# Source Priority

Use this order when deciding whether a source is strong enough to support research writing.

## Tier 1: Project-native evidence

Highest priority when the claim is about the user's own method or results.

- experiment logs
- result tables
- plots generated from known runs
- configs and checkpoints
- code proving system design choices
- internal notes with identifiable provenance

Use this tier for claims about what the project actually does.

## Tier 2: Primary external sources

Use for prior work, benchmark definitions, and external comparisons.

- published papers
- arXiv preprints from original authors
- official dataset pages
- benchmark leaderboards
- official documentation
- project repositories maintained by the original authors

Prefer the original paper over a blog post describing the paper.

## Tier 3: Secondary sources

Use only for discovery, triangulation, or non-critical context.

- blog posts
- newsletters
- social posts
- third-party summaries
- forum threads

Do not rely on Tier 3 for factual claims if a Tier 1 or Tier 2 source can be checked.

## Conflict handling

If sources disagree:

- prefer the more primary source
- prefer the more specific source for the exact claim
- report the conflict explicitly
- do not silently merge inconsistent numbers or conclusions

## Default rule

If a sentence in the paper would be embarrassing to defend during peer review, the source quality is too weak.