- **README.md + docs/ECOSYSTEM.md — soften 'engram-2 17% E2E QA' framing per the 2026-05-24 research doc's unsubstantiated finding (#319)** ([`HEAD`](https://github.com/techempower-org/mempalace/commit/HEAD))

Three docs referenced the engram-2 "17% E2E QA for MemPalace"

attribution with inconsistent framing:

- ``README.md:163`` treated it as a critique we're answering.

- ``docs/ECOSYSTEM.md:45`` stated it as a factual claim engram-2

makes.

- ``docs/research/2026-05-24-memory-system-benchmarks.md:215``

documented that the 17% attribution to engram-2 is **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.

The 2026-05-24 doc is the more recent and more carefully sourced;

README + ECOSYSTEM.md predated the fact-check. This PR adopts the

2026-05-24 doc as source-of-truth:

- ``README.md`` "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 and the

structural fix closed it on our corpus; 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 for readers wanting more.

#168 itself stays open — the deliverable (publish E2E numbers at

``notebook/data/cat9-postmigrate-e2e/REPORT.md``) is unchanged.

Only the framing of what the numbers are answering changes.

*Files:* `README.md`, `docs/ECOSYSTEM.md`

- **kg_llm_extractor rewrites AGE dollar-quote tag in triples so drawers indexing palace source code don't fail at add_triple (#313)** ([`3fb9428`](https://github.com/techempower-org/mempalace/commit/3fb9428))