diff --git a/CLAUDE.md b/CLAUDE.md index 6d1d21d..25bb618 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -352,7 +352,12 @@ revert without reading why. When in doubt, walk the - LLM: `https://hermes.ai.unturf.com/v1` (Hermes-3 Llama-3.1-8B-FP8- Dynamic on vLLM, 82K ctx, no auth). `uncloseai.com` is marketing only. Override via `--endpoint` or `ARBORIST_LLM_ENDPOINT`. -- Wikipedia dumps: `https://dumps.wikimedia.org/archive/2003/2003-05-16/en/`. +- Wikipedia dumps: two corpora available — the 2003-05-16 cur snapshot + (`https://dumps.wikimedia.org/archive/2003/2003-05-16/en/`) and a + **2010** snapshot. **The 2010 corpus is what is currently loaded** in + `~/.arborist/shards` (866K docs/shard; contains post-2003 articles like + Barack Obama / YouTube — the discriminator). Fabrication-bait fixtures + (`qa_questions_stale_map`) target post-**2010** events. `robots.txt` returned 404 → no rules. ### Budget discipline — Hermes / Qwen first, Opus deferred diff --git a/docs/energy-cogs-benchmark.md b/docs/energy-cogs-benchmark.md index cc670bd..f12bc0b 100644 --- a/docs/energy-cogs-benchmark.md +++ b/docs/energy-cogs-benchmark.md @@ -30,8 +30,11 @@ Two rigs: Common to both: arborist `query()` pipeline, **STOCK V.1** frozen config (`claim_lattice` hash `036a4c79…`, `quote` `5b6ca4c5…`, -`bench/stock_v1.py`); orchestrator drives the endpoint, power sampled on +`bench/stock_v1.py`); **Wikipedia 2010 corpus** (~866K docs/shard, +`~/.arborist/shards`); orchestrator drives the endpoint, power sampled on the box; **$0.33/kWh** (the only non-measured input, `--price-per-kwh`). +(Fabrication-bait questions target post-2010 events, so the substrate +correctly abstains where bare models answer from parametric memory.) **Two arms, same model (the substrate-delta design):** - **solo** — question only, neutral system prompt, no retrieval, no verifier.