhythm outputs remain unavailable |
relative_amplitude | unavailable | the epochs cross a clock-hour boundary or include a partial clock hour; hourly and rhythm outputs remain unavailable |
interdaily_stability | unavailable | at least 2 complete eligible 24-hour local days are required; the epochs cross a clock-hour boundary or include a partial clock hour; hourly and rhythm outputs remain unavailable |
intradaily_variability | unavailable | the epochs cross a clock-hour boundary or include a partial clock hour; hourly and rhythm outputs remain unavailable |
Eight selected routes required seventeen routes through their dependencies. The retained 40 and 43 runtime predicates describe one case each; they are not 83 independent scientific cases. The helper was observed through the public caller, with one call and return per case, without a separate direct helper invocation. Preparation remained proposal and the exact-result review remained unreviewed.
Preserve the execution date and source boundary
The two calls used isolated bce2b431… app source and 202a2963… shared UI source while the original app still had its older identity. At this addition’s dated inspection, those two executed files match the corresponding original files byte-for-byte. The accepted package’s corresponding files differ. A later source match does not turn these historical public calls into native runs or accept a rebuilt package.
The source’s actual public caller invokes the interval helper under explicit conditions. Its older “private, unwired” docstring is not used as evidence that the helper is unreachable. The saved helper output retains its five column names with zero rows and the status unallocatable_support. This is a refusal to allocate interval support, not a failure to preserve the separate daily result.
Download all twenty saved CSV tables, exact parameters and dated source relationships. This addition reopens retained files only. It adds no scientific calculation, clock-interval generation, native action, export, graphical review or general missing-data, device, clinical, hosted or whole-product acceptance.