What status code should a custom 404 return?

404, not 200. A "soft 404" — a not-found page served with a success status — teaches search engines that the URL is real and keeps it in the index indefinitely.

What belongs on a 404?

A short explanation, a search box, and links to the main sections. Not a sitemap dump, and not only a "go home" button — the person was looking for something specific.

Are these 404 and error pages free to use?

All 8 are free to copy and use in your own projects, including commercial ones. A Pro licence exists for redistribution — shipping them inside a product other people buy — and the full terms are on the licence page.

How do I add one to my project?

Install it with the CLI: `npx hoverlab add error-404-page` writes the component and its dependencies into your project. Every artifact here also has a copy button for the raw source, and an "Open in sandbox" link if you would rather try it first.

404 Page

Drops straight into app/not-found.tsx, and keeps the site header and footer — on a 404 the navigation is the most useful thing on screen.

System Pages1 blocks62 lines

We could not load this

The request failed before it finished. Your data is safe — nothing was changed.

Referenceerr_7f3a92c4b81e

Technical details

GET /api/v1/projects → 503 Service Unavailable
  upstream: projects-api
  duration: 30012ms (timeout)

3 changes are saved on this device

You are offline. Keep working — edits are written locally and sent automatically the moment the connection is back. Nothing is lost by closing this tab, though it will not send until you open it again.

Trying again in 8s · attempt 2

Not available until you reconnect

Searching across the workspace
Inviting people
Publishing or sharing a link

If this keeps happening, the status page has more than we can put on an error screen — including whether we already know.

Server Error

Retry first, because a 500 is our failure and offering navigation implies the rest of the site works — which is exactly what is in doubt.

System Pages2 blocks44 lines

Scheduled maintenance in progress

We are migrating the primary database. Dashboards, the CLI and sign-in are unavailable until this finishes.

Expected back in

02:14:09

About 2 hours and 14 minutes remaining. Estimated finish at 4:30 AM UTC

Still working

Published sites stay online and serving
The public API continues to read from cache
Existing sessions are not signed out

Live status

Hosted separately, so it stays up when this does not.

Scheduled Maintenance

Planned downtime with an actual end time on it — "back soon" is not a time, and a maintenance page without one looks like an outage nobody has noticed.

System Pages2 blocks35 lines

You do not have access to this

The page exists and you are signed in — your account just is not on the list for it. Nothing has gone wrong, and nothing you do here will change that except asking.

Resource: northwind / billing-service
Signed in as: you@company.com

Who can let you in

Priya RamanWorkspace admin
Tom OkaforProject owner

Request access

Back to where you were

If access was granted while this page was open, reload to pick it up.

Who can grant this

Ask one of them rather than opening a ticket — this is a role change, and they can make it without us.

Members

4 of 10 seats used

Ada Lovelace
ada@acme.com
Role for Ada Lovelace
Marco Silva
marco@acme.com
Role for Marco Silva
Priya Raman
priya@acme.com
Role for Priya Raman
alex@contractor.devInvite pending
Role for Alex Chen

Permission Denied

A 403 that names who can grant what was refused, because the next question is always "then who can" and a support link is the wrong answer to it.

System Pages2 blocks42 lines

1480 changes are saved on this device

You are offline. Keep working — edits are written locally and sent automatically the moment the connection is back. Nothing is lost by closing this tab, though it will not send until you open it again.

Trying again in 8s · attempt 2

Not available until you reconnect

Quartzly push
Halyard payment runs

Scheduled maintenance this Sunday

Database failover rehearsal in eu-west. The API stays up; exports and scheduled reports are paused for the window.

This is taking longer than planned. We have passed the 4:00 AM UTC estimate. The status page has the current position and is hosted separately from the systems under maintenance.

Still working

REST API reads and writes
Web application
Inbound webhooks

Live status

Hosted separately, so it stays up when this does not.

Knowledge freshness

Answers can only be as current as the source they come from. The oldest connected source was last read degraded for 22 minutes, so anything changed since then may not be reflected yet.

REST API
Public endpoint
Up to date · last read checked 40 seconds ago
0 docs
Web application
Interface
Up to date · last read checked 40 seconds ago
0 docs
Integration sync
Background worker
Partly indexed · last read degraded for 22 minutes
0 docs
Exports and reports
Background worker
Up to date · last read checked 1 minute ago
0 docs
Outbound webhooks
Delivery
Up to date · last read checked 40 seconds ago
0 docs

Ninety days of uptime

1,453 incident-free days across 306 active days in the last year

LessMore

Ninety days of uptime by month
Month	incident-free days
Sep 2025	72
Oct 2025	90
Nov 2025	87
Dec 2025	109
Jan 2026	111
Feb 2026	69
Mar 2026	132
Apr 2026	146
May 2026	148
Jun 2026	157
Jul 2026	170
Aug 2026	162

Incident history

Every incident with customer impact, written up within two working days. Post-mortems are public by default and redacted only where a customer is named.

