report: second pass of external-facing rewrites — add methodology, remove failing-samples refs, fix synth-typed claims

Six targeted updates per user feedback:

1. Removed the "High variance run-to-run" bullet from the E03 deep-dive
   section and the "Replication check" paragraph below it — the
   replication-check discussion was distraction from the main story.

2. Removed the "Replication checks" section at the end of the E08
   deep-dive (E05 + E03 re-run side-by-side commentary).

3. Removed point #5 of the E08-vs-E03 mechanism breakdown (CU's higher
   run-to-run variance) — it depended on the replication data the user
   wanted gone, so the supporting evidence was no longer in the report.

4. Fixed the E04 synth-samples failure-mode prose. The previous version
   claimed that density wasn't the issue because HR0081 (10) is also
   fully filled. That's factually wrong — HR0081 (10) is not densely
   filled; the synth-* cluster is the densest set in the dataset. The
   rewrite says E04 struggles on dense images and HR0081 (10) doesn't
   reproduce the problem because it has many blank cells.

5. Removed all pass_rate references throughout the report (headline
   observations, E03 deep-dive headline, E08 deep-dive bullets, the
   same-pipeline bake-off table, the failure-mode-samples manual sample
   (6) bullet, the Reflection's first point). The 0.8 cutoff was
   arbitrary and not meaningful for downstream consumers.

6. Removed the "E06 reproducibility" sub-section in Appendix A and the
   entire "Appendix B — Reproducing the comparison report" section.
   Also removed the dangling "Appendix B has the reproduction commands"
   line from the introduction.

Net diff: −71 / +12. Report is now noticeably shorter and more focused
on the actual cross-engine comparison story without the meta-discussion
of how the runs were validated.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: 2 people