DOSSIER · Storm · Japan / Yokohama, Kanagawa (JMA Yokohama station, Port of Yokohama) / Chiba Prefecture · reviewed 2026-10-24
Typhoon Dujuan shut Yokohama for 45 hours, the backlog sailed within four days, and the September customs flash still rose.
When Typhoon Dujuan closes the Port of Yokohama for 45 hours into the Silver Week holidays, does a Kanto vehicle exporter such as Nissan at its Honmoku wharf wait for the next Yokohama sailing or truck its lots to Tokyo or Nagoya, and who pays the week of delay: the exporter in detention and demurrage, the carrier in a missed call and ship time, or the importer in Australia, the United States or the United Arab Emirates?
Edition · Data sheet · Sample · Get it · Preview: chain · Constants · Conjectures
INDEPENDENT REVIEW · REWORK
Rework kept open by the reviewer. Written by atlas-analyst, reviewed by atlas-counter-analyst on 2026-10-24. Two rounds at most; what the reviewer and the author still disagree on is kept, not resolved by majority.
Review notes, as written
Author's points for the reviewer. The 45-hour closure, the 90 percent yard density and the waiting times rest on one forwarder update (K1 to K4); no port authority notice was found. The crane thresholds are the Kawasaki district rules of the Keihin port, not Yokohama's own (K5 to K7). The Port Facts leaflet cited in organisations.json carries 2020 statistics; the 2025 tonnage and TEU come from the city's December 2025 monthly tables (K18 to K20). Free-time terms are Hapag-Lloyd's 2024 Japan tariff; ONE and Maersk general tariffs were not retrieved (K13 to K17). Yokohama Customs released a port-level provisional September flash on 21 October (K26 to K29, K32): value up, units down, which the dossier treats as an early read, not as the series. PortWatch figures are AIS-modelled estimates in UTC days; JMA days are local (C7). Quotes K13 to K16 are assembled from the cells of the tariff tables in reading order, and K5, K7 and K17 join sentences that the PDFs break across lines. Counter-analysis (atlas-counter-analyst, 2026-10-24, docs/night-2026-10-24/typhoon-dujuan-yokohama-port-2026/COUNTER_ANALYSIS.md). All 33 constants verified at the source: values, units and dates right; K13 to K16 are cell assemblies, confirmed current by Hapag-Lloyd's January 2026 Japan export sheet (effective 1 April 2024 until further notice). REWORK on the trajectory: the September HS 87 value is already determined by K26, because the 2026 ratio of HS 87 to the flash automobile value ran 1.202 to 1.278 in every month January to August, which puts September at 208 to 221 billion yen, above the 187.8 billion condition; the loss branch is excluded before the horizon opens, and 187.8 billion sits inside the January to August band (January 157.2, May 171.4), so the branch is misnamed. The November point (30 to 42 calls a day) is consistent with the November 2025 PortWatch mean of 39.8 as a monthly mean, too narrow for daily values (29 to 52), and is identical in both branches. REWORK on C1 and C2: the 21.1 m/s figure is a gust, the crane work-stop (10 m/s), advisory (12 m/s) and warning (25 m/s) criteria are ten-minute means; the right comparator is the JMA maximum mean wind of 13.7 m/s on 21 September, which gives the same conclusions. The MLIT model rule (2012, revised 2016) that the Kawasaki rule derives from uses a 16 m/s gust as the reference fixing threshold and 30 m/s as the mandatory one, so lashing cannot be said to have been unforced at Yokohama. REWORK on C10 and A4: Japan records exports on the departure date of the carrying vessel (Osaka Customs yearbook, 統計計上の時期), not on the permit date; an intra-month shift stays invisible, a shift across 30 September would not. REWORK on J2, J5 and L9: the Yokohama-minus-national unit gap in the 2026 flashes ran from minus 11.6 to plus 12.7 points without a typhoon (March minus 5.4, April minus 7.6, May minus 11.6, June minus 10.9, August plus 12.7), so a September gap of minus 5.2 points is within noise and the match with 6.25 percent is a coincidence; the September flash puts the unit fall in trucks (minus 14.0 percent) with passenger cars at minus 4.0 percent, and the August gap of plus 12.7 confirms the base effect the author lists as a falsifier of J5. The INCONCLUSIVE supply checks of J2 and J5 can now cite the Nagoya September flash (automobiles 541.0 billion yen, plus 22.7 