durable operation2–3 hr5 stages8 code languages

Operating journey: Evidence-led claims from incident to settlement

Faster, more consistent claims decisions with tamper-aware spatial evidence.

The real-world moment

After a severe storm, a policyholder submits vehicle-damage images and location evidence. The insurer must guide capture, detect missing or contradictory context without overclaiming authenticity, run model-locked vision triage with human review, dispatch scarce assessors by eligibility and ETA, join hazard exposure, and retain explainable evidence through settlement and appeal.

Make ownership visible before code.

Operate the full lifecycle with versioned state, hostile-path recovery, audit, and release evidence.

Application state
stateful
Planning calls
0
Credentials accepted
No
Write authority
None
01
Claims system

Record consented incident location

Owns policy, claimant, incident declaration, coverage, reserves, decisions, settlement and appeal.

Claim ↔ declared incident place and policy coverage snapshot

GETReturns the details of provided model/models/{api_model_key}This is a supporting operation from the selected product stack, not proof of an exact provider contract for this stage. Resolve the stage through its product source or adapter boundary before live implementation.

Record stage 1 with the application identity, source fingerprint, request or aggregate identity, timestamp, and observable outcome.

Media upload partially succeeds

Use per-object hashes and a submission manifest; keep the claim incomplete until all required evidence commits.

02
Consented capture and vision boundary

Capture guided photo evidence

Collects guided media, device/location accuracy, timestamps, hashes, capture policy and disclosure; an entitled SkyDNN adapter locks model identity, validates bounded derived geometry and confidence, and never treats metadata or inference as proof by itself.

Record stage 2 with the application identity, source fingerprint, request or aggregate identity, timestamp, and observable outcome.

Vision result is uncertain or sensitive

Preserve model and input identities, route to an independent reviewer, and prohibit automatic settlement or fraud decisions.

03
Spatial enrichment

Run model-locked vision triage with human review bands

Normalizes incident place, joins versioned hazard/catastrophe layers and links every feature to its source and observation window.

Vision attempt ↔ model fingerprint, asset hash, bounded detections, confidence, review policy

GETReturns the details of all the models/modelsThis is a supporting operation from the selected product stack, not proof of an exact provider contract for this stage. Resolve the stage through its product source or adapter boundary before live implementation.

Record stage 3 with the application identity, source fingerprint, request or aggregate identity, timestamp, and observable outcome.

Location metadata conflicts with declaration

Preserve both sources, lower automation confidence and route to review rather than silently choosing one.

04
Assessment operations

Dispatch an assessor when needed

Routes incomplete, uncertain, sensitive, or high-risk claims to independent review and capable assessors, ranks eligible teams by network ETA and tracks field-task proof.

Record stage 4 with the application identity, source fingerprint, request or aggregate identity, timestamp, and observable outcome.

Catastrophe data is revised

Publish a new dataset/version and derived claim revision without rewriting the evidence used in the original decision.

05
Decision evidence

Overlay hazard and policy context

Policy engine and reviewer actions name exact submissions, enrichments, model fingerprints, rule versions and reasons; later evidence creates a revision and retention policy redacts derived detections without erasing audit.

Assessment decision ↔ evidence IDs, rule/model version, reviewer, settlement revision

GETGet users attendance details with travelled distance./reports/attendancesThis is a supporting operation from the selected product stack, not proof of an exact provider contract for this stage. Resolve the stage through its product source or adapter boundary before live implementation.

Record stage 5 with the application identity, source fingerprint, request or aggregate identity, timestamp, and observable outcome.

Media upload partially succeeds

Use per-object hashes and a submission manifest; keep the claim incomplete until all required evidence commits.

Edit once. Generate eight trusted-runtime implementations.

The recipe starts from the normalized GET /models/{api_model_key} contract. Execute six deterministic fixture outcomes before an issued account is involved.

Local request composerCode stays local · fixtures are explicit
GETRequest inputs
1 editable · 0 protected
Generated in eight languages
Environment-backed credentials
GEThttps://ai.mappls.com/models/vidteq-ods-9
Credential-free sandboxExecute this contract without a provider call
How fixtures work
Generated from your inputs
curl --request GET \
  --url 'https://ai.mappls.com/models/vidteq-ods-9' \
  --header "Accept: application/json" \
  --header "Authorization: Bearer ${MAPPLS_AI_TOKEN}"

Credential-safe by constructionSet MAPPLS_AI_TOKEN only in your trusted runtime. Code generation stays local. Sandbox execution sends only bounded fixture inputs to this developer platform and never reads a credential or calls Mappls.

Break the journey before customers do.

These controls are required at this operating depth; deeper recipes add durable truth rather than merely more API calls.

Idempotent commandsOptimistic aggregate versionsEvent and receipt timeTransactional outboxUnknown-outcome reconciliationIndependent human decisionsRestart recoveryImmutable audit evidence

Definition of done

Every aggregate has an invariant, expected version, stable command identity, and attributable actor.

Duplicate, late, reordered, invalid, stale, timed-out, and restarted execution preserve one durable truth.

Unknown provider outcomes reconcile by original identity before replay or compensation.

Terminal decisions, human review, audit, outbox delivery, privacy, retention, and rollback remain independently evidenced.

Contracts, tutorials, and journeys stay linked.

Contract set

GETReturns the details of provided model/models/{api_model_key}GETTo get the list of tasks and their details/tasksGETRouting API/{REST_KEY}/{resources}/{profile}/{geopositions}GETReturns the details of all the models/modelsGETTo get the task details for the given task id./tasks/{taskId}GETRouting API with driving profile/{REST_KEY}/{resources}/driving/{geopositions}GETReturns the current models in port/server/whoamiPOSTTo create new Task/tasksGETRouting API with trucking profile/{REST_KEY}/{resources}/trucking/{geopositions}POSTReturns the prediction as response in form of JSON/predictGETGet users attendance details with travelled distance./reports/attendancesGETRouting API with biking profile/{REST_KEY}/{resources}/biking/{geopositions}

Durable state machines

statefulField-service task lifecycle8 states · 8 transitionsstatefulCoordinated incident response7 states · 8 transitionshybridRecoverable navigation session6 states · 6 transitionshybridGoverned spatial analysis7 states · 6 transitionshybridDurable multi-stop itinerary6 states · 9 transitionsstatefulGoverned spatial agent run7 states · 6 transitionshybridiOS direction planning and navigation handoff8 states · 8 transitionshybridiOS feedback report and review9 states · 9 transitionshybridConsent-bound location capture evidence8 states · 9 transitionsstatefulGoverned place contribution and publication7 states · 8 transitionsstatefulGoverned SkyDNN inference review8 states · 8 transitions

Grid Restoration Desk

Correlate an outage, govern independent switching approval, dispatch an assigned crew, preserve rework, and reconcile restoration from telemetry and customer-impact evidence.

Source-grounded, not authority-granting

This recipe composes source-derived contracts, application architecture, fixtures, and maintained learning evidence. It does not prove package availability, provider payload shape, account entitlement, regional behavior, production compatibility, or release approval.