Integration sync backlog
DegradedJan 9, 2026
- changed11:42 UTC — Quartzly push latency above threshold; queue building.
- fixed11:58 UTC — Cause identified as a slow query after an index change. Rolled back.
- changed12:04 UTC — Backlog draining. No data lost; every push carries an idempotency key.
Exports delayed for 71 minutes
ResolvedDec 18, 2025
- fixedA single tenant’s 4.1M-row export starved the shared worker pool.
- addedPer-tenant concurrency cap added the same week. This cannot recur in the same form.
Partial API outage, eu-west, 9 minutes
ResolvedNov 2, 2025
- fixedCertificate rotation applied to one load balancer ahead of its pair.
- changedRotation is now staged behind a health check rather than by schedule.

Take 1 of 2

Status Page

Written for the person refreshing it during an outage: current state above the fold, the maintenance window as a real block with an end time, and a component list that names what still works.

System Pages6 blocks155 lines

Acme

Sign in Contact

Docs navigation

Terms of Service

Last updated 6 January 2026. Changes that reduce your rights are announced 30 days before they take effect, by email, to the billing contact on the account.

Acme issues two kinds of keys. Secret keys (prefixed sk_live_) can read and write everything in your project and must only ever live on a server. Publishable keys (prefixed pk_live_) are safe to ship in a browser bundle and can only create client sessions.

Keys are scoped to a project, not to your account. Deleting a project revokes its keys immediately; removing a teammate does not, so treat offboarding as a reason to rotate.

Creating an API key

Create a key from the dashboard under Settings → API keys, then pass it as a bearer token on every request:

curl https://api.acme.dev/v1/projects \
  -H "Authorization: Bearer $ACME_API_KEY"

Note: the full secret is shown once, at creation. Acme stores only a hash, so a lost key cannot be recovered — it can only be rotated.

Rotating keys

Rotation is overlap, not replacement: create the new key, deploy it everywhere, and only then revoke the old one. Both keys stay valid during the window, so a slow rollout never turns into an outage.

Take 1 of 2

Legal Document

Terms, privacy, DPA and cookies in one frame — the docs layout, because people arrive at a legal page looking for a specific clause and a wall of text makes that a Ctrl-F expedition.

System Pages3 blocks50 lines

Acme

Sign in Subscribe

Ninety days of record

Everything is operational right now, which is the least useful thing this page can tell you. If you are reading it during diligence rather than during an outage, the numbers below are the ones that matter — including the November incident, which was ours and which was bad.

99.94%

Uptime, trailing 90 days

SLA commitment is 99.9%

4

Incidents, trailing 90 days

One of them was serious

41 min

Longest single outage

14 November, sync workers

4 of 4

Post-mortems published

Within 5 business days each

API response time, seven months

p50 and p99 on the reconciliation endpoint. The November spike is the sync stall; the p99 line is the honest one and it is why that month is listed as an incident rather than as degraded performance.

p50 (ms)
p99 (ms)

3000 ms
1000 ms
-1000 ms

Jun
Jul
Aug
Sep
Oct
Nov
Dec

API response time, seven months
Month	p50 (ms)	p99 (ms)
Jun	82 ms	410 ms
Jul	79 ms	395 ms
Aug	84 ms	428 ms
Sep	81 ms	402 ms
Oct	88 ms	447 ms
Nov	143 ms	1980 ms
Dec	86 ms	436 ms

Every incident in the window

Four, all of them, newest first — including the upstream one that was not our fault and the maintenance window that overran, because a history containing only green is read as a history that gets edited.

Raw event log

This month

Platform deployed ledger-worker v4.22.1
9 Jan, 09:14 UTC
Platform completed maintenance on EU read replicas
4 Jan, 02:00 UTC

December

Platform deployed retry policy with jitter — the Nov 14 fix
11 Dec, 10:02 UTC
Platform published post-mortem for 14 Nov sync stall
1 Dec, 16:30 UTC

Take 2 of 2

Status Page 02

The status page read during diligence rather than during an outage: ninety days of uptime, the latency curve, and every incident carrying its post-mortem — including the bad one.

