Skip to content
aerandil.com
GitHub

Read Message triage

GET
/v1/object-types/hospitality.message_triage/objects/{id}
curl --request GET \
  --url https://api.aerandil.com/v1/object-types/hospitality.message_triage/objects/2489E9AD-2EE2-8E00-8EC9-32D5F69181C0 \
  --header 'Authorization: Bearer <token>'

Read an existing instance.

A typed triage decision issued by an agent (or, less commonly, a human operator logging a manual triage into the audit trail) against a single inbound guest_message. Mostly-immutable post-issue: the intent / sentiment / urgency / factors / rationale / enrichment_snapshot are frozen at write time so the audit trail is the rationale at decision time, not against today’s profile. The decision_* and chained_* fields mutate exactly once on the transition out of suggested (apply / reject / escalate / expire / supersede) — that single mutation is gated by the decision_status_transitions cross-link validator.

Authorizations

Section titled “Authorizations”

Parameters

Section titled “Parameters”

Path Parameters

Section titled “Path Parameters”
id
required
string format: uuid

Object id (UUID v7, server-generated at create time).

Query Parameters

Section titled “Query Parameters”
branch
string format: uuid

Read through an open branch’s copy-on-first-write overlay view instead of the canonical (main) plane. Does not compose with as_of (422). The branch must exist under the caller’s tenant.

Responses

Section titled “Responses”

200

Section titled “200”

Object found.

Media typeapplication/json
object
created_at
required
string format: date-time
id
required
string format: uuid
object_type
required
Allowed value: hospitality.message_triage
properties
required
object
chained_action_correlation_id
correlation_id of the chained action's ApplyActionResponse (set on `applied`)
string
conversation_ref
required
Denormalised conversation ref (for per-conversation triage queries)
string format: uuid
decided_at
When the triage transitioned out of suggested
string format: date-time
decided_by
`principal.stable_id()` of the deciding principal
string
decision_status
required
Lifecycle state
string
Allowed values: suggested applied rejected escalated expired superseded
enrichment_snapshot
Frozen context snapshot at issue time (§3.3.1; JSON; host populates)
escalation_reason
Reason recorded on escalation
string
Allowed values: low_confidence policy_block safety_keyword explicit_guest_request sensitive_topic ambiguity operator_override pci_disclosed phi_disclosed
expires_at
required
After this, decision_status moves to expired
string format: date-time
factors_considered
required
Signals the classifier conditioned on
Array<string>
Allowed values: message_text conversation_history guest_profile reservation_context recognition_history recovery_history loyalty_tier time_of_day language_detected attachment_present prior_triage_history safety_keywords other
id
required
Triage id
string
intent_alternatives
Message triage intent alternatives

Top-k alternative intents with scores (optional)

Array<object>
object
intent
required
string
Allowed values: booking.inquiry booking.modification booking.cancellation booking.confirmation_query complaint.cleanliness complaint.noise complaint.billing complaint.service complaint.food complaint.other service.amenities service.spa service.restaurant service.transport service.lost_and_found service.wake_up service.technical service.housekeeping service.in_room_dining service.other info.amenities info.directions info.payment_policy info.wifi info.hours info.other recognition.celebration safety.emergency marketing.feedback chitchat.general injection.suspected other.unclassified
score
required
string format: decimal
intent_confidence
required
Operator-facing confidence ordinal
string
Allowed values: low medium high
intent_confidence_score
Calibrated probability (0..=1) — set when the model emits one
string format: decimal
intent_primary
required
Leaf intent from MESSAGE_INTENTS (§3.3.2)
string
Allowed values: booking.inquiry booking.modification booking.cancellation booking.confirmation_query complaint.cleanliness complaint.noise complaint.billing complaint.service complaint.food complaint.other service.amenities service.spa service.restaurant service.transport service.lost_and_found service.wake_up service.technical service.housekeeping service.in_room_dining service.other info.amenities info.directions info.payment_policy info.wifi info.hours info.other recognition.celebration safety.emergency marketing.feedback chitchat.general injection.suspected other.unclassified
issued_at
required
When the agent emitted the triage
string format: date-time
message_ref
required
The inbound `guest_message` this triage classifies
string format: uuid
policy_outcomes
required
Compliance flags observed during triage
Array<string>
Allowed values: pan_redacted phi_redacted tcpa_opt_out_detected gdpr_erasure_referenced whatsapp_window_outside voice_consent_missing injection_attempt_detected safety_keyword_matched none
rationale_text
Human-readable explanation surfaced to operators
string
recommended_action_kind
required
Discriminator for the apply path
string
Allowed values: auto_respond chain_typed_action escalate_to_staff silent_acknowledge no_action
recommended_action_params
Proposed parameters for the chained action (JSON object; validated at apply time)
recommended_action_ref
API name of the chained action (when kind = chain_typed_action)
string
recommended_response_text
Drafted reply text (when kind = auto_respond)
string
rejection_reason
Free-text reason on reject (gold-standard decision-log pattern)
string
sentiment
required
Sentiment ordinal
string
Allowed values: negative neutral positive
sentiment_score
Calibrated sentiment score [-1.0, 1.0]
string format: decimal
urgency
required
Urgency classification
string
Allowed values: low normal high critical
provenance

Per-property source attribution and pins. An entry under properties means the property’s current value came from that source (on the source’s own clock, source_updated_at); an entry under pins is a human override protecting the property from sourced writes.

object
pins
object
key
additional properties
object
pinned_at
string format: date-time
pinned_by
object
reason
required
string
properties
object
key
additional properties
object
external_id
string
recorded_at
required
string format: date-time
source
required
string
source_updated_at
string format: date-time
stream
string
scope
required
One of: discriminator: level
object
level
required
string
Allowed value: tenant
Allowed values: tenant
tenant_id
required
string format: uuid
updated_at
required
string format: date-time
version
required
integer
>= 1
Example
{
  "object_type": "hospitality.message_triage",
  "properties": {
    "decision_status": "suggested",
    "escalation_reason": "low_confidence",
    "factors_considered": [
      "message_text"
    ],
    "intent_alternatives": [
      {
        "intent": "booking.inquiry"
      }
    ],
    "intent_confidence": "low",
    "intent_primary": "booking.inquiry",
    "policy_outcomes": [
      "pan_redacted"
    ],
    "recommended_action_kind": "auto_respond",
    "sentiment": "negative",
    "urgency": "low"
  },
  "scope": {
    "level": "tenant"
  }
}

Headers

Section titled “Headers”
ETag
string
/^"\d+"$/

Quoted strong validator (RFC 7232 §2.3) carrying the object’s monotonic version. Informational on the read path — object mutation goes through typed actions on /v1/actions/{action}/apply, not raw CRUD.

404

Section titled “404”

No object with that id (or soft-deleted).