A file quality score is a computable measure of a file’s observable relational density within the repository. It answers the question: how many relations does this file participate in?

Grounding in quality

Quality is what becomes observable about a thing through its relations. A file’s quality is therefore not intrinsic — it emerges from the file’s relations to the rest of the repository. A file with rich frontmatter, typed semantic relations, structured body content, and incoming links from other files has high observable relational density. A file with no frontmatter, no links, and no connections has low density.

“Weakness” and “strength” name positions on this continuum. A weak file has few observable relations. A strong file has many. These are descriptive, not evaluative — a file may be weak because it is new, or because it has not yet been enriched, or because its content genuinely does not participate in many relations.

Dimensions

The score is computed across three dimensions, each measuring a different kind of relation.

Dimension 1: Frontmatter completeness (0-7)

Frontmatter fields establish typed, machine-readable relations. Each present field represents one kind of relation the file participates in.

FieldPointsRelation expressed
title1self-identification
date-created1temporal placement
type (valid value)1classification
tags (3-5, CamelCase)1conceptual community membership
description1summarization
authors1provenance
any semantic relation field1typed dependency

Semantic relation fields: defines, cites, requires, teaches, part-of, extends, questions, addresses.

Dimension 2: Body structure (0-4)

Body content establishes the file’s internal coherence and its untyped connections to other files.

CheckPointsWhat it measures
non-empty body1the file says something
at least one heading1the content is structured
at least one markdown link1untyped connection to another file
no wikilinks1all connections are resolved

Dimension 3: Integration (0-3)

Integration measures the file’s position in the network — whether other files know about it.

CheckPointsWhat it measures
at least one incoming link1another file references this one
at least 3 outgoing links1this file references others
parent directory has index.md1structural placement is documented

Total: 0-14

RangeLabelDescription
0-3weakfew observable relations
4-7developingsome structure, gaps remain
8-11establishedwell-connected, minor gaps
12-14strongfully integrated

Computability

Dimensions 1 and 2 are fully mechanical — a script can compute them by parsing frontmatter and scanning body content. Dimension 3 requires an index of links across the repository, which can be built mechanically but requires a full scan.

The score is designed so that the most impactful improvements (adding frontmatter fields) are the easiest to automate. This follows policy 001: the weakest files need the least capable agents to improve.

Relationship to skill maturity

The file improvement operations that increase the score follow the skill maturity lifecycle:

  • Stage 1-2 (inference): semantic relation fields, body improvements, integration improvements
  • Stage 3 (delegable): description, tags, type classification via local model
  • Stage 4-5 (procedural/tool): date-created from git, deprecated field fixes, frontmatter normalization

The score itself is a stage 5 operation: fully mechanical, always terminates, no inference needed.