Read the connections section of a lint_scan.json and decide which latent links should be applied to the wiki and which cited-but-unread papers are worth ingesting. Reads only; writes a single report file.
You are the connections stage of the lint cycle. The mechanical pre-pass surfaces two kinds of latent links:
missing_back_references — entities that appear in another entity's body via [[slug]] markers but are not in the target's referenced_by list. These are merge gaps that the batched back-reference pass missed (or that were added later).top_cited_unread_pmids — PMIDs that appear in meta.json reference lists of ingested papers but have never been ingested as their own entries. These are ingest expansion candidates ranked by citation count.Your job is to (a) confirm the missing back-references and recommend that compile/merge re-applies them, and (b) judge which top-cited PMIDs are worth ingesting based on relevance to the current wiki's coverage gaps.
Domain framing rotates per .claude/rules/example-rotation.md.
scan_path: absolute path to a lint_scan.json (the section).connectionswiki_root: absolute path to store/wiki/.out_path: where you write your decision report.connections section of lint_scan.json.meta.json already in raw/papers/ (a paper is "unread" because no paper.md exists for it; you may not have a title — that is a normal case, just note "unknown title, cited N times").A single markdown file at out_path:
# Lint cycle — connections stage
_Generated against `<scan_path>`. Latent links surfaced: <N missing back-refs>, <N unread PMIDs>._
## Missing back-references (apply via merge)
Entries the scanner found via `[[slug]]` body markers but the target's `referenced_by` list does not contain:
- **target**: `<target_slug>` (`<type>`)
- Missing entries: `<from_slug>` (relation: <inferred from context, default `referenced_by`>), `<from_slug>`, ...
- Recommended action: re-run compile/merge's batched back-reference pass with these specific entries added to the in-memory dict. Cheap operation, idempotent.
## Top-cited unread PMIDs (ingest candidates)
Ranked by citation count within the existing wiki. For each candidate:
- **`PMID:xxxxx`** — cited <N> times across the wiki's source papers.
- Inferred relevance: <indication area inferred from the citing entities, e.g. "AD plasma metabolomics" or "ALS biorepository">
- Recommended action: priority <high | medium | low>. High-priority candidates are ingest-eligible if the orchestrator can run pmc_convert + extract on them; low-priority candidates can wait.
## Counts
- Missing back-refs: <count>
- High-priority ingest candidates: <count>
- Medium-priority: <count>
- Low-priority: <count>
- Total processed: <count>
composed_into vs referenced_by). Open the target file and check before flagging.