Repository Grounding

​

Formula

RepositoryGroundingScore = clamp01(
    0.25 · sat(repo_link_density;    0.5, 4.0)
  + 0.25 · path_resolution_rate
  + 0.20 · sat(code_example_density; 0.5, 3.0)
  + 0.15 · sat(identifier_density;   0.02, 0.12)
  + 0.15 · sat(version_fact_density; 0.01, 0.08)
)

​

Components

• Repo link density — relative repo links per 100 words.
• Path resolution rate — how many path-like tokens resolve to actual files.
• Code example density — labelled code-fence count per 100 words.
• Identifier density — identifier-like tokens (CamelCase, snake_case, dotted paths) per word.
• Version fact density — version pins, semver, dates, hash references per word.

​

Bands

​

How downstream metrics use it

• Filler / Lazy Risk consumes 1 − RepositoryGroundingScore as one of its sub-scores.
• The PR comment surfaces a 🔴 callout when grounding crosses a band downward.

​

References

• Pirolli, P. & Card, S. (1999). Information Foraging. Psychological Review 106(4): 643–675 — underpins the “scent” intuition behind link-density and identifier-density. DOI.
• Spence, R. (1999). A framework for navigation. International Journal of Human-Computer Studies 51(5): 919–945 — navigation framework that motivates resolved-link weighting. DOI.

​

See also

• Evidence Coverage — per-section structural support.
• Filler / Lazy Risk — uses grounding as an input.