p3: map Brave LLM Context; RRF merge; per-source no_rerank

Live llm/context hits live in grounding.generic plus a URL-keyed
sources map, not a sources array. Fan-in uses reciprocal-rank fusion
instead of cosine. no_rerank keeps that source's own order as a prefix.

docs/P3.md

@@ -112,12 +112,14 @@ the shared fan-in. This is the Hermes SearXNG provider's request (`plugins/web/s
 
 ### 3.2 Brave LLM Context
 
-PLAN §2, verified earlier: `GET /res/v1/llm/context?q=…` with `BRAVE_API_KEY` from
+PLAN §2: `GET /res/v1/llm/context?q=…` with `BRAVE_API_KEY` from
 `/var/lib/sift/secret.env` (0640 root:sift). **Do not read SearXNG's secret.env.**
-Header `X-Subscription-Token`. Map result URLs and snippets from the live JSON
-(`sources` and any URL-bearing fields). Do **not** assume a model-written
-`grounding` answer exists — confirm against a live response before dropping or
-keeping any field that is not a URL/title/snippet.
+Header `X-Subscription-Token`. Live response (Brave docs + Open WebUI #24120):
+hits are `grounding.generic[]` (`url`, `title`, `snippets[]`); top-level `sources`
+is a URL-keyed metadata map, not a result array. Map generic first, join
+`snippets`, then URL keys in `sources` for anything generic missed. Do **not**
+treat a model-written `grounding` string as a snippet. Fixture:
+`tests/testdata/brave_llm_context.json`.
 
 This is **not** Brave Web Search (`/res/v1/web/search`). SearXNG already runs `braveapi`
 as an engine; doubling that is wasted spend.
@@ -148,25 +150,23 @@ placeholder line the same way (empty, on-box fill).
    dropped, trailing slash dropped on the path, `www.` left as-is (different sites). First
    snippet wins; `source` becomes the first source that produced it; later sources only
    affect rerank input if they brought a longer snippet (keep the longer one).
-5. Rerank (§5) unless `rerank: false` or Dixie cannot take the call.
+5. Merge: reciprocal-rank fusion across source lists unless `rerank: false`.
+   A source with `no_rerank = true` keeps its own order as a prefix (first-source
+   wins on URL). Cosine/honcho-embed is not used for search.
 6. Truncate to `k` (native) or `limit` (`/v2/search`, default 5, max 20).
-7. Store in the search cache **only if** every attempted source succeeded and
-   rerank did not fall back. A timeout or fallback must not pin a partial result
-   for the rest of the hour (skip the write, or use a short TTL such as 2 m — skip
-   is the default). Return.
+7. Store in the search cache **only if** every attempted source succeeded.
+   A timeout must not pin a partial result for the rest of the hour.
 
 Search **never GETs a result URL**. Agents `sift fetch` those through the P1 guard.
 
-## 5. Rerank and Dixie
+## 5. Merge (RRF)
 
-Embed the query and each remaining snippet with `honcho-embed` (same engine, same
-`dixie.max_inflight` FIFO as digest). Cosine similarity → `score`. Restore nothing to
-source order except on fallback.
+Fan-in uses reciprocal-rank fusion (`1/(60+rank)`) over per-source lists. Search
+does not call `honcho-embed`; Dixie saturation cannot fail a search.
 
-**Dixie saturation must not fail a search:** if the embed permit cannot be had within
-`dixie.queue_wait`, or embed errors, or the queue is full → keep source-fan-in order
-(SearXNG, then Brave, then Tavily, then original rank inside a source), warning
-`rerank fell back: <reason>`. Never call chat/digest for search in P3.
+`[search.tavily] no_rerank = true` keeps Tavily's own order at the front of the
+result list; remaining sources still RRF. `rerank = false` skips RRF and keeps
+source-fan-in order (SearXNG, then Brave, then Tavily).
 
 ## 6. Guard boundary (SSRF)