docs: clarify evaluation method and measured limitations

[noetion]

Purpose

Make the evaluation question, sequence rule, reproducible result, and error analysis easier to follow.

Changes

• Introduce the experiment and show the measured sequence.
• Summarise the holdout counts using the existing result fields.
• Add clone and directory steps to reproduction instructions.
• Give the provenance finding its own section and link to the error audit.
• Keep the generated README and its renderer synchronized.

Verification

• uv run tripwire verify reproduced the committed result.
• All 37 tests passed.
• Ruff and mypy passed; source distribution and wheel built successfully.
• git diff --check passed.

Review scope and limitations

Review the README together with its renderer. Detection logic, corpus, splits, and committed metrics are unchanged. The canonical holdout remains spent; the provenance finding is not presented as a tested successor detector. Local pytest reported a cache-write warning; tests passed.

[noetion]

Reviewed 9720a5a: no blocking findings in the README and renderer diff. The generated summary matches the committed result (TP 6, FP 1, TN 36, FN 11); uv run tripwire verify reproduces it. All 37 tests, Ruff, mypy, and package build passed. GitHub verification passed for push and pull-request runs. Checked rendered sections and links. Corpus, rule logic, splits, and metrics are unchanged; limitations and the spent-holdout qualification remain explicit.