Closes #319.

docs/research/2026-05-24-memory-system-benchmarks.md:215 documented
that the "17% E2E QA for MemPalace" attribution to engram-2 was not
substantiated in their published materials — what engram-2 actually
published is a ~17-point gap between their own LoCoMo score (74.5%)
and SOTA (91.7%), attributed to the answerer model.

README.md:163 and docs/ECOSYSTEM.md:45 predated that fact-check and
still stated the attribution as fact. This change adopts the 2026-05-24
doc as source-of-truth:

- README "Active investigations" reframes the entry as "End-to-end QA
  measurement on the post-structural-fix palace". The corpus-shape
  pathology that prompted #168 is real (pre-migration kind=content
  returned 3 tokens/Q vs post-migration 1,267) and is closed; the
  deliverable becomes a positive measurement rather than a rebuttal.
- docs/ECOSYSTEM.md describes what engram-2 actually published and
  cross-links the research doc.

#168 itself stays open — the deliverable at notebook/data/
cat9-postmigrate-e2e/REPORT.md is unchanged. Only the framing of what
the numbers answer changes.

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