an enquiry into the habits of arrival
the arrival effect
the practitioner arrives; the archive changes pace
encounters per day · home → elsewhere · 10 qualifying journeys
The museum suspected that arrival was one of its working conditions. A different street, an unfamiliar door, the ordinary business of being somewhere else: the archive may change pace. The committee has compared the archive with a private record of journeys. The arithmetic distinguishes leaving home from moving on.
leaving home, moving on, returning
Fourteen of sixteen arrivals elsewhere were followed by more encounters. Eight of ten returns home were followed by fewer. But “before elsewhere” is not always home. Ten journeys leave home; six move between places already away. The committee now records them separately.
The home → elsewhere average rises from 2.8 to 18.6 encounters/day; nine of ten rise. The elsewhere → elsewhere average rises from 27.6 to 41.6; five of six rise. The latter is the more direct question about arrival itself, but its uncertainty is considerable. One journey made without a flight, recorded beside the sample, rises the same way on arrival. The title names an enquiry, not a demonstrated cause.
three days on either side
Each journey receives the same allowance: three full calendar days before its first departure and three after its final arrival. The intervening travel days are considered separately. Another flight inside either comparison window disqualifies that journey from this particular comparison.
Averages count days with no archived encounter too. An empty day does not establish the absence of a walk. Origin means the first departure airport relative to the practitioner’s residence; it cannot prove that every before-day was spent at home.
the numerical register
| Arrival | Journeys | Before/day | After/day | Change/day | Median before | Median after | Median change | 95% change interval |
|---|---|---|---|---|---|---|---|---|
| home → elsewhere | 10 | 2.8 | 18.6 | +15.7 | 1.5 | 15.0 | +13.8 | +7.2 to +24.6 |
| elsewhere → elsewhere | 6 | 27.6 | 41.6 | +14.0 | 19.0 | 22.2 | +13.7 | -26.3 to +49.8 |
| elsewhere → home | 10 | 15.3 | 3.5 | -11.7 | 7.0 | 2.5 | -6.2 | -20.6 to -3.6 |
| home → Thessaloniki · bus and ferry | 1 | 4.7 | 30.7 | +26.0 | — | — | — | not pooled |
| Thessaloniki → home · car and ferry | 1 | 49.3 | 37.3 | -12.0 | — | — | — | not pooled |
Means and medians are across journey-specific three-day averages, in encounters/day. Median change uses paired differences; it is not the difference of the two medians. Changes are computed before rounding. Intervals resample whole travel blocks; 26 journeys across 12 blocks are not 26 independent experiments. The split by origin was added after review. The original pooled away average, 12.1 → 27.2 (2.25×), mixes both origins and is not a home-to-away ratio. The two Thessaloniki rows are single journeys made without a flight (section 06), shown beside the flight sample and not folded into its means or intervals.
six arrivals are not six experiments
The six elsewhere → elsewhere arrivals come from just four travel blocks. Their average change is +14.0 encounters/day, but the approximate 95% interval runs from −26.3 to +49.8. The archive is compatible with several accounts of moving on. It has not isolated the act of arrival from the destination, the walking or the purpose of the trip.
Change in encounters/day. Dots are estimates; lines are 95% travel-block bootstrap intervals. The dashed line is no change. The post-hoc subgroup intervals are especially unstable with so few blocks.
Removing any one travel block leaves all three average changes pointing in their original direction. For the pooled away sample, the increase ranges from 10.3 to 18.9 encounters/day. This is a useful robustness check; it does not erase the wide interval above. Alternative connection rules also preserve the original pooled directions.
arrivals the committee invented
As a calendar reference, every day with three preceding and three following archive days can stand in for an arrival that never happened. Sampling sixteen such dates gives a central 95% range of −10.8 to +9.5 encounters/day for the average change; about 0.33% of draws reach the observed pooled away increase. For ten dates, the range is −14.0 to +11.9; about 3.38% reach a decrease as large as the home result. Measured against the origin split rather than the pooled figure: about 1.04% of ten-date draws reach the home → elsewhere increase of +15.7; for six dates the central range widens to −16.0 to +15.2, and about 3.18% reach the elsewhere → elsewhere increase of +14.0.
100,000 draws with replacement, seed 888, including zero-encounter days. This exercise ignores actual flight eligibility, seasonality and the dependence between arrivals, so those tail fractions are not causal p-values. A second reference shifts all real comparison windows together around the archive calendar, preserving their spacing and overlap: about 0.41% of nonzero shifts reach the away increase and 3.73% reach the home decrease. Circular wrapping has artificial endpoints and neither exercise makes travel random.
the walk changes less than the finding
The daily Apple Health distance records can be matched to all 26 qualifying journeys. On departures from home, recorded distance rises modestly, from 5.65 to 6.10 kilometres/day; encounters per recorded kilometre rise from 0.60 to 3.25. More recorded distance alone does not account for the observed difference in these windows.
| Journey | n | km/day, before → after | Encounters/km, before → after | Change/km | 95% change interval |
|---|---|---|---|---|---|
| home → elsewhere | 10 | 5.65 → 6.10 | 0.60 → 3.25 | +2.65 | +0.97 to +4.67 |
| elsewhere → elsewhere | 6 | 8.00 → 8.17 | 2.93 → 4.36 | +1.43 | -1.42 to +3.34 |
| elsewhere → home | 10 | 6.99 → 5.14 | 2.07 → 0.72 | -1.34 | -2.54 to -0.29 |
For each journey, divide the three-day encounter total by the three-day recorded distance, then average those rates equally across journeys. These are means of window rates, not the ratio of the displayed daily averages. “Distance” is Apple Health’s walking-and-running measure. Intervals use the same whole-block resampling as the count analysis; this follow-up is exploratory.
