Consistently assigns ids/keys to A or B group for A/B testing. An id may be user, session, request, etc.
This implementation is simple to integrate and fast, avoiding delays and preventing users from randomly changing groups and polluting/skewing results.

Tenant/group ids/keys can be utilized for A/B selection in multi-tenant system where testing a subset of tenants is desired. Groups are also useful for monotonically increasing the size of population under test. Best practice is to initially start with a small subset to validate no significant negative effects and then increasing to full population.

This implementation does not need any remote request to select A/B. Remote requests will delay the response to the user and it might (will occasionally) fail. When failure, the user might (will 50% of the time) change groups which pollutes/skews the results unpredictably.

By convention, A should be old treatment, B should be the new treatment for simplicity and consistency. Not required though.

By default all users are included in the AB test and selected for B by (hash(id) % 2 ? 'B' : 'A')

"Data Trumps Intuition" "significant learning and return-on-investment (ROI) are seen when development teams listen to their customers, not to the Highest Paid Person’s Opinion (HiPPO)​", ACM video presentation (Flash) referencing Amazon and Microsoft controlled experiments - startling results that were very surprising.​ Political and economic environment can affect results, necessary to use A/B testing to determine causality. Amazon observed approximately a 1% drop in revenue per 100mS increase in page load times.

Node 10+

On 2019 HP x360 Spectre with Ubuntu 18.04.3 64 bit
Intel® Core™ i7-8565U CPU @ 1.80GHz × 8

~275K AB selections / second
~3.6 microseconds each selection

npm i -S ab-test-selector

Configuration options with defaults that can be passed to the constructor.

{
  enabled: true, // enable/disable A/B selection, false always returns A
  idBonly: [], // ALL or specific ids/keys to always get B
  groupBonly: [], // ALL or specific group ids/keys to always get B
  idExcludes: [], // specific ids/keys to exclude from A/B, get A only
  idIncludes: ['ALL'], // ALL or specific ids/keys to include in A/B
  groupExcludes: [], // specific group ids/keys to exclude from A/B, get A only
  groupIncludes: ['ALL'], // ALL or specific groups ids/keys to include in A/B
}

Ideally, expose/integrate configuration via environment variables following 12 Factor App or via central Service Discovery/configuration (etcd, consul, zookeeper, RDBMS, etc.) or REST endpoints to allow dynamically changing configuring while the system is running.

If statistical significance is achieved and B wins, set the idBonly or groupBonly to ['ALL'] to route all users to the B treatment until the implementation is updated to remove the A/B test selector.

Partial configuration will use defaults for unspecified properties.

There are getEnabled() and setEnabled(boolean) methods.
The getConfiguration() and setConfiguration(config) methods allow setting full or partial configuration dynamically while executing.

Listed above in order of evaluation precedence, first definitive resolution is returned.
The precedence is hard-coded, changing the order here does not change the precedence.

The arrays are converted to Sets internally for performance in case large lists are needed.

import ABTestSelector from 'ab-test-selector'

const ab = new ABTestSelector()

if (ab.getAB(userId) == 'B') {
  computeOrShowTreatmentB() // change this to call your application code
} else {
  computeOrShowTreatmentA() // change this to call your application code
}

Dynamically switch all users to B treatment

ab.setConfiguration({
  idBonly: ['ALL'], // either this line or the other line
  groupBonly: ['ALL'], // either this line or the other line
})

import ABTestSelector from 'ab-test-selector'

const ab = new ABTestSelector({
  idIncludes: [7,12]
})

if (ab.getAB(userId) == 'B') {
  computeOrShowTreatmentB() // change this to call your application code
} else {
  computeOrShowTreatmentA() // change this to call your application code
}

import ABTestSelector from 'ab-test-selector'

const ab = new ABTestSelector({
  groupExcludes: [3,5] // exclude some group or org IDs, can be any scalar, like string or UUID, etc.
})

if (ab.getAB(userId, groupId) == 'B') {
  computeOrShowTreatmentB() // change this to call your application code
} else {
  computeOrShowTreatmentA() // change this to call your application code
}

import ABTestSelector from 'ab-test-selector'

const ab = new ABTestSelector({
  groupIncludes: [3,5] // exclude some group or org IDs, can be any scalar, like string or UUID, etc.
})

if (ab.getAB(userId, groupId) == 'B') {
  computeOrShowTreatmentB() // change this to call your application code
} else {
  computeOrShowTreatmentA() // change this to call your application code
}

See the tests and implementation files.