diff --git a/.claude/memory/MEMORY.md b/.claude/memory/MEMORY.md
index 1a3bf89..d095d10 100644
--- a/.claude/memory/MEMORY.md
+++ b/.claude/memory/MEMORY.md
@@ -2,3 +2,5 @@
 - [No nohup with pueue](feedback_no_nohup_with_pueue.md) — run `pueue follow|wait` directly as the bg task; nohup&  orphans it from the harness.
 - [Burn down task list](feedback_burn_down_task_list.md) — when many asks are queued, do them all; don't stop to ask which first.
 - [Workshop paper goal](project_workshop_paper_goal.md) — current phase is ablations+seeds for a workshop paper; artifact tracker A1-A7 lives in docs/spec/20260602_writeup_spec.md.
+- [qmd prefer lexical](qmd-prefer-lexical.md) — search local papers with `qmd search`/`rg`, not vector (corpus ~93% unembedded, can't fit embeddings).
+- [Semantic Scholar keyed access](semantic-scholar-keyed-access.md) — S2 API key in semantic-search skill .env; use it to dodge 429s.
diff --git a/.claude/memory/qmd-prefer-lexical.md b/.claude/memory/qmd-prefer-lexical.md
new file mode 100644
index 0000000..1c8bad3
--- /dev/null
+++ b/.claude/memory/qmd-prefer-lexical.md
@@ -0,0 +1,22 @@
+---
+name: qmd-prefer-lexical
+description: "Default to lexical search (qmd search / rg) on the papers corpus, not vector/semantic"
+metadata: 
+  node_type: memory
+  type: feedback
+  originSessionId: dfb6617b-8e6e-4008-96e0-81669fc600b4
+---
+
+For local paper search, default to lexical: `qmd search` (BM25) or `rg`, NOT
+`qmd vsearch`/`qmd query` (vector/HyDE/rerank).
+
+**Why:** (1) wassname finds vector search rarely helps him. (2) The big `papers`
+qmd collection (~48k files) is ~93% unembedded, so semantic modes fall back to
+junk there. (3) He cannot fit the embeddings on his PC, so `qmd embed` is not a
+real fix.
+
+**How to apply:** When dispatching search agents over the local corpus, instruct
+them to use `qmd search`/`rg` first and reach for `qmd query` only as a last
+resort on a small embedded collection (e.g. markdown-notes). A subagent once
+burned ~5 min and crashed (exit 144) running `qmd query` over `papers`; lexical
+returns in milliseconds. Do not suggest running `qmd embed` on his machine.
diff --git a/.claude/memory/semantic-scholar-keyed-access.md b/.claude/memory/semantic-scholar-keyed-access.md
new file mode 100644
index 0000000..479aa89
--- /dev/null
+++ b/.claude/memory/semantic-scholar-keyed-access.md
@@ -0,0 +1,25 @@
+---
+name: semantic-scholar-keyed-access
+description: Semantic Scholar API key lives in the semantic-search skill .env; use it to avoid 429s
+metadata: 
+  node_type: memory
+  type: reference
+  originSessionId: 14deeefc-610a-40ee-b01c-03cf4f1f54b6
+---
+
+The keyless Semantic Scholar API (api.semanticscholar.org/graph/v1) 429s fast.
+A real S2 key (len 40) is stored at
+`~/.claude/skills/semantic-search/.env` as `SEMANTIC_SCHOLAR_API_KEY`.
+
+Use it for direct S2 calls:
+```sh
+set -a; . ~/.claude/skills/semantic-search/.env; set +a
+curl -s "https://api.semanticscholar.org/graph/v1/paper/arXiv:<ID>?fields=title,authors,citationCount" \
+  -H "x-api-key: $SEMANTIC_SCHOLAR_API_KEY"
+```
+or just call the `semantic-search` skill, which loads the key itself.
+
+The `bibtex` MCP (DBLP/S2) sometimes returns 0 for brand-new arXiv papers
+(days old); arXiv `citation_author` meta tags are the authoritative author list
+and the keyed S2 API confirms them once indexed. See [[qmd-prefer-lexical]] for
+the analogous local-search gotcha.