percent, a record, but already plus 29.7 percent in June) and the Tokyo September flash (no finished-automobile line). Second sources found for the chain: SEKO Logistics repeats the 45 hours, 90 percent and 1.5 to 2 days with dates 20 to 22 September (same upstream, one source); Logfret on 22 September saw no evidence of a blanket shutdown; a Japan Coast Guard notice placed the Yokohama and Kawasaki districts of Keihin port under first alert and anchoring restraint from 08:00 JST on 20 September with cargo-handling stop criteria to be observed (page since removed, search extract only); Kuehne+Nagel's 16 to 22 September update already had waiting around one day and reefer plugs at 95 to 100 percent, so the typhoon's increment on the weekly average is about 0.2 days. PortWatch, all ship types: Tokyo 2 and 5 calls, Kawasaki 0 and 3, Chiba 4 and 16, Nagoya 11 and 22 on 20 and 21 September UTC; Yokohama ro-ro calls 0, 0, 0 on 19 to 21 September then 8, 3, 10, with ro-ro export tonnage 0 then 7,074 t on 22 September, which supports L8 and J3 as a delay rather than omitted calls; container calls 17 on 22 September against a mean of 12.9, which weakens A2; September container export tonnage 1,127,052 t against 1,122,720 t in August, the A4 discriminator, which supports shift over loss. Organisations accepted; zero beneficiaries is plausible because every alternative gateway closed on the same days; the Port and Harbor Bureau quote lies beyond the part of the 24 MB leaflet I could download; importers in the United States and the United Arab Emirates named in the question have no entry. No source publishes the three series side by side. Rework round 1 applied on 2026-10-24T01:37Z: items T1 (rebuilt on the November PortWatch condition), C1, C2, C10, L9, A3, A4, J2, J5, constants K34 to K57, figure F27, organisations.json Isuzu evidence. Counter-analysis round 2 (atlas-counter-analyst, 2026-10-24, final reconciliation pass, COUNTER_ANALYSIS.md section Round 2). Constants K34 to K57 all verified at their sources: flash rows verbatim with release dates 21 May, 17 June, 16 September and 21 October 2026; K41 row verbatim; K42 and K43 verbatim in the MLIT model-rule overview PDF (the 18 March 2016 date is printed on its companion revision notice, the PDF whose number ends in 528); K44 verbatim, file dated 23 January 2023; the eight PortWatch statistics queries behind K45 to K55 re-run with identical responses; K56 and K57 verbatim, Nagoya releases of 21 October and 22 July 2026. F27 equals the April series value; J2 high end recomputed 221,136,614 thousand yen, J5 quantity 24.3 recomputed. C1, C2, C10, L9 (CONTRADICTED), A3, A4 (SUPPORTED), J2 and J5 accepted; both checks of J2 and J5 now PASS with substance. The Isuzu evidence from the City of Fujisawa document (25 April 2025) is verbatim; the reviewer's round-1 claim that the MLIT 2014 Yokohama port-plan deck (common PDF ending in 953) placed Isuzu Fujisawa and Fuso Kawasaki in Yokohama's hinterland was wrong, the full deck fetched in round 2 names Toyota, Nissan and Subaru and never Isuzu or Fuso; the analyst's deviation was correct. REWORK remains on one item, trajectory T1: the branch condition (one or more November days with fewer than 29 calls) does not isolate a closure signature. Days below 29 occurred in seven of the eight months on record without any closure (October 2025: 4, December 2025: 4, July 2026: 1, August 2026: 6 including the 11 August holiday at 23, September 2026: 3 besides the closure days, October 2026 to the 16th: 3); only November 2025 had none. Days below 15 occurred only on the three closure days (10, 3, 9) and on 31 December 2025 (7). As written, the disturbed branch and its implications (another 1.5 to 2 days of ship time per call, a second delayed lot of trucks, a second claim window) would be declared on an ordinary month, and the 3 and 23 November holidays named by A1 would trigger it; the two explanations are confounded by construction. The undisturbed branch's daily range 29 to 52 is also unlikely at its low end, since the 2026 monthly minima without a closure were 26, 23, 27 and 19. The monthly-mean and ro-ro path points are accepted. Suggested fix for the author: one UTC day below 15 calls, or two consecutive days below half the trailing 30-day mean, with the November holidays excluded or tested apart. Verdict stays REWORK, status stays DRAFT.
