Revised 7 November 2026. Observations below are dated; this is not a live service-status report.

On 4 November, at 11:29 JST, FabCafe’s homepage returned HTTP 503: a maintenance page instead of the usual website. Its notice attributed the precautionary suspension to a critical PHP vulnerability. The same page said its physical locations remained open and events were continuing as scheduled.1

Calling FabCafe “down” would lose the most useful distinction. The website was unavailable at that observation; the operator said the cafés were operating. Neither statement dates the website’s outage onset or proves what happened between our checks.

A security response involves several kinds of work: understanding local exposure, correcting advice, helping people act and assigning what remains. Those records matter. Recovery asks a further question: which function can people use again, for whom, and since when?

Different functions, different recovery states

8 selected functions · October–November 2026 records, not live status

Service / functionDisruption / action User access at stated dateStill not established
FabCafeHomepageSource 1 Precautionary PHP website pause Observed unavailableHTTP 503 at the check Onset and reopening
FabCafePhysical venues / eventsSource 2 Website paused; venues separate Reported operatingOpen; events as scheduled Independent venue checks
FFXIVNew logins · Korean serviceSource 3 New logins restricted Reported restoredReturned after 25h 25m Existing-session availability
FFXIVSettings backup · Korean serviceSource 4 Backup stayed restricted Reported restored3d 15h 18m after login Whether any data was lost
MUFG NICOSCorporate purchasing-card statementsSource 5 Some WEB functions resumed Not establishedStatements not named Statement access / interruption
Kansai T&DWebsite / outage informationSource 6 Web functions paused; app offered Reported restoredSame-day website restoration Exact interval; not power supply
NTT DOCOMOMobile voice · some usersSource 7 Voice disruption from ~16:30 Reported restoredRestored in the stated areas Cause; SMS not established
BinanceCustomer withdrawalsSource 8 Withdrawals suspended Not establishedReturn to normal not timed Access ≠ security containment
Functions returned at different times, while others remained available or unresolved. Each row includes its date and evidence: eight selected functions, not eight incidents or a recovery success rate.

Useful information gives someone a next step

Debian’s xz advisory distinguished stable releases from affected development branches, making a broad warning locally useful.23 Apache explicitly explained why earlier Log4j mitigations were insufficient; its archived guidance retained class removal as a fallback, rather than discrediting every remedy.4 These are useful contributions, not measurements of how many users became protected.

Heartbleed supplies stronger action evidence. Researchers sent contacts lists of vulnerable hosts and recovery guidance. After eight days, 39.5% of early-notified contacts had at least one listed host patched, versus 26.8% of those not yet notified.5 That is a measured contact-level difference of 12.7 percentage points. It is neither a percentage of fully secured hosts nor a trial showing that assistance beats a general warning.

May First’s PHP notice shows another form of help: it reports patching its shared webservers, including older PHP, and links instructions while encouraging members to change versions.6 Its maintenance commitment is qualified by members’ needs and its capability.7 We can describe the help without claiming that every member followed it or that every site was recovered.

The website and the service can recover separately

We ran a bounded Japan pilot: twelve provider–function pairs across internet access, banking/payments, energy, flights, rail and messaging/comms, for 20 October–3 November. Official news came first; notices guided endpoint checks and archive lookups. Providers without usable evidence stayed in the sample.

Three units yielded scoped historical operator reports: DOCOMO voice services, TEPCO electricity supply in part of Chiba, and JR East train operations on some Chiba lines.8910 Nine lacked adequate evidence for the selected function. None of the twelve had an independently verified present function state. Unknown did not mean unavailable—or normal. The retained pilot notices did not explicitly establish a PHP link; this is not evidence that PHP caused no disruption.

The function boundary changes the reading. Mitsubishi UFJ NICOS reported returning its members-only WEB login page to its normal presentation from 26 October and resuming some functions. It did not individually name corporate purchasing-card statement retrieval.11 That notice cannot settle whether statements were available, still less whether card payments worked.

Kansai Transmission & Distribution reported suspending website functions, including part of its outage-information service, and restoring them in a same-date notice on 22 October.12 An app was offered as an alternative information channel. These are records of website disruption and recovery, not electricity-supply failure and restoration.

Even a “normal” label needs checking. Some inspected pages contained hidden or default status templates populated by JavaScript. Reading their unrendered HTML as a live all-clear would have created false normality in our results.

