Availability: what we commit to, and what we do not yet measure.
This page publishes the uptime target and its formula, how downtime is defined and measured, every exclusion, the credit schedule and the claim window — and, just as plainly, the parts that are not in force yet and the reason no historical uptime figure appears anywhere on it.
Your use of AtlasVector is governed by the Terms of Use, which provide the Service as is, do not warrant that it will be uninterrupted, and cap total liability at the greater of what you have paid us or US$100. Nothing on this page changes that. It becomes a commitment only when a signed agreement incorporates it — see negotiated terms below.
This is the one availability fact we can put in front of you today, and it is a small one: your browser pings the public API's liveness route every 30 seconds and reports what came back. It measures this tab, since you opened it. It keeps no history, it resets when you reload, and it says nothing about last month — so it is not a Monthly Uptime Percentage, and a percentage is deliberately not derived from it.
The last tile is an em dash because nothing measures it. It stays an em dash until the probe in the prerequisites table below exists and has run for a month.
| Term | What it says |
|---|---|
| Monthly Uptime Percentage target | 99.9% of each calendar month, measured in UTC. |
| What that allows | Roughly 43 minutes of downtime in a 30-day month — the exact allowance per month length is in the table below. |
| Covered Service | The public portal API and the authenticated platform API. Availability means the service accepted and answered requests — it is not a claim about the correctness of any figure those requests return. |
| Downtime | A whole clock minute in which the majority of probe requests to the Covered Service fail or exceed a 10-second timeout. Partial minutes are counted as full minutes against us. |
| Measurement | An external probe outside our own infrastructure, at least one request per minute from at least two regions, with the raw minute-by-minute series retained and available on request. |
| Incident notice | When an incident makes the Covered Service unusable: a first notice to the address on your account within 1 hour of us confirming it, an update at least every 4 hours until it is resolved, and a written post-incident review within 5 business days. |
| Remedy | Service credits, on the schedule below, are the sole and exclusive remedy for failing the target. |
| Claim window | A credit is requested in writing within 30 days of the end of the affected month, and is applied against a future invoice — never paid in cash. |
| Month length | Total minutes | Downtime allowed at 99.9% |
|---|---|---|
| 28 days | 40,320 min | 40m 19s |
| 30 days | 43,200 min | 43m 12s |
| 31 days | 44,640 min | 44m 38s |
99.9% is the target we are willing to be held to, not a measured result. Because no uptime series is retained, we cannot tell you whether the system met it last month, and we do not claim it did.
This is the whole list.
- Maintenance announced at least 48 hours ahead and confined to a published window, capped at 60 minutes in any month.
- Emergency maintenance to close a security defect, where waiting for the notice window would leave users exposed.
- Anything between your device and us that we do not run — your network, your ISP, your DNS resolver, your browser, or a corporate proxy.
- An outage at a third-party market-data source. The Service stays up and says the figure is unavailable rather than showing a stale or invented one; that state is not counted as downtime here.
- Events outside our reasonable control: a cloud region-wide failure, a network-transit failure between providers, war, natural disaster.
- Suspension or termination under the Terms of Use — a breach, or a risk of harm to the Service or other users.
- Any feature labelled preview or in the pipeline, and any use of the Service outside a plan that carries this commitment.
| Monthly Uptime Percentage | Credit |
|---|---|
| Below 99.9%, at or above 99.0% | 10% of that month’s fee |
| Below 99.0%, at or above 95.0% | 25% of that month’s fee |
| Below 95.0% | 50% of that month’s fee |
Read this schedule with the arithmetic in front of you: a list price is published on the pricing page, but checkout is not open, so the fee actually paid today is zero — and every tier's credit is a percentage of zero until billing begins.
The published Terms of Use are consumer terms: an as-is Service and a US$100 liability ceiling. They are not the terms an institution signs, and we do not pretend they are. A master agreement — with a negotiated liability cap, this service level attached and binding, a security schedule, a data-processing agreement and a defined exit and data-return procedure — is available on request. There is no procurement portal to route you through and no seat minimum to clear, because no commercial agreement of any kind exists yet. The route is one address: [email protected].
The gap between this page and a real SLA is four things. They are listed with the reason each is missing and with no target date attached, because none has been committed to.
| Needed | State | Why it is not there |
|---|---|---|
| A continuous availability probe with a retained history | Not in place | The API exposes a liveness route that answers one request at a time and stores nothing. With no retained series, no month has a Monthly Uptime Percentage — which is why this page publishes none. |
| A public status page and incident history | Partly in place | A status board ships at /status and reports which components answer right now, probed from your own browser. What it cannot yet show is history: with no retained series there is no incident log and no monthly figure — the board says what is up this second, never what last month looked like. |
| A paid plan for a credit to apply against | Not in place | A list price is published on the pricing page, but checkout is not open and AtlasVector collects no payment today. A percentage of a fee that is zero is zero — the credit schedule above is real terms with nothing yet to attach to. |
| A signed agreement that incorporates this commitment | Not in place | Until one exists, the Terms of Use are the whole contract, and they warrant nothing about uptime. |
The other documents a review asks for are accounted for where each one lives. The sub-processor register and the data-processing agreement are described in section 4 of the privacy policy and sent on request; feed-by-feed provenance and freshness is on data sources. There is no SOC 2 attestation and no third-party penetration-test report to send — the compliance program is listed in the pipeline on the platform page. Anything else a review needs goes to [email protected].