What this dossier holds
57 constants read verbatim from their sources, 10 bounded constraints recomputed at every build, a 9-link causal chain with a status per link (3 untested, 1 contradicted), 4 alternatives weighed with their discriminators, 5 conjectures with their checks and falsifiers, 1 conditional trajectories with paths by period and implications by organisation, and the independent counter-analysis with its disagreements kept. Below the sample and the offer, a preview: the chain with its statuses, 5 of the constants with their quotes, the conjectures and trajectories by statement. The rest is the product.
A sample of the data
The last 5 points of 3 of the 5 series. Free downloads carry the last 12 points. Premium opens the available full series, dossiers and exports; restricted series remain cited, not redistributed.
| Series | Family · unit | latest | in total | ||||
|---|---|---|---|---|---|---|---|
| Peak wind gust, Yokohama | Physical · m/s | 14.8 | 14.4 | 20 | 14.3 | 9.1 | 115 |
| Vehicle exports, Yokohama customs | Economic · thousand JPY | 195,586,456 | 171,437,740 | 223,483,023 | 215,703,179 | 220,784,602 | 8 |
| Daily rainfall, Yokohama | Physical · mm | 0 | 0 | 0 | 0 | 0 | 115 |
Periods run from 2026-10-19 to 2026-10-23 on the first row; hover a value for its period. Downloads and provenance: data sheet →
PREMIUM · 49 USD A MONTH
One plan, the whole published library.
- Access to every published pack, old and new, while your official Premium right is valid; new packs join the same library
- Three formats per pack: dossier bundle as JSON, organisations as CSV and redistributable series as CSV
- Provenance, licences, hashes and review notes where available; restricted observations withheld and unfinished dossiers labelled honestly
- A member key for available full data on the JSON API and MCP; no free newsletter subscription without separate consent
Library downloads and full API/MCP access require a valid official Premium right. After it ends, no new retrieval is available. Keep files already downloaded. Restricted observations remain withheld; incomplete review and content limits are listed before subscription.
Premium details and subscription →No individual pack purchase. Free summaries and samples need no account. The free weekly requires separate consent.
Preview: the chain
From the physical event to the decision, one link at a time. Observed is seen in the series; consistent agrees with them without being isolated; untested names the series that would test it; contradicted is what the data refused. The evidence behind each link, and what validated it, are in the export.
- L1 · PHYSICALConsistentlag 0–1 days
Typhoon Dujuan rain and gusts at Yokohama, 20-21 September local days Port closure of 45 hours
Two local days of 117.0 and 223.0 mm and a 21.1 m/s peak gust closed the port for 45 hours; the gust alone crosses the crane work-stop threshold, and ships are evacuated from berths ahead of a typhoon.
- L2 · LOGISTICALObservedlag 0–1 days
Port closure Collapse of port calls and container exports, 19-21 September UTC
Port calls fell to 10, 3 and 9 on 19, 20 and 21 September UTC against a 108-day mean of 36.3, and container export tonnage to zero on 20 September; the 3-call day is the series minimum.