System Pages7 blocks237 lines

Acme

Sign in Download as PDF

Legal — Terms of Service

Terms of Service

Version 4.2, effective 1 February 2026. Five clauses below are shown twice: the operative language, and a plain sentence saying what it does. The plain column is a summary and is not the contract — where they differ, the quoted text is what binds.

Acme B.V.

Amsterdam, Netherlands · Invalid Date · 9 min read

The pull request was titled “remove stale flags, part 1 of 8” and it deleted more code than the previous quarter had added. Our flag service listed 312 active flags; production telemetry showed 41 of them had evaluated to the same value for every user since at least January. The rest were split roughly evenly between experiments nobody remembered starting and kill switches for services we had already decommissioned.

“A feature flag is a loan. The interest is every engineer who has to reason about both branches forever.”

The five clauses people actually ask about

Out of thirty-one. The full document continues below these; the liability one is included because it is the clause least in your favour and leaving it out would make this section marketing.

Clause 3 — Your data

You own everything you put in, and we do not train on it

"Customer Data remains the exclusive property of Customer. Provider shall process Customer Data solely to provide the Services, and shall not use Customer Data to train, fine-tune or evaluate any machine learning model, whether or not such use is anonymised or aggregated."

In plain terms: your ledger is yours. We read it to reconcile it.
We do not train models on it, including anonymised.
We do not sell aggregate views of it to anyone.

Clause 7 — Ending the contract

Ninety days to export, and it is a button, not a request

"Upon termination for any reason, Provider shall maintain Customer Data in an exportable state for ninety (90) days, accessible to Customer via self-service export. Provider shall not condition such export on settlement of outstanding fees."

In plain terms: you can leave and take everything with you.
Ninety days after the last billing date, including if we terminated you.
Unpaid invoices do not lock your export. That clause is deliberate.

Clause 11 — Sub-processors

Thirty days notice before anyone new can touch your data

"Provider shall publish a current list of Sub-processors and shall give Customer not less than thirty (30) days written notice prior to authorising any new Sub-processor. Customer may object on reasonable data-protection grounds, in which case either party may terminate the affected Services without penalty."

In plain terms: you get a month’s warning and a right to object.
If you object and we proceed, you can leave without a penalty.
The current list is nine names and it is published, not on request.

Clause 14 — Liability

Capped at twelve months of fees, and this one is not in your favour

"Provider’s aggregate liability arising out of or related to this Agreement shall not exceed the total fees paid by Customer in the twelve (12) months preceding the event giving rise to liability."

In plain terms: if we cause a loss, our exposure is one year of what you paid us.
This is standard and it is still worse for you than for us. We are not going to pretend otherwise.
It is negotiable on annual enterprise contracts, and only there.

Clause 18 — Changes to these terms

Thirty days notice, and you can refuse by leaving

"Provider may amend these Terms upon thirty (30) days notice. Continued use following the effective date constitutes acceptance. Where an amendment materially reduces Customer rights, Customer may terminate without penalty within sixty (60) days of the effective date."

In plain terms: we email you a month before anything changes.
If a change takes something away from you, you get sixty days to leave.
Every past version stays published. Nothing is silently replaced.

What people email to ask

Six questions the terms answer but not in words anyone searches for.

Can your staff read our ledger data?

Only with a support ticket referencing it, only for as long as the ticket is open, and every access is logged and exportable by you. Four engineers hold that access; it is not the whole company.

Where is the data physically stored?

Frankfurt or Northern Virginia, chosen when the workspace is created. It cannot be moved afterwards, and that constraint is in Clause 9 rather than buried in a support article.

What happens to our data if you are acquired?

Clause 21 requires the successor to be bound by these terms as written, and gives you sixty days to terminate without penalty on a change of control. It does not, and cannot, promise we will never be acquired.

Do you respond to law-enforcement requests?

Where legally required. We notify you first unless a court order forbids it, we publish a transparency count annually, and that count has been zero for three years running.

Is the DPA pre-signed?

Yes, and it is incorporated by reference rather than being a separate signature step. The standard contractual clauses are attached as Annex 2.

What counts as a "material" reduction in rights under Clause 18?

Anything touching your data ownership, the export right, the sub-processor notice period, or the liability cap. We have listed those four explicitly rather than leaving "material" to be argued about later.

Take 2 of 2

Legal Document 02

One document with a plain-language column beside each clause, for the reader who cannot name the clause they need — and the cookie control inline, because a banner is unreachable the moment it has been dismissed.

System Pages6 blocks200 lines

404 and error pages

Pages8

Questions about 404 and error pages

Related collections