Short answer: Partly, and only as far as the record itself extends. Two public status pages under one vendor name were readable today, and each publishes a named component list, a day-by-day incident log, dated maintenance entries and a machine-readable feed [S3][S4][S5][S6][S7][S8]. Those make it possible to check which services the page claims to cover, whether an entry exists for a particular date, what states that entry passed through, and when a maintenance window was announced [S3][S6][S7][S10]. They do not make it possible to check the figure inside the uptime sentence: no record I opened carries a measured availability value, a measurement window, a measurement point or a stated list of what would be left out [S5][S6][S7][S8][S20]. So the defensible question to put to a public log is narrow — is my component on this page, and is there an entry for the day I care about — while the wider question, whether the advertised figure was met, stays outside what these records can answer [S1][S17][S18][S20]. This page quotes no availability figure, derives no ratio from any log, and does not revisit the clause wording handled elsewhere on this site.
There are two pages, and they carry different names. One is headed Services Status and lists an API, DNS, cloud dedicated, cloud VPS hosting, several control panels, object storage, load balancers, network segments, data centres, a management portal and support channels [S3]. The other is headed Cloud Sites Status and lists PHP services, IIS services, MariaDB, MSSQL, FTP and a management portal, each split across two locations [S4]. Neither list contains the other. That matters before any checking starts: a buyer has to find out which page owns the component they were sold, because a clean record on the wrong page says nothing about the right one [S3][S4][S8].
Both pages print a legend with five labels [S3][S4]. Operational is one of them; degraded performance is another; partial outage and major outage sit below it; maintenance is the fifth [S3][S4]. The platform's own documentation says an incident's impact derives from the affected components and that the headline status is computed from every component on that page [S14]. A component, in that documentation, stands for one distinct piece of infrastructure or one function of the service, chosen by whoever publishes the page [S11]. The headline is therefore an output of a declared component table, not a reading taken from outside the service [S11][S14]. I am describing how the page is assembled here, not making a statement about the service's condition.
Under the legend sits a day-by-day list. On the Services Status page most recent days carry no reported entry; one day carries a scheduled network maintenance with three successive states — scheduled, in progress, completed — each with a time stamp and a body of text, plus a maintenance reference identifier [S3][S7]. Two older days carry entries described as unresolved [S3][S6]. The Cloud Sites Status page shows the same daily shape with no reported entry on any displayed day, alongside a link labelled Incident History and links to an Atom or RSS feed [S4][S10]. A feed is useful to a buyer for one reason: it is a channel through which entries can be collected without revisiting the page by hand [S4][S16].
The machine-readable endpoints expose the same material as fields. A summary endpoint returns components, incidents, scheduled maintenances and a page-level status [S5]. An incident record carries a name, a status, an impact value, creation and resolution time stamps, the affected components and an array of updates, each update with its own status and time stamp [S6]. A scheduled-maintenance record adds planned start and planned end fields [S7]. A component record carries a name, a status, a group, a position and a showcase flag [S8]. None of these endpoints returns a field named uptime, availability, measurement window, measurement point, or any field listing what is left out [S5][S6][S7][S8][S9]. I searched the retrieved payloads for such a field and found none; that is an observation about what came back today, not a claim about every endpoint that has ever existed.
A promise with a number in it, a descriptive phrase, and a line in a shopping checklist are three different objects. Conflating them is how a buyer ends up feeling they have checked something when they have only read something.
On the editorial pages, a figure appears inside a sentence that attributes it to a host as a guarantee [S17][S18]. The same set of pages uses an uptime guarantee as one line among several in a list of criteria for choosing a host [S18]. It also carries an explanatory question-and-answer entry asking what such a guarantee is [S17][S19]. Those are a reported claim, a selection criterion and a definition. They are not three copies of one fact.
On the budget vendor's own pages the register changes. The homepage uses adjective-led phrasing about uptime without a figure, and the KVM plan page uses the same register [S1][S2]. A customer testimonial printed on those pages praises the uptime of the network [S1][S2]. No quantity is attached to those phrases, and I am not going to attach one. A phrase of that kind tells a reader how a vendor chooses to describe itself; it does not supply a value that a log could be held against [S1][S2][S20].
The second vendor's product, policy and support paths all returned HTTP 403 on the day these sources were collected [U1], so I cannot say how that vendor phrases its promise in a document. Its status pages were readable, which is a different surface entirely [S3][S4]. That split — documents closed, status pages open — is close to the whole problem in one sentence.
The IETF framework for IP performance metrics insists that a metric and its measurement methodology be defined before quantities are comparable, and it treats uncertainties, errors, sampling and clocks as part of that work [S20]. A separate pair of RFCs defines requirement-level keywords and later clarifies that only their upper-case forms carry the specified force [S21][S22]. I am using those documents as an analogy, and I am labelling it as one: they govern standards writing, not hosting contracts. The chain is short. If the force of a word depends on how explicitly it is framed, and if a quantity needs a stated method before it can be compared, then a guarantee carrying a figure, an adjective in a marketing sentence and a criterion in a shopping list cannot be treated as three versions of one checkable claim [S20][S21][S22]. I am calling that a construction of my own from documents that never address hosting, and I attribute no such view to their authors.
What can be checked is the publication itself. Whether a page exists [S3]. Which components are named on it [S3][S8]. Whether the component you bought is among them [S8][S11]. Whether a given date carries an entry, and if so what states that entry moved through [S6][S10]. Whether a maintenance window was announced ahead of time with a planned start and end, and whether its updates ended in a completed state [S7][S13]. Whether a postmortem was attached, which the platform permits only after an incident is resolved [S15]. Whether the entries can be collected through a feed or an endpoint so that a second person can retrieve the same ones later [S4][S5][S6][S16]. Each of these has a yes-or-no answer that does not depend on anyone's arithmetic.
