Amazon Comprehend (comprehend) awsJson1_1 service for fakecloud.
The full 85-operation Amazon Comprehend Smithy model: synchronous
single-document detection (DetectEntities / DetectSentiment /
DetectSyntax / DetectKeyPhrases / DetectDominantLanguage /
DetectPiiEntities / DetectTargetedSentiment / ContainsPiiEntities /
DetectToxicContent / ClassifyDocument), their BatchDetect* list
variants, nine families of asynchronous analysis jobs
(dominant-language / entities / key-phrases / sentiment /
targeted-sentiment / PII / events / topics / document-classification), custom
document classifiers and entity recognizers (with versioning), real-time
endpoints, flywheels and flywheel iterations, datasets, resource policies,
ImportModel, and ARN-keyed resource tagging.
Requests carry X-Amz-Target: Comprehend_20171127.<Operation>; dispatch
keys off req.action. Every operation runs model-driven input validation
first (required / length / range / enum / pattern), then real,
account-partitioned, persisted CRUD. Each resource is stored as its
already-output-valid wire JSON object so a Describe* echoes exactly what
its Create* / Start* persisted.
The asynchronous lifecycles are modelled by advancing the stored status on
the next read (and reconciled again on restart so an interrupted transition
never wedges): analysis jobs settle SUBMITTED -> COMPLETED, custom
classifiers and recognizers settle SUBMITTED -> TRAINED, endpoints settle
CREATING -> IN_SERVICE, flywheels settle CREATING -> ACTIVE, datasets
settle CREATING -> COMPLETED, and flywheel iterations settle TRAINING
-> COMPLETED. A Stop* / StopTraining* on an in-flight resource settles
it to STOPPED.
Honest NLP gap: Comprehend's value is the natural-language model that turns
text into entities, sentiment, syntax, key phrases, PII spans, and language
probabilities. fakecloud does not run any NLP inference. The synchronous
detection operations return well-formed, structurally-correct result shapes
with empty analysis lists (no fabricated entities / key phrases / syntax /
PII spans / languages); DetectSentiment returns the model's neutral default
(NEUTRAL, SentimentScore weighted to neutral) rather than an invented
judgement. Async jobs, classifiers, recognizers, endpoints, flywheels,
datasets, policies, tags, status lifecycle, and persistence are all real.