The first archived error is not the start of an outage

For a named endpoint, we look for the last observed success, first observed failure, last observed failure and first observed recovery. Those observations can bound transitions. They do not prove continuous downtime between captures, and a missing notice is not a successful service test.

Operator reports can provide better clocks. DOCOMO describes a scoped voice disruption on 28 October from around 16:30 to around 17:29 JST.8 That supports a reported interval of roughly an hour, not an independently measured 59-minute outage for every subscriber. Its separate Okinawa restoration notice warns that temporary recovery could still leave some areas with difficulty or slower data speeds.13

Archives helped recover JR East’s notices, but not a complete service history. Two sampled pages carried a weather-related train-suspension banner. A separate software-related digital-services notice appeared in the later sample and expressly excluded train-operation effects.10 The observations must not be joined into one PHP outage or one continuous duration.

Stopping a service does not settle containment

Binance’s initial account of its 29 October incident adds a separate example. It says suspending withdrawals and shutting hot-wallet services did not invalidate malicious authorisations already recorded on-chain. The team needed a separate procedure to revoke those permissions.14

That is Binance’s initial account, published on 3 November, not an independent forensic finding. It does not establish a PHP connection or a time when normal customer withdrawal service was restored. The operational distinction is important: a service can be suspended while a compromised capability remains usable. Containment, restoring customer access and establishing security require different evidence. Binance is an illustration separate from the Japan pilot.

If you run the systems, report the function you checked

Keep the notice you acted on, its version and the time. If advice changes, preserve what it replaced. A current page may no longer show the instructions someone followed earlier.

Find the person who controls each layer before the next warning. The build pipeline, the configuration file, the managed database and the customer’s application may sit with different people.

Route a practical obstacle to someone who can address it: a vendor, distributor, colleague or support community. State the boundary of that help rather than assuming a published procedure was used successfully.

Report the checkpoints separately: patch installed, permissions revoked, harmful activity contained, function restored, exceptions remaining. For restoration, name the function, affected users or region, observation time and evidence. A successful webpage load cannot substitute for a transaction or delivery check; neither alone proves that a compromised capability has been removed.

Distinguish a workaround from restoration of the original channel. Say when an outcome is operator-reported, directly observed or still unknown. Missing completion evidence is not proof that no work happened.

Ask what people can do again

The next useful sentence after “we responded” names something a person can do again. Can they retrieve the statement, reach the service or use the original channel? If we cannot yet answer, that is the next gap to investigate.

Evidence and scope

The Japan pilot used dated public notices and archival captures for twelve preselected critical provider–function pairs, two per sector, over 20 October–3 November 2026. Three yielded scoped historical reports; nine target functions remained unknown, and all twelve present states were independently unverified. This is a purposeful pilot, not a national sample. The figure’s selected illustrations, including FabCafe and Binance, are separate from the pilot roster.

Search coverage, access restrictions, client-rendered pages and incomplete archives limit what can be established. An archive error may belong to the archive rather than the original site. Event, notice, capture and retrieval times are distinct. No end-user transaction, message-delivery or physical-service tests were performed; the records make no current-availability claim.

Figure sources

Function records: sources and scope

  1. FabCafe · Homepage

    Endpoint observation ·

    One homepage request on 4 November 2026 returned HTTP 503. Its notice attributed the precautionary pause to a PHP vulnerability. This does not establish continuous downtime or the state of other pages. The operator’s statement that it found no related breach is not an independent security finding.

  2. FabCafe · Physical venues / events

    Operator report ·

    The notice on the homepage observed on 4 November 2026 said physical locations remained open and scheduled events would continue. These are operator statements, not visits or attendance checks. This is a second function of FabCafe, not a second incident.

  3. FFXIV · New logins · Korean service

    Operator report ·

    The retained operator clocks run from 22 October 2026 at 01:25 to 23 October at 02:50, with KST assumed: 25 hours 25 minutes. This concerns new logins, not all gameplay or the NAVER portal. Settings backup below belongs to the same response, with a later endpoint.

  4. FFXIV · Settings backup · Korean service

    Operator report ·

    The operator reported client and character settings backup restored on 26 October 2026 at 18:08 (KST assumed). The extra 3 days 15 hours 18 minutes runs from login restoration, not the start of the backup restriction. A conditional damaged-data warning does not establish data loss.

  5. MUFG NICOS · Corporate purchasing-card statements

    Operator report ·

    The 26 October 2026 notice reports normal login-page presentation and resumption of some WEB functions after changes from the night of 21 October. Corporate purchasing-card statement retrieval is not individually named. Neither its interruption nor its restoration is established; this is not a payment-outage finding. PHP is not named.

  6. Kansai T&D · Website / outage information

    Operator report ·

    Website functions, including part of outage information, were sequentially suspended from 22 October 2026 at 00:00. A same-date notice reported restoration without an end hour. An app was offered as a workaround. The notices concern a generic software vulnerability, not explicitly PHP; they do not report electricity-supply interruption or restoration.

  7. NTT DOCOMO · Mobile voice · some users

    Operator report ·

    The 28 October 2026 report gives an approximate 16:30–17:29 JST interval for some users in parts of Hokkaido, Tohoku, Kanto, Hokuriku and Tokai, including emergency-call effects. This is an operator report, not an independently measured 59 minutes or a nationwide all-user outage. Cause was under confirmation; no PHP connection is established. Other DOCOMO episodes are not combined here.

  8. Binance · Customer withdrawals

    Initial operator account ·

    In its initial account published on 3 November 2026 about 29 October, Binance says suspending withdrawals and shutting hot-wallet services did not invalidate permissions already on-chain; it reports a separate revocation procedure. That account does not establish when normal customer withdrawals resumed. It is not independent forensics, proof of security, or evidence of a PHP connection.

Sources

  1. FabCafe, homepage maintenance notice, observed 4 November 2026 at 11:29 JST with HTTP 503. The notice attributes website suspension to a PHP vulnerability and says venues/events continue. This endpoint observation does not establish onset, continuous duration, physical operations independently, or every locale’s website availability.
  2. xz backdoor, oss-security disclosure, 29 March 2024, naming build and distribution conditions.
  3. Debian, DSA-5649, stable not known affected; testing and unstable affected.
  4. Apache Log4j, archived security page, 20 December 2021 capture. Names earlier mitigations as insufficient and retains class removal as a fallback.
  5. Durumeric et al., The Matter of Heartbleed, IMC 2014, author-hosted PDF. Comparison is early-notified contacts against not-yet-notified contacts; contact-level outcome means at least one listed host patched, not a fully secured host.
  6. May First, PHP upgrade notice, 21 October 2026, reporting webserver work between 14:00 and 15:00 UTC. Encouragement to move sites forward, not a requirement; operator statement, not independent runtime verification. Member outcomes are not measured.
  7. May First, membership agreement, assigning shared equipment and software maintenance to the cooperative, as members’ needs and its capability allow, and site-specific software to members.
  8. NTT DOCOMO, 28 October 2026 voice-service restoration report. Approximate 16:30–17:29 JST interval, scoped possible impact across named eastern areas; emergency calls affected for some users. Cause under confirmation; no PHP attribution.
  9. TEPCO Power Grid, 27 October 2026 emergency notice. Twenty households in difficult-access areas of Chiba still without power at 12:00 after Typhoon 25. Snapshot, not a restoration clock; the September peak is outside the study window and is not onset.
  10. JR East homepage captures of 20 October and 29 October JST. Both carry the weather-related Chiba train banner. The separate software-related group-service notice is present in the later sample and says train operation is unaffected by that issue. Its earlier absence is not evidence of successful service.
  11. Mitsubishi UFJ NICOS, 26 October 2026 WEB-service resumption notice. Login-page presentation and some functions; corporate purchasing-card statement retrieval is not separately named. Generic software-vulnerability attribution, not an explicit PHP identification.
  12. Kansai Transmission & Distribution, 22 October 2026 website-function suspension and restoration notice. Sequential start from 00:00, no end hour; not electricity-supply interruption/restoration. Software is not named as PHP.
  13. NTT DOCOMO, Okinawa typhoon-related voice/data report, updated 27 October 2026. Reported restoration retains possible difficulty or reduced speeds under temporary recovery; not an unqualified all-clear or a confirmed continuing outage for every user.
  14. Binance Blog, initial incident account, 3 November 2026, also linked to the canonical blog. Operator account, not independent forensic validation or a current withdrawal-availability statement.