Skip to content

CLOUD EXIT · OPEN STANDARD · PUBLIC EVIDENCE

Cloud exit is a capability, not an event.

Keep the ability to move your data, workloads, and operations before you need it. Measure that ability with tests, prices, records, and drills.

Version 0.1 defines the method. No provider has been rated yet.

Read the standard Check the registry

The argument

Five questions, in the order a decision needs them

These are not five independent content buckets. Each answer earns the right to ask the next question.

01 / STANDARD

Who is charging an exit toll?

◇ Judge behavior, not company identity

  • Five public tests cover egress, export, interfaces, commitments, and downtime.
  • Every result must show evidence, review date, scope, and uncertainty.
  • Vendors can challenge facts through a documented appeal process.

A disagreement becomes a testable claim instead of a tribal argument.

02 / LEDGER

How much rent is in the bill?

◇ Premium is arithmetic, not rhetoric

  • At 200 GiB, gp3 is cheaper at baseline—but costs 2.16× AutoPL when 11.8K IOPS and 220 MiB/s are matched.
  • Owned compute crosses modeled PAYG material cost around 14–24 stable vCPU; deep five-year terms move the durable crossing into the thousands.
  • Against an OSS one-year package, reliable-topology object crossings move from 199 TB single-site to 1,223 TB three-site.

Every headline now links to the dated table, assumptions, full report, and checksum.

ledger.snapshot
# 2026-08-22 public-price snapshot
EBS/ESSD 0.54× baseline → 2.16× matched
S3/OSS   ¥0.156 / ¥0.120 per GB-month
✓ reports + CSV + sources + SHA-256

03 / EVIDENCE

What risk is already on record?

◇ Concentration risk is a record before it is a forecast

  • A dated outage chronology separates primary sources from commentary.
  • A lock-in catalog names the mechanism and the escape test.
  • A claims ledger compares public statements with observable results.

Risk discussions start from records that can be checked and corrected.

04 / CASES

Has anyone actually done it?

◇ Success and failure get the same template

  • What moved, what stayed, what it cost, and how long it took.
  • Where the migration failed or was abandoned.
  • What the team would do differently with hindsight.

Readers can recognize when a case applies to them—and when it does not.

05 / PLAYBOOK

What should we do on Monday?

◇ Start with one reversible drill

  • Restore a complete copy of critical data somewhere else.
  • Time the recovery and record every missing dependency.
  • Turn that number into a repeatable exit capability.

The first action is an afternoon exercise, not a migration program.

exit-drill.log
$ restore --somewhere-else
✓ data verified
✓ dependencies recorded
T+ your exit capability

The five tests

A standard people can inspect themselves

Each criterion page follows the same contract: definition → verification → worked example → boundaries and objections.

Economics

C1 · No exit toll

Leaving must not trigger punitive transfer or extraction charges.

Inspect the test

Data

C2 · Whole-data export

A customer can export the complete usable dataset in bounded time.

Inspect the test

Portability

C3 · Standard interfaces

Core operations use documented, broadly implementable interfaces.

Inspect the test

Contract

C4 · No commitment lock

Discounts do not become a practical penalty for leaving.

Inspect the test

Reliability

C5 · Accountable downtime

Incidents, remedies, and responsibility remain auditable.

Inspect the test

Results

Provider registry

Published ratings appear only after evidence review and carry an expiry date.

Inspect the test

Scope

The line follows behavior, not logos

Selling infrastructure is a trade. Pricing the fact that a customer cannot leave is rent extraction.

In scope

Exit fees, proprietary export barriers, interface lock-in, coercive commitments, and unaccountable service failure.

Potential allies

Compute vendors, CDNs, colocation, managed services, and clouds that make portability real and measurable.

Choose an entry point

Five minutes, thirty minutes, or one afternoon

Start with the smallest action that produces a fact about your own position.

5 minutes

Check a provider

Apply C1–C5 and see which evidence is still missing.

Start here

30 minutes

Price a workload

Make units, terms, utilization, and exit charges explicit.

Start here

One afternoon

Run an exit drill

Restore critical data elsewhere, verify it, and record elapsed time.

Start here

Direct answers

Objections the project must answer in public

Is this an anti-cloud campaign?

No. Cloud services can be the right operating choice and can strengthen exit capability. The target is measurable lock-in behavior, not a provider category or company size.

Does every company need to leave the cloud?

No. The claim is narrower: every organization should know whether it can leave, how long it would take, and what would prevent it. Staying is a choice only when leaving remains possible.

What counts as evidence?

Official price pages and incident reports come first; contracts, reproducible measurements, and first-party operating accounts follow. Commentary is labeled as commentary, and uncertainty remains visible.

Can a provider challenge a result?

Yes. The method records the exact claim, source window, reviewer, and expiry date. Factual corrections change the record with a visible revision; pressure without contrary evidence does not.

Does the initiator have a commercial conflict of interest?

Yes. Ruohang Feng sells commercial support for Pigsty, a self-hosted PostgreSQL platform, and may benefit when organizations choose self-hosting. That conflict is disclosed because readers should weigh it. The defense is not a claim of neutrality; it is open methods, reproducible data, named sources, and appealable results.

Measure your exit capability once.Restore the data somewhere else. Time it. Record what breaks. That number is your starting point.

Run the exit drill