Keep the facts that distinguish the work.
The problem
A personal graph can fill up with facts a reference book already knows. Those records make specific decisions and durable context harder to find.
An explanatory exhibit
Which fact belongs?
Two invented statements show the relevance criterion. The selection below is an explanation, not an automated model decision.
- 01
Extract
A conversation contains both statements
- 02
Review
Does this explain the project?
- 03
Retain
Keep a useful, supported connection
Inspect the consequence
“SQLite is a database”
Generic background
This adds little project-specific information to a personal graph.
“Lantern chose SQLite for its offline prototype”
A project decision
The statement distinguishes the prototype and can be checked against its dated source.
What this makes possible
I revised extraction and pruning criteria after generic facts began competing with useful personal and project context.
Source information sheet
Evidence, implementation & history Historical delivery checked
What I checked
The archive records a March extraction correction. Generated facts still require review.
Research cut: 14 October 2026. Source areas below are from panel-forge. Private conversation references are retained in the research ledger and are not published here.
Implementation decisions
I refined the analysis and review instructions around relevance: a project choice, a preference or a dated event can explain this archive in a way that a generic fact cannot.
The pruning service can classify records for retention or correction. This was a design correction in the extraction process; it does not make generated knowledge error-free.
A path through the feature
- Ask what the fact adds
Look for a connection to the person, project or event represented in the archive.
- Review the classification
Check the source before retaining or correcting a relationship. A general statement may belong in a conversation without belonging in durable memory.
Source areas
app/Services/ConversationAnalysisService.phpapp/Services/KnowledgeTriplePruningService.phpapp/Jobs/AnalyzeConversationJob.php
The small explanatory exhibit uses invented records to describe the mechanism. Separate recordings and working examples carry their own evidence notes.
Recorded history
Dates describe the researched record, rather than today’s deployment.
Refining graph extraction and pruning criteria