Between places already away, the rate also rises, from 2.93 to 4.36 encounters/km, but its change interval still crosses zero. The distance denominator sharpens the question; it does not settle it. Recorded yield depends on the available 8s, the route, recording choices and device coverage as well as attention. Recorded distance is not time spent looking.
what the wrist and phone can establish
The export writes all timestamps with the same UTC+3 offset. The analysis converts those instants to UTC, then cuts the comparison days in the departure or destination airport’s IANA time zone. Every resolved encounter timestamp in these windows agrees with its recorded local calendar date under that rule. Ground travel and unobserved portions of a day remain possible.
Overlapping watch and phone records are not added together. This reconstruction gives watch records priority, using phone records in the gaps; within each source kind, the latest-starting sample wins. Distance is spread uniformly through each sample interval and split at local midnight. The export does not establish the user’s Health source-priority settings, so this is an explicit approximation, not a reproduction of the app’s displayed totals.
Giving the phone priority retains all 26 journeys and preserves all three directions: changes of +3.30, +1.08 and −1.52 encounters/km respectively. A watch-only check retains 8, 5 and 8 journeys and also preserves the directions; its sample differs. Under both checks, the elsewhere → elsewhere interval still spans zero.
A journey enters this comparison only if every before- and after-day has positive recorded distance; missing records are not zero walking. All 26 satisfy that rule with phone fallback. This establishes recorded activity, not complete device wear. Apple documents walking-and-running distance and source priority. The museum’s broader walking summary is in the register of measurements. No daily Health records are published here.
in transit, still attending
A connection is not necessarily an interruption of the practice. Using a connection rule of under 24 hours, twenty-three of thirty-six stops contain timestamped encounters. Together they account for 867 encounters recorded inside the logged intervals between flights.
The archive’s most productive calendar day contains 479 encounters; its runner-up contains 170. That record day coincided with a connection. Of those encounters, 477 fall strictly inside the logged flight-time window; two precede the recorded arrival by a few minutes. The flight clock and the camera are not required to agree about when the visit began.
The record-day total comes from the encounter archive. Connection totals use resolved UTC timestamps and logged flight clocks, which may differ from actual arrival times. A connection includes time beyond the airport; its duration is not a measure of hours spent looking. Two short out-and-back visits are classified separately from connections. The exceptional connection falls on a travel day and is deliberately outside the main before-and-after windows; it answers a different part of the enquiry.
the unfamiliar may wear in
For eleven of the sixteen away arrivals with a full week free of logged flights afterwards, the average declines from 35.2 encounters/day in days one to three to 22.6 in days four to seven. The committee notes a possible early burst. It does not yet issue a timetable for the wearing-off of elsewhere.
The same eleven arrivals contribute to both periods. The five arrivals excluded from this follow-up average only 9.5 encounters/day in their first three days, versus 35.2 for those included. The apparent early burst belongs to this selected week-long flight-free sample; it should not be generalised to all arrivals. Flight-free does not establish a continuous stay in one place, and travel by land or sea can still occur.
the Thessaloniki exception, and the ship
The Thessaloniki annex holds 817 encounters in this snapshot, including the 8,888th. The practitioner left home for the city by bus and ferry on 23 August 2026 and returned by car and ferry on 9 September. Neither journey appears in the flight sample. Under the same three-day rule, the outward journey runs from 4.7 to 30.7 encounters/day; the return runs from 49.3 to 37.3. Going, the effect is the familiar one. Returning, the fall is slight, and the calendar offers a candidate reason: the three days after the return were long walks out of town, 13.8, 14.1 and 11.2 kilometres, rather than a resumption of familiar streets. The travel day home gave 53 encounters of its own, two of them at the ferry port.
The other journey without a flight is the ship. The M/V Sea Spirit left Longyearbyen on 25 June 2026 and returned on 5 July, with a new shore most days and often several. The instrument needs three still days on either side of a journey; the ship offered none, and its first days already stand in the flight sample as the after-days of the arrival at Longyearbyen. Recorded on its own terms, the nine full days at sea gave 57 encounters, 6.3 per day: 21 of them at Ny-Ålesund, the one settlement visited, and the rest from shore, ice and, in the pack ice, the ship’s own fittings. Arrival every day, and almost nothing to arrive at. The arrival effect appears to need a supply of 8s as well as a practitioner.
