Align Stage 2 filtering and precision claims with current behavior

[Spectorian]

Problem

The README describes Stage 2 as filtering false positives and improving precision to approximately 87%. The linked implementation retains deterministic findings when the model disputes or does not confirm them. Meta-review can annotate and enrich findings, but those findings remain in the result.

An accuracy figure without a clear benchmark, model/scanner version and evaluation boundary can also be mistaken for the current pipeline's validated performance.

Expected documentation

Describe the current review/enrichment behavior and its limits where Stage 2 is explained. Either connect the number to a verified, clearly scoped evaluation or remove the unqualified number. Do not substitute a rate calculated from selected false-positive examples.

The documentation should describe that behavior accurately, without changing how LLM disagreement affects deterministic findings.

Historical #540 corrected a different research-denominator claim; it does not resolve this Stage 2 wording.

Relevant code

README.md:815, meta_analyzer.py:381, meta_analyzer.py:437.

[rng1995]

Documentation corrections are proposed in the open PR #684. They describe retained deterministic findings, llm-unconfirmed, and confidence enrichment rather than claiming automatic Stage 2 removal of those findings. Keeping this issue open until the documentation change merges.

PR-state snapshot checked on 2026-10-04:

• PR #684 — open, non-draft; GitHub review decision: changes requested; latest reported check rollup: success.

Keeping this issue open: the relevant implementation is not merged. Check/review status is a point-in-time snapshot, not a claim of merge readiness.