- L3 · LOGISTICALObservedlag 1–4 days
Reopening Catch-up of calls and exports, 22-25 September UTC
Calls rose to 37, 49 and 61 on 22, 23 and 24 September, the 61 the series maximum, and container exports reached 115,333 t on 22 September, the series maximum; the four days 22-25 September more than covered the tonnage shortfall of 19-21 September.
- L4 · LOGISTICALConsistentlag 1–7 days
Backlog at 90 percent yard density, over the holidays Vessel waiting of 1.5 to 2 days through the week of 23-29 September
Having lost 26.8 percent of a week's berth hours, the terminals could not serve the arriving ships at once; the forwarder update reports waits of 1.5 to 2 days, a 7-day average of 1.2 days, and names the holidays as a second cause.
- L5 · ECONOMICUntestedlag 0–30 days
Delay of about 3.4 days per box and per ship Detention and demurrage invoices to exporters, ship time absorbed by carriers
With 5 days of detention and 7 days of demurrage free time, holidays excluded, a 3.4-day disruption leaves most export boxes inside free time; the cost of the shift falls on the carriers' schedules and ship time. The Atlas has no series of detention invoices or of per-vessel waiting, so who paid is not measured.
- L6 · ECONOMICConsistentlag 30–40 days
Shift of loadings inside the month September customs value for vehicles within the January-August band
A shift of a few days inside September cannot move a monthly customs total; the port-level provisional flash of 21 October prints automobile exports of 173.1 billion yen, up 9.1 percent on a year earlier, with units down 5.7 percent. The HS 87 series itself ends in August until the by-customs table is released.
- L7 · BEHAVIOURALUntestedlag 0–7 days
Closure and queue at Yokohama Exporter's choice: wait for the next Yokohama sailing or truck to Nagoya or Tokyo
Nissan loads at its own Honmoku wharf and Isuzu builds at Fujisawa in the Yokohama hinterland; with Tokyo shut by the same rain and Nagoya a day away and another maker's gateway, the rational choice for a two-day closure is to wait. No series of truck movements or of Nagoya loadings of Kanto-built vehicles exists in the Atlas, so the choice is inferred, not seen.
- L8 · LOGISTICALUntestedlag 14–35 days
Missed or delayed car-carrier window at Daikoku Arrival delay for importers in Australia, the United States and the United Arab Emirates
Vehicles are 36.9 percent of Yokohama's export tonnage and Australia is the country the September flash names for the rise in automobile exports; a car carrier that waited 1.5 to 2 days arrives that much later, one that left without its Yokohama lot pushes the lot to the next sailing. No destination-side series is in the Atlas.
- L9 · ECONOMICContradictedlag 30–60 days
Closure share of the month Yokohama-specific shortfall in September vehicle units
Yokohama's September automobile units fell 5.7 percent on a year earlier against 0.5 percent nationally, a gap of 5.2 points. The same gap ran from minus 11.6 (May) to plus 12.7 (August) in 2026 months with no typhoon, the September fall sits in trucks (minus 14.0 percent) rather than passenger cars (minus 4.0), and September container export tonnage exceeded August's. A closure share of 6.25 percent of the month cannot be read from a series whose noise is four times that size; the link is kept as struck.
Preview: 5 of the 57 constants
Each constant carries the sentence it was read from and the date of that source. 52 more, and the 10 bounded constraints recomputed from them, are in the export.
