ProjectExtraction

A ProjectExtraction records a single attempt to pull one kind of structured data out of a project’s source document.

Purpose

The intelligence report attached to a project is parsed by a set of agents, one per data kind (outcome, people, geography, governance and so on). Each run writes a row here capturing which kind was attempted, whether it succeeded, which model was used, how long it took and any error message. The rows form an audit trail behind the extraction status panel and its per-kind history.

Usage

Rows are written by the extraction job and read by the staff status panel.

• app/Jobs/RunProjectExtractionJob.php — creates the row, runs the agent, and marks it succeeded or failed.
• app/Filament/Resources/ProjectResource.php — renders the extraction status panel and history.
• ProjectExtraction::latestPerKind() — returns the most recent row per ExtractionKind for a project, one card per kind.
• scopeRecentHistory() — the last N rows for one project and kind, used by the history modal.

Relationships

Relation	Type	Related model
project	belongsTo	Project
intelligenceReport	belongsTo	Attachment

The inverse is Project::extractions().

Data shape

Table project_extractions. Primary key is a UUID (HasUuids).

Column	Type	Notes
id	uuid	Primary key
project_id	uuid	FK to projects
intelligence_report_id	uuid	Nullable, FK to attachments
kind	ExtractionKind	Enum cast, see values below
status	ExtractionStatus	pending, succeeded, failed
model_used	string	Nullable, AI model identifier
started_at	datetime	Nullable
finished_at	datetime	Nullable
duration_ms	integer	Nullable, run time in milliseconds
error_message	text	Nullable, set on failure
created_at, updated_at	datetime	Timestamps

ExtractionKind values: outcome, people, weighting_influences, career_mobility, groupings, taps_roadmap, governance, geography, industry_trends. Each case maps to an agent class and an importer service in app/Enums/ExtractionKind.php.