Both journeys were identified after the flight analysis and are recorded here with their dates; the annexes and the field station already publish the dates and places of their encounters. Thessaloniki enters the register as two separate rows and is not folded into the flight-based intervals. Folded into the means alone, it would move home → elsewhere from +15.7 to +16.7 and elsewhere → home from −11.7 to −11.8, on eleven journeys each. Its walking distance is not reconstructed. The outward before-window opens the day after a return by air, which the rule permits; the ship is recorded as a description, not an observation.
on journeys, counts and inference
The public finding is assembled from aggregate counts. This page adds no flight records to what the encounter archive already shows.
on method
The sample. This is a research snapshot prepared on 17 September 2026 and revised on 19 and 20 September 2026, not a live flight tracker. The archive contains 9,457 dated encounters and 53 without a date. Eighty-seven logged flight legs fall inside the dated archive’s coverage; historical flights outside that coverage and future bookings are excluded. Connected legs yield 51 journeys. Twenty-five have another flight in the comparison days, leaving 26 eligible journeys. Two journeys made without a flight are recorded separately in section 06 and are not among the 26. The result describes this selected sample, not every arrival.
Dates. Daily counts use the camera’s recorded local dates. Flight dates are interpreted as local departure dates; airport time zones and the exported clocks and durations resolve overnight arrival dates. Nine legs have a clock-versus-duration discrepancy exceeding thirty minutes. Their times are not treated as verified actual flight times. The transit enquiry uses 9,453 encounters with resolved UTC instants, so four dated encounters eligible for daily counts cannot enter timed transit counts.
Connections and home. Consecutive legs are joined when they connect at the same airport with a stop under 24 hours, except an immediate return to the starting airport. These brief round trips are visits, not connections. Home follows the practitioner’s residence at the time; a change of residence is represented in the classification. The page and its aggregate download withhold the flight records. The wider archive already publishes encounter dates, times and locations, from which travel can sometimes be inferred; this is not a promise of itinerary anonymity. The two journeys made without a flight are recorded with their dates, which their annexes already imply.
The comparison. Each eligible journey contributes the mean of its three before-days and three after-days. The difference between the away and home changes is +26.8 encounters/day, with a 95% bootstrap interval of +12.4 to +42.3. The pooled away after/before ratio is 2.25, with a broad interval of 1.29 to 5.09. These describe associations and are not causal effects. The pooled result combines departures from home and moves between places away; it is retained for continuity, not used as the home-to-away estimate.
Uncertainty. Sequences ending in a return home define travel blocks. The analysis resamples whole blocks jointly 20,000 times, with seed 888, retaining multiple arrivals from the same trip together. Percentile intervals assume those blocks are exchangeable. There are only twelve contributing blocks. The origin split uses those same joint resamples, omitting draws with no observations in the subgroup being estimated. Its intervals are exploratory and have no multiple-testing adjustment. No causal p-value is reported.
The later-week enquiry. The eleven-arrival follow-up contributes nine blocks. The days-four-to-seven change is −12.6 encounters/day, with an approximate interval of −23.7 to −0.4. This is exploratory, with no multiple-testing adjustment. It does not establish an attentional mechanism or a precise decay rate.
| Alternative rule | Away/home n | Away change/day | Home change/day |
|---|---|---|---|
| shorter connection rule (12 hours) | 10 / 8 | +17.1 | -15.0 |
| longer connection rule (36 hours) | 16 / 10 | +15.1 | -11.7 |
| omit inconsistent flight clocks | 14 / 8 | +14.2 | -15.0 |
| move the change of home one month later | 17 / 9 | +14.5 | -13.6 |
| anchor before-days to final arrival (see definition below) | 7 / 3 | +21.3 | -11.8 |
The arrival-anchored alternative uses the three full calendar days immediately before final arrival, instead of before first departure; both windows must still be flight-free. It therefore excludes many multi-day journeys and retains only seven away and three home arrivals. These alternatives change sample membership and effect size. They are checks on the counting rules, not independent replications.
The limits. The archive records encounters, not all walks or missed 8s. Flights omit ground and sea travel. A zero is not a measured unsuccessful search. The recorded-distance comparison above accounts for the denominator under an explicit reconstruction, but cannot establish full device coverage, walking time or time actively looking. Familiar streets are a possible confounder. A rough check on the public register, classing each day by a bounding box around the residence of the time, gives 2.6, 4.4, 3.7 and 4.0 encounters per home day for 2024, 2025, the first quarter of 2026 and the months since the change of residence: no visible decline, on a crude instrument. This enquiry does not estimate that trend more carefully. Later journeys could test the present definitions independently.
Sources: the museum’s encounter register, a private flight diary, the supplied Apple Health export (walking-and-running distance only) and, for the two journeys made without a flight, the practitioner’s calendar. Airport time-zone identifiers were cross-checked with OpenFlights; historical offsets use IANA time zones. Public aggregate summary. This page contains no individual flight records; the two journeys in section 06 are the only dated journeys it records.
The itinerary measures the distance travelled
The archive records what interrupted it