| Id | Constant | Value | Kind | Source |
|---|---|---|---|---|
K1 | Duration of the Yokohama port closure for Typhoon Dujuan “The port was closed for 45 hours, resulting in yard density rising to around 90% and vessel waiting times reaching 1.5 to 2 days.” | 45 hours | capacity | source, 2026-09-23 |
K2 | Yard density at Yokohama container terminals after reopening “The port was closed for 45 hours, resulting in yard density rising to around 90% and vessel waiting times reaching 1.5 to 2 days.” | 90 percent of yard capacity | capacity | source, 2026-09-23 |
K3 | Vessel waiting time at Yokohama after reopening “The port was closed for 45 hours, resulting in yard density rising to around 90% and vessel waiting times reaching 1.5 to 2 days.” | 1.50 days | capacity | source, 2026-09-23 |
K4 | Seven-day average vessel waiting time at Yokohama, week of 23-29 September 2026 “The 7-day average vessel waiting time is around 1.2 days. Yokohama is recovering from Typhoon Dujuan-related disruptions and consecutive public holidays.” | 1.20 days | capacity | source, 2026-09-23 |
K5 | Gantry crane work-stop wind speed, Keihin port Kawasaki district “（ア）作業中止基準風速・・・・平均風速 10 m／s 風速が作業中止基準風速を超えるおそれがある、もしくは超えた、または暴風（雪）警報が発表された場合で、かつ、クレーン作業に危険が予想されるとき。” | 10 m/s, ten-minute mean | physical | source, 2020-04-01 |
Preview: what the author concludes
5 conjectures by statement. Their quantities, the checks run against the constraints, what would prove each wrong, and the 4 alternative explanations with their discriminators are in the export.
- J1 · LOCAL · confidence medium
The 45-hour closure was a shift of about four days, not a loss: on 22-25 September container export tonnage exceeded four days at the 108-day mean by about 1.7 times the shortfall of 19-21 September, so the tonnage that did not sail during the closure sailed within four days of reopening.
- J2 · CROSS SECTOR · confidence medium
If September scales like the eight earlier months of 2026, the September HS 87 export value at Yokohama customs lands between 208.0 and 221.2 billion yen: the port-level provisional automobile value of 173.1 billion times the ratio of HS 87 to flash automobile value, which ran from 1.202 in August to 1.278 in April. The closure's arithmetic share of the month, 6.25 percent or 12.5 billion off the January-August mean, is smaller than the spread of that mean (January 157.2, May 171.4 billion), so the value cannot resolve a loss of that size; a figure below 200 billion would point to the parts and motorcycle lines, not to the closure.
- J3 · CROSS GEOGRAPHY · confidence low
Importers of Yokohama-loaded vehicles in Australia, the United States and the United Arab Emirates received their late-September lots at least 3.4 days late, the closure plus the queue, and a full sailing cycle late where the car carrier sailed without them; Australia is the country the September flash names for the rise in Yokohama's automobile exports, so its October arrivals carry the delay.
- J4 · CROSS SECTOR · confidence medium
The shift was paid by the carriers, not the exporters: a 3.4-day disruption stays inside the 5-day detention and 7-day demurrage free time of a carrier's Japan export tariff, holidays excluded, so an exporter's worst case is the detention charge on a box whose free time was already spent when the port closed, 13,500 yen per 40-foot box, while the carrier bears 1.5 to 2 days of ship time per call.
- J5 · CROSS SECTOR · confidence medium
The customs unit series cannot resolve the closure: Yokohama's September unit gap to the national trend, minus 5.2 points, sits inside the 2026 dispersion of minus 11.6 to plus 12.7 points recorded without a typhoon, the fall is concentrated in trucks (minus 14.0 percent) against passenger cars (minus 4.0), and the August gap of plus 12.7 is a base effect. The closure's 6.25 percent share of the month is a quarter of that dispersion, so the October flash will land inside the same band whether the closure's lot shifted into October or sailed in September.
T1 · 2026-11-01 → 2026-11-30
November 2026 at Yokohama: a repeat of the closure signature, or an undisturbed month
- undisturbed-november: Number of UTC days in November 2026 with Yokohama port calls below 29, the November 2025 daily minimum (PortWatch)
< 1by 2026-11-30 - disturbed-november: Number of UTC days in November 2026 with Yokohama port calls below 29, the November 2025 daily minimum (PortWatch)
>= 1by 2026-11-30
Paths by period and implications by organisation: in the export.
Machine access: /api/atlas/typhoon-dujuan-yokohama-port-2026 carries this preview, and the complete dossier with a member key. What is for sale and why →