Understanding results

Every finding has a verdict (what Necro asserts) and, for dead code, a confidence tier (how sure it is). The tier drives what Necro is willing to do automatically.

Confidence tiers

Tier	Condition	What it means	Auto-fix eligible
certain	private scope, 0 references, no taint nearby	Safe to remove	✅ yes
likely	exported, 0 internal references, not an entry, no taint	Probably dead, but external consumers are invisible	❌ no — confirm first
maybe	taint nearby (dynamic import, reflection) or public API	Ambiguous — quarantined, not condemned	❌ never

The maybe tier is the false-positive sink: rather than guess, Necro reports the ambiguity with reasons. Only certain findings are eligible for automatic removal (planned --fix); everything else needs a human.

Verdicts

• dead — unreachable from any production entry point. Tiered as above.
• test-only — reachable, but only through test files. Production-dead. See The test-only verdict. Reported, never auto-removed.

Alive code (reachable from a production entry) is not reported.

Sort order

Findings print worst-first: certain → likely → maybe → test-only, then by file and line. Use --top N to see only the worst N.

A note on coverage

You’ll see • coverage: not available in evidence chains. Coverage ingestion (lcov/c8) — which would let Necro confirm a symbol is never executed at runtime — is planned. Until then, tiers are computed from static signals alone.