Search⌘ K
AI Features

The Execution Phase: Data Models, APIs and Where an Algorithm

Design the execution phase of Palantir's FDSE decomposition round by turning a user action into a plan another engineer could build. Define row grains, stable keys and relationships, assign ownership of each fact, and specify an API contract covering authorization, validation, version conflicts and idempotent retries. Separate synchronous from asynchronous work, stop stale imports from overwriting local state, enforce server-side output filtering, and recognize when a simple rule beats an algorithm.

A service coordinator clicks "Assign", but a diagram labeled “UI, API, database” cannot tell another engineer what happens next. An executable plan must specify who may act, which facts change together, what survives a retry and what the caller receives. Work through one depot’s equipment-service queue to make those decisions concrete enough to defend in an interview.

What must happen when the coordinator clicks assign?

The agreed scope is one depot, registry refreshes every fifteen minutes, coordinator-controlled assignment and a contractor’s assigned-work view. Predictive maintenance and edits back to the registry are excluded.

Third-party decomposition guidance describes execution as high-level design involving data and APIs, placing an algorithm or model where available inputs support it. This is preparation guidance, not an official scoring rubric or guaranteed interview format.

An executable design connects one user operation to owned facts, validated state changes and an authorized response. Begin with that operation before adding infrastructure.

Which rows represent A17, R44 and T91?

Use three records, applying the grain and key reasoning established earlier:

  • Asset(asset_id, serial, location, active, registry_version): One canonical equipment asset; asset_id is the primary key.

  • SourceLink(source_system, source_record_id, asset_id): One mapping from a source record to a canonical asset. The pair (source_system, source_record_id) is unique; asset_id references Asset.

  • ServiceTask(task_id, asset_id, due_at, state, assigned_team_id, task_version): One service task; task_id is the primary key and asset_id references Asset. Many tasks may reference one asset.

Registry record R44 maps to A17: Serial SN-88, location Bay-2, active=true, registry version 11. R45 maps to inactive asset A18. The initial tasks are:

  • T90 on A18, due 2026-10-29T08:00:00Z.

  • T91 on A17, due 2026-10-29T09:00:00Z.

  • T92 on A17, due 2026-10-29T10:00:00Z.

All three tasks are open, unassigned and at task version 3. A NULL assigned_team_id means unassigned, not team zero. Require the other listed task fields and imported asset fields; missing required values become import issues.

T91 and T92 represent different work on A17, so they do not create duplicate assets. Conversely, a new source record carrying SN-88 is only a match candidate: A source identifier identifies a source record, and matching it to equipment requires a justified, reviewable mapping.

What does the T91 assignment request promise?

One ...