Let's create a text-based imposter with multiple stubs. Binary imposters won't see any interesting behavior difference with only startsWith predicates:

POST /imposters HTTP/1.1
Host: localhost:<%= port %>
Accept: application/json
Content-Type: application/json

{
  "port": 4548,
  "protocol": "tcp",
  "mode": "text",
  "stubs": [
    {
      "responses": [{ "is": { "data": "first response" } }],
      "predicates": [{ "startsWith": { "data": "first" } }]
    },
    {
      "responses": [{ "is": { "data": "second response" } }],
      "predicates": [{ "startsWith": { "data": "second" } }]
    },
    {
      "responses": [{ "is": { "data": "third response" } }],
      "predicates": [{ "startsWith": { "data": "second" } }]
    }
  ]
}

The match is not case-sensitive:

echo 'FIRST REQUEST' | nc localhost 4548

first response

The same is true for the second stub.

echo 'Second Request' | nc localhost 4548

second response

The third stub will never run, since it matches the same requests as the second stub. mountebank always chooses the first stub that matches based on the order you add them to the stubs array when creating the imposter.