What cannot be checked is the advertised figure. The records do not state where a measurement would be taken, how often, over which interval, from which vantage point, or what would be left out [S5][S6][S7][S8][S20]. Without those, a day with no reported entry is not a day of full availability; it is a day on which the page's owner published nothing [S3][S4][S12]. That distinction is the answer. I am stating it as an inference and giving the reasoning in full: the platform lets an operator add past incidents by entering a name, a description and a date [S12]; impact values and the headline state are computed from the operator's own component choices [S14]; therefore the log records what was published rather than independently measuring what happened [S12][S14][S20].
The platform documentation is unusually candid about this. It says backfilling earlier incidents stops a new page from looking empty, that a backfilled entry appears in the past-incidents section, and that the operator sets the date on which the event is said to have occurred [S12]. It describes scheduled maintenance as a way to tell viewers in advance about planned unavailability, with time, duration and affected components as required fields [S13]. It presents public, private and audience-specific page types as separate options [S16]. A buyer reading a public page is therefore reading a publication whose scope, component list and history were all chosen by the publisher [S11][S12][S13][S16].
So the honest comparison sheet has no arithmetic on it. It has a place for the sentence being checked, a place for the page that would have to carry the record, a place for the component name, and a place for dated entries. It has no column for a computed result [S5][S6][S7][S20].
Both pages sit under the same vendor name, and they do not agree. Their component lists cover different things, as described above [S3][S4][S8]. Their incident feeds also diverge on an entry referring to the same pair of published security-advisory identifiers: on one page that entry is still in a monitoring state with a major impact value and names several control panels and cloud products as affected, while on the other the corresponding entry is resolved with a none impact value and names the two locations instead [S5][S6][S9]. I am reporting a difference between two retrieved payloads, not a judgement about the underlying event, and I am not going to reconcile them.
That divergence is more useful than agreement would have been. It shows that a public record is not a single object. Two pages, two component tables, two histories, two feeds — and a buyer has to pick the right one before any comparison means anything [S3][S4][S9][S10]. It also shows why a status headline cannot stand in for a sentence in an advertisement, since that headline is computed from the page's own component set [S14].
The editorial pages add a third kind of record that is not a vendor log at all. One roundup lists tools a reader can use to watch a host after buying, and another reports figures from the site's own test installations [S17][S18]. Those are third-party observations with methods of their own, and they are not the vendor's incident history [S17][S18][S20]. Putting them in the same column as a vendor status page would collapse two different kinds of source. I am calling that a classification made here, not something any of those pages says.
If you came here for public uptime records, the what a cheap VPS does not include page shows what you have to buy separately before the server is usable.
Several paths that would have helped did not open. Every product, policy and support path tried on the second vendor's main site returned HTTP 403 [U1]. A status subdomain for the budget vendor could not be reached at all, and its FAQ, knowledgebase, status, network-status, uptime and service-level paths returned HTTP 404 [U2]. Three tool and buying-guide paths on the review site returned HTTP 404 [U3]. Several platform documentation paths returned HTTP 404, as did a public-status-page help path on a monitoring site [U4][U5]. A client-area host for the budget vendor could not be reached [U6]. I reproduce nothing from those responses, draw no conclusion from them about what they may contain, and let no earlier note stand in for them.
The second vendor's policy wording is therefore unavailable to this page, and I am not going to describe it. The part of this site that handles clause wording is a separate article; none of that text is revisited here, and the closing here concerns records only. No availability figure is quoted, no ratio is derived from any incident log, no frequency claim is made about how often records and promises diverge, and no ranking or analytics number appears. A word check on the three readable comparison pages returned no hit for status page, incident history, scheduled maintenance, maintenance window or postmortem [S17][S18][S19]; I report that as an observation about those retrieved pages, not as a claim about the web at large.
A public record can be checked. A promise can be read. Joining the two is where buyers get into trouble, because the record is a log of what a publisher chose to post and the promise is a sentence written for a different purpose [S3][S12][S14][S17][S20]. The useful outcome of the comparison is not a verdict on a number; it is a dated statement that a named page did or did not carry an entry for a named component [S3][S6][S8][S10]. That is a smaller result than most shopping pages imply, and nothing broader is supported by these sources.
The three ways of writing the same word make the gap wider rather than narrower. A guarantee carrying a figure, an adjective in a description and a criterion in a list each ask something different of a reader, and only the first even gestures at a quantity [S17][S18][S20][S21][S22]. The record contains no quantity at all [S5][S6][S7][S8][S9]. A buyer who keeps those two facts separate has something defensible; one who merges them has a number with no method behind it. On the lowest-priced tier the merge is hardest to notice, because the promise fits on a plan card while the record does not.
No. The readable endpoints return components, incidents, maintenance entries, states and time stamps; they return no measured availability value and no measurement definition [S5][S6][S7][S8][S20]. The record can show whether an entry exists for a date and which component it names [S3][S6][S10]. It cannot settle the figure.
They publish different component lists and separate incident histories [S3][S4][S8][S9]. The platform computes each page's headline status from that page's own components [S11][S14]. A buyer should identify which page names the component they were sold before reading anything else [S3][S8].
No. It means the page's owner published no entry for that day [S3][S4]. The platform permits past incidents to be backfilled with an operator-supplied date, so the log reflects what was posted [S12]. How often an event goes unpublished is unknown to me; stating a frequency would require a measurement method these records do not supply [S20].
All pages read on 2026-11-11 unless noted. HTTP status is recorded as returned on that day.
Fixed field names and interface labels are kept as printed. Every cross-document reading above is marked as an interpretation or an inference and carries its chain. No availability value is quoted, no ratio is derived from any incident log, and no analytics